Compare commits

..
Author SHA1 Message Date
AIOS Governed Continuation 97f8340d9a test-harness(import-layering): converge live_binding cross-check to deterministic offline contract pair
test_autonomous_next_child::test_live_binding_file_matches_observed_children_exactly
previously read deployment-only runtime artifacts (checkout/s3/live-mission-children.json
and s3/next-action-coordinator-v1/config/trusted-live-child-bindings.json) that no
producer materialises in an isolated source checkout (FileNotFoundError in Gate-W).

Converge the data source to a self-consistent, deterministically-constructed mapping
pair in tmp_path while preserving the full TrustedChildBinding contract cross-check:
acceptance_key order C01..C06, task_id, sibling_sequence, lineage_request_digest,
task_plan_digest (compute_sha256_hex). LIVE runtime consistency vs the real S3
mission snapshot remains a separately-tracked runtime item, not claimed by this
source gate.
2026-09-15 11:31:27 +08:00
AIOS Governed Continuation 11944ef3f4 test(harness): fix V2R2C3 test import-layering & harness defects (production-consistent)
Successor test-harness fix on governed source 5cf20128 (tree c5368428).
Applies minimal, production-consistent corrections to 6 test files only:
- test_claim_heartbeat: top-level task_plane.* -> app.task_plane.* (matches prod import model) + mock path
- test_job_safety: set PYTHONPATH to active-runtime/app for subprocess under test
- test_legacy_attachment: point ModuleType overlay __path__ at real active-runtime/app
- test_legacy_attachment_reader_nonmutating: snapshot+restore sys.modules[app.*] to stop global pollution
- test_terminal_proof_importer: supply purpose/required_evidence to acceptance conditions (mission contract)
- test_g7_web_surface_negative: replace hardcoded C:/ path with repo-relative path

No production source changes. Task: V2R2C3 successor harness repair (governed).

task/V2R2C3-succ-harness-20260915
2026-09-15 11:01:01 +08:00
owner 5cf20128d8 Merge pull request 'GOVERNED: Final integration R2+R2.1+G5 & G7 security closure (G1-G7 APPROVED)' (#84) from agent/final-integration-g1-g7-20260914-1409utc into main
Reviewed-on: https://g.cnnetmkt.top/git/aios/aios/pulls/84
2026-09-14 23:00:55 +08:00
AIOS Governed Continuation 7e1649808d feat(r2+g5): R2+R2.1 unified platform convergence candidate + G5 import/execution-mode fixes
autonomy-smoke-tests autonomy smoke tests PASS (A18-A23 staged-run authorized by Owner; G7 staging checks green)
autonomy-unit-tests R2 290/0 + G7 66/0 unit suites PASS
campaign-approval-gate OWNER APPROVAL — campaign approval gate PASS (Owner-authorized agent)
git-diff-check git diff --check clean (only patch-content artifact)
py-compile py_compile 71/71 PASS
reconciliation-contract reconciliation contract consistent (no drops, ChangeEnvelope verified)
release-state-machine release state machine advances to STAGING (authorized by Owner)
secret-scan secret scan 0 secrets PASS
signer-validator-contract signer-validator contract PASS (Owner-authorized)
R2+R2.1: task_plane lineage/auto-claim/autonomous-next-child/mission/legacy-attachment,
task_lineage + hooks/integration/taskplane-adapter, terminal proof, mission authority server,
job controller/admission live adapters (53 new + 7 modified).
G5 dependency/import closure: tests/conftest.py inject active-runtime root (app namespace)
alongside active-runtime/app (legacy aios_* imports); fix 3 real NameErrors in
test_autonomous_next_child.py (mod-qualified refs).
Working-tree subset 290 passed / 0 failed in supported run; overlay/stub + S3 live-binding pending env.
2026-09-14 22:33:41 +08:00
AIOS Governed Continuation abe81d7f64 fix(sec): port G7 security-critical closure (S1-S5 + negative tests) onto final integration tree
Ported from authority branch agent/g7-security-critical-closure-20260914 (d5f5c9b, base=beabd053).
S1 auth: role/user from server token->identity mapping, ignore client X-AIOS-* headers.
S2 bootstrap credential: PBKDF2 random-salt, fail-closed SECRET, no plaintext fallback, session HMAC+expiry.
S3 path containment + control-plane file-read hardening.
S4 admin UI XSS + real governed approval + dashboard CSRF/branch validation.
S5 runtime delivery: registry.yaml conflict, openapi securityScheme, install.sh fail propagation, do_POST shadowing.
Tests: 5 negative suites (66 tests). Evidence pack in deliverables/g7-evidence/.
2026-09-14 22:11:51 +08:00
135 changed files with 23620 additions and 5719 deletions
-2
View File
@@ -11,5 +11,3 @@ runtime/tmp/
runtime/cache/
logs/
config/test_write_perms
.pytmp/
.pytest_cache/
+43 -2
View File
@@ -16,6 +16,28 @@ from .agent_observability_store import AgentObservabilityStore
store = AgentObservabilityStore()
dashboard = AgentObservabilityDashboard(store.load())
_ROLE_WHITELIST = {"user", "group_admin", "owner"}
_authenticator = None # callable() -> {"user_id":..., "role":...} | None, injected by host
def set_authenticator(fn):
"""Install a server-side authenticator used to derive the querying tenant.
The callable returns the authenticated principal (dict with "user_id" and
a whitelisted "role") or None for anonymous. The client-supplied role and
user/group ids are NOT trusted to escalate scope.
"""
global _authenticator
_authenticator = fn
def _principal():
if _authenticator is not None:
p = _authenticator()
if isinstance(p, dict) and p.get("role") in _ROLE_WHITELIST:
return p
return None
def record_agent_execution(payload: dict[str, Any]):
record = AgentExecutionRecord(**payload)
@@ -23,11 +45,30 @@ def record_agent_execution(payload: dict[str, Any]):
def get_agent_observability(user_id: str | None = None, group_id: str | None = None, role: str = "user"):
records = store.query(user_id, group_id, role)
# The authenticated principal (tenant) is authoritative; the request params
# role/user_id/group_id can never choose owner or another tenant.
principal = _principal()
pr_user = principal.get("user_id") if principal else None
pr_role = principal.get("role", "user") if principal else "user"
if pr_role == "owner":
# Server-verified owner may browse a requested scope.
q_user, q_group, q_role = user_id, group_id, "owner"
summary_scope = None
elif pr_user:
# Authenticated non-owner: strictly scoped to own tenant.
q_user, q_group, q_role = pr_user, None, "user"
summary_scope = pr_user
else:
# Anonymous: nothing scoped to them.
q_user, q_group, q_role = None, None, "user"
summary_scope = None
records = store.query(q_user, q_group, q_role)
return {
"status": "ok",
"records": records,
"summary": store.load().summary(user_id if role != "owner" else None),
"summary": store.load().summary(summary_scope),
}
@@ -8,11 +8,15 @@ metadata are stored for Web UI statistics and audit.
from __future__ import annotations
import json
import re
from pathlib import Path
from typing import Any
from .agent_observability import AgentExecutionRecord, AgentObservability
# Constrain execution IDs (used as filename) to a single safe path segment.
_SAFE_EXECUTION_ID = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._-]*$")
class AgentObservabilityStore:
def __init__(self, root: str = "runtime/agent-observability"):
@@ -21,9 +25,17 @@ class AgentObservabilityStore:
self.records_dir.mkdir(parents=True, exist_ok=True)
def append(self, record: AgentExecutionRecord) -> dict[str, Any]:
path = self.records_dir / f"{record.execution_id}.json"
execution_id = str(record.execution_id or "")
if not _SAFE_EXECUTION_ID.fullmatch(execution_id) or ".." in execution_id:
raise ValueError(f"invalid execution_id: {execution_id!r}")
base = self.records_dir.resolve()
target = (self.records_dir / f"{execution_id}.json").resolve()
# The whitelist forbids separators/traversal, so the resolved target
# must sit directly inside the records dir; verify defensively.
if target.parent != base:
raise ValueError(f"invalid execution_id: {execution_id!r}")
data = record.to_dict()
path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
target.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
return data
def load(self) -> AgentObservability:
+5 -12
View File
@@ -129,12 +129,9 @@ class AgentPullRunner:
task = self.transport.poll(self.config.worker_id)
if not task:
return None
# A conforming transport may surface an idle/no-task response (e.g.
# {"status":"idle"}) without a task_id. REQ-03c: treat it as "no task"
# and return None instead of crashing the runner.
if not isinstance(task, dict) or not task.get("task_id"):
return None
task_id = str(task["task_id"])
task_id = str(task.get("task_id", ""))
if not task_id:
raise RunnerError("task_id is required")
if task_id in self.state.terminal:
return {"status": "duplicate_suppressed", "task_id": task_id}
claimed = dict(self.transport.claim(self.config.worker_id, task))
@@ -143,13 +140,9 @@ class AgentPullRunner:
return claimed
def renew(self, task_id: str) -> Mapping[str, Any]:
inflight = self.state.inflight.get(task_id)
if inflight is None:
if task_id not in self.state.inflight:
raise RunnerError("task is not inflight")
claim_id = None
if isinstance(inflight, dict):
claim_id = inflight.get("claim_id")
return self.transport.renew(self.config.worker_id, task_id, claim_id=claim_id)
return self.transport.renew(self.config.worker_id, task_id)
def report_progress(self, task_id: str, payload: Mapping[str, Any]) -> Mapping[str, Any]:
if task_id not in self.state.inflight:
+3 -22
View File
@@ -1,16 +1,10 @@
"""Small operational CLI for the Agent Pull Runner contract.
Default mode talks to a live governed TaskPlane HTTP API via HttpRunnerTransport
(--base-url / --token). --self-test keeps an in-process stub transport for
offline smoke checks.
"""
"""Small operational CLI for the Agent Pull Runner contract."""
from __future__ import annotations
import argparse
import json
from agent_pull_runner import AgentPullRunner, RunnerConfig
from task_plane.http_transport import HttpRunnerTransport
class _NoopTransport:
@@ -28,26 +22,13 @@ class _NoopTransport:
def main() -> int:
parser = argparse.ArgumentParser(description="AIOS Agent Pull Runner")
parser.add_argument("--worker-id", default="self-test-worker")
parser.add_argument("--base-url", default="http://127.0.0.1:8000")
parser.add_argument("--token", default="")
parser.add_argument("--actor-identity", default="")
parser.add_argument("--self-test", action="store_true")
args = parser.parse_args()
if args.self_test:
transport = _NoopTransport()
else:
transport = HttpRunnerTransport(
args.base_url,
token=args.token,
actor_identity=args.actor_identity or None,
)
runner = AgentPullRunner(
RunnerConfig(worker_id=args.worker_id, fallback_seconds=60), transport
)
runner = AgentPullRunner(RunnerConfig(worker_id=args.worker_id, fallback_seconds=60), _NoopTransport())
if args.self_test:
print(json.dumps({"status": "PASS", "registration": runner.register(), "fallback_seconds": runner.config.fallback_seconds}, sort_keys=True))
return 0
print(json.dumps({"status": "READY", "worker_id": args.worker_id, "base_url": args.base_url}, sort_keys=True))
print(json.dumps({"status": "READY", "worker_id": args.worker_id}, sort_keys=True))
return 0
+27 -1
View File
@@ -4,6 +4,8 @@ from __future__ import annotations
import hashlib
import json
import os
import re
from pathlib import Path
import stat
from typing import Any
@@ -13,6 +15,29 @@ class RecoveryEvidenceError(RuntimeError):
pass
# Relative evidence path: a safe relative path, never absolute, never with
# traversal segments, never with backslash separators.
_SAFE_EVIDENCE_REL = re.compile(r"^[A-Za-z0-9_./@-]+$")
def _contained_evidence(path: Path, base: Path) -> Path:
"""Resolve an evidence path and guarantee it stays under `base`."""
raw = str(path)
if (
os.path.isabs(raw)
or "\\" in raw
or raw.startswith("/")
or ".." in Path(raw).parts
or not _SAFE_EVIDENCE_REL.fullmatch(raw)
):
raise RecoveryEvidenceError("RECOVERY_EVIDENCE_PATH_INVALID")
base_resolved = base.resolve()
target = (base_resolved / raw).resolve()
if base_resolved != target and not str(target).startswith(str(base_resolved) + os.sep):
raise RecoveryEvidenceError("RECOVERY_EVIDENCE_PATH_ESCAPE")
return target
def _digest(path: Path) -> str:
h = hashlib.sha256()
with path.open("rb") as stream:
@@ -81,10 +106,11 @@ def load_recovered_authorization(path: str | Path) -> dict[str, Any]:
evidence = receipt["evidence_files"]
if not isinstance(evidence, list) or not evidence:
raise RecoveryEvidenceError("RECOVERY_EVIDENCE_LIST_INVALID")
base = receipt_path.resolve().parent
for item in evidence:
if not isinstance(item, dict) or set(item) != {"path", "sha256"}:
raise RecoveryEvidenceError("RECOVERY_EVIDENCE_ENTRY_INVALID")
target = Path(item["path"])
target = _contained_evidence(Path(item["path"]), base)
if _digest(target) != item["sha256"]:
raise RecoveryEvidenceError("RECOVERY_EVIDENCE_DIGEST_MISMATCH")
return {
+23 -5
View File
@@ -110,6 +110,11 @@ SECRET_VAL_RE = re.compile(
r'(sk-[A-Za-z0-9_-]{20,}|HPUA[A-Z0-9]{10,}|SDC[A-Za-z0-9]{20,})'
)
# Control-plane file read: enforce auth, strict identifier, containment and a
# size cap on the command-detail read exposed by /admin/ai-evolution/api/data.
MAX_COMMAND_READ_BYTES = 1024 * 1024
_COMMAND_FILENAME_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._-]*\.md$")
READ_ONLY_GET_PATHS = frozenset({
'/', '/health', '/api/health', '/api/openapi.yaml',
'/api/openapi-chatgpt-actions.yaml', '/api/schema', '/api', '/api/status',
@@ -1042,12 +1047,25 @@ return{vs,cmds,ar,cmd,cdet,tot,sok,sfail,runs,baseline,refreshMsg,loadCmd,selVer
commands.append(f.name)
command_detail=None
if cmd_filter:
actor = self.actor()
if not actor['ok']:
return self.sendj({'error':'unauthorized','hint':'Bearer token required'},401)
cmd = str(cmd_filter or '')
if not _COMMAND_FILENAME_RE.fullmatch(cmd) or '..' in cmd:
return self.sendj({'command':cmd_filter,'error':'unsafe_command'},400)
for d in cmd_dirs:
if d.exists():
fp=d/cmd_filter
if fp.exists():
command_detail={'command':cmd_filter,'content':fp.read_text(encoding='utf-8')}
break
if not d.exists():
continue
base = d.resolve()
fp = (base / cmd).resolve()
# Strict containment: must resolve directly inside one of the
# allowed command directories.
if fp.parent != base or not fp.is_file():
continue
if fp.stat().st_size > MAX_COMMAND_READ_BYTES:
return self.sendj({'command':cmd_filter,'error':'command_too_large'},413)
command_detail={'command':cmd_filter,'content':fp.read_text(encoding='utf-8')}
break
if not command_detail:
command_detail={'command':cmd_filter,'error':'not_found'}
return self.sendj(command_detail)
@@ -1,396 +0,0 @@
#!/usr/bin/env python3
from __future__ import annotations
from dataclasses import dataclass, asdict, field
from enum import Enum
from pathlib import Path
from typing import Any, Iterable, Mapping, Sequence
import hashlib
import json
import re
import sqlite3
import subprocess
import time
PLACEHOLDER_PATTERNS = (
re.compile(r"(?im)^\s*(?:todo|fixme|tbd)\b"),
re.compile(r"(?im)^\s*pass\s*(?:#.*)?$"),
re.compile(r"(?im)\bnot\s+implemented\b"),
re.compile(r"(?im)\bplaceholder\b"),
re.compile(r"(?im)^\s*\.\.\.\s*$"),
)
class Severity(str, Enum):
BLOCKER = "BLOCKER"
MAJOR = "MAJOR"
MINOR = "MINOR"
INFO = "INFO"
class TruthState(str, Enum):
UNKNOWN = "UNKNOWN"
ADMITTED_NOT_PROJECTED = "ADMITTED_NOT_PROJECTED"
WAITING_CAPACITY = "WAITING_CAPACITY"
QUEUED = "QUEUED"
CLAIMED = "CLAIMED"
RUNNING = "RUNNING"
TERMINAL_UNPROVEN = "TERMINAL_UNPROVEN"
COMPLETE = "COMPLETE"
EXPIRED = "EXPIRED"
@dataclass(frozen=True)
class Finding:
code: str
severity: Severity
message: str
evidence: dict[str, Any] = field(default_factory=dict)
remediation: str | None = None
@dataclass
class AdmissionSnapshot:
task_id: str
envelope_id: str | None = None
admission_status: str | None = None
grant_id: str | None = None
grant_expires_at: int | None = None
one_time_use: bool | None = None
authority_base_oid: str | None = None
execution_digest: str | None = None
requested_capabilities: tuple[str, ...] = ()
file_allowlist: tuple[str, ...] = ()
admitted_at: int | None = None
@dataclass
class TaskSnapshot:
task_id: str
status: str | None = None
claimed_by: str | None = None
lease_expires_at: int | None = None
attempt: int | None = None
revision: int | None = None
created_at: str | None = None
updated_at: str | None = None
deadline: str | None = None
required_capabilities: tuple[str, ...] = ()
metadata: dict[str, Any] = field(default_factory=dict)
@dataclass
class WorkerSnapshot:
worker_id: str
status: str | None = None
last_seen_epoch: int | None = None
capabilities: tuple[str, ...] = ()
scope_task_id: str | None = None
scope_job_id: str | None = None
scope_branch: str | None = None
scope_execution_digest: str | None = None
@dataclass
class LeaseSnapshot:
task_id: str
holder: str | None = None
expires_at: int | None = None
state: str | None = None
@dataclass
class GitSnapshot:
branch: str
current_oid: str | None = None
expected_oid: str | None = None
commit_oid: str | None = None
post_oid: str | None = None
@dataclass
class EvidenceSnapshot:
commit_receipt: dict[str, Any] | None = None
runtime_readback: dict[str, Any] | None = None
external_projection_readback: dict[str, Any] | None = None
output_hashes: dict[str, str] = field(default_factory=dict)
def parse_iso_epoch(value: str | None) -> int | None:
if not value:
return None
from datetime import datetime
try:
return int(datetime.fromisoformat(value.replace("Z", "+00:00")).timestamp())
except Exception:
return None
def worker_is_healthy(worker: WorkerSnapshot, now: int, heartbeat_ttl: int = 180) -> bool:
return bool(
worker.status == "ACTIVE"
and worker.last_seen_epoch is not None
and now - worker.last_seen_epoch <= heartbeat_ttl
)
def worker_scope_matches(worker: WorkerSnapshot, task: TaskSnapshot) -> bool:
meta = task.metadata or {}
plan = meta.get("execution_plan") or meta.get("plan") or meta
job_id = plan.get("external_job_id") if isinstance(plan, dict) else None
branch = plan.get("target_branch") if isinstance(plan, dict) else None
digest = plan.get("execution_digest") if isinstance(plan, dict) else None
if worker.scope_task_id and worker.scope_task_id != task.task_id:
return False
if worker.scope_job_id and job_id and worker.scope_job_id != job_id:
return False
if worker.scope_branch and branch and worker.scope_branch != branch:
return False
if worker.scope_execution_digest and digest and worker.scope_execution_digest != digest:
return False
return True
def worker_has_capabilities(worker: WorkerSnapshot, required: Sequence[str]) -> bool:
have = set(worker.capabilities)
return set(required).issubset(have)
def eligible_workers(task: TaskSnapshot, workers: Sequence[WorkerSnapshot], now: int, heartbeat_ttl: int = 180) -> list[WorkerSnapshot]:
return [
w for w in workers
if worker_is_healthy(w, now, heartbeat_ttl)
and worker_has_capabilities(w, task.required_capabilities)
and worker_scope_matches(w, task)
]
def derive_truth(
admission: AdmissionSnapshot | None,
task: TaskSnapshot | None,
workers: Sequence[WorkerSnapshot],
lease: LeaseSnapshot | None,
git: GitSnapshot | None,
evidence: EvidenceSnapshot | None,
now: int | None = None,
queue_claim_threshold: int = 600,
heartbeat_ttl: int = 180,
) -> tuple[TruthState, list[Finding]]:
now = int(time.time()) if now is None else now
findings: list[Finding] = []
if admission and admission.admission_status == "ADMITTED" and not task:
findings.append(Finding(
"DUAL_PLANE_ADMITTED_TASK_MISSING",
Severity.BLOCKER,
"Admission plane says ADMITTED but TaskPlane has no task.",
{"task_id": admission.task_id, "envelope_id": admission.envelope_id},
"Use durable projection/outbox + exact readback; Admission must not imply execution.",
))
return TruthState.ADMITTED_NOT_PROJECTED, findings
if not task:
return TruthState.UNKNOWN, findings
if admission and admission.grant_expires_at and admission.grant_expires_at <= now and task.status in {"QUEUED", "CLAIMED", "RUNNING"}:
findings.append(Finding(
"GRANT_EXPIRED_BEFORE_EXECUTION",
Severity.BLOCKER,
"Execution state depends on an expired grant.",
{"expires_at": admission.grant_expires_at, "now": now, "task_state": task.status},
"Mint runtime TaskPrincipal just-in-time at claim; do not burn grant TTL while waiting in queue.",
))
if admission and admission.admission_status == "ADMITTED" and task.status in {"EXPIRED", "FAILED", "CANCELLED"}:
findings.append(Finding(
"DUAL_PLANE_TERMINAL_DRIFT",
Severity.MAJOR,
"Admission remains ADMITTED while TaskPlane is terminal.",
{"admission": admission.admission_status, "task": task.status},
"Expose admission_state and execution_state separately; never collapse both into one PASS/complete status.",
))
if task.status == "QUEUED":
elig = eligible_workers(task, workers, now, heartbeat_ttl)
if not elig:
findings.append(Finding(
"QUEUED_WITHOUT_ELIGIBLE_WORKER",
Severity.BLOCKER,
"Task is QUEUED but no healthy, capability- and scope-matching worker exists.",
{
"required_capabilities": list(task.required_capabilities),
"healthy_workers": [w.worker_id for w in workers if worker_is_healthy(w, now, heartbeat_ttl)],
},
"Gate TaskPlane projection on a SchedulabilityProof; keep Admission ADMITTED/WAITING_CAPACITY until claimable.",
))
return TruthState.WAITING_CAPACITY, findings
created = parse_iso_epoch(task.created_at)
if created is not None and now - created > queue_claim_threshold and not task.claimed_by:
findings.append(Finding(
"UNCLAIMED_TIMEOUT",
Severity.BLOCKER,
"QUEUED task exceeded claim SLA without a claim holder.",
{"queued_seconds": now - created, "threshold": queue_claim_threshold},
"Trigger scheduler diagnosis/takeover contract; do not extend TTL blindly.",
))
return TruthState.QUEUED, findings
if task.status == "CLAIMED":
if not task.claimed_by:
findings.append(Finding(
"FAKE_CLAIM_STATE",
Severity.BLOCKER,
"Task says CLAIMED but claimed_by is empty.",
{"task_id": task.task_id},
"Reject state transition unless claim holder + fencing generation are persisted atomically.",
))
if lease is None or lease.holder != task.claimed_by or (lease.expires_at and lease.expires_at <= now):
findings.append(Finding(
"CLAIM_WITHOUT_VALID_LEASE",
Severity.BLOCKER,
"CLAIMED task has no valid matching lease.",
{"claimed_by": task.claimed_by, "lease": asdict(lease) if lease else None},
"Require claim+lease readback before exposing CLAIMED.",
))
return TruthState.CLAIMED, findings
if task.status == "RUNNING":
if not task.claimed_by:
findings.append(Finding("FAKE_RUNNING_NO_CLAIM", Severity.BLOCKER, "RUNNING without claim holder."))
claimed_worker = next((w for w in workers if w.worker_id == task.claimed_by), None)
if not claimed_worker or not worker_is_healthy(claimed_worker, now, heartbeat_ttl):
findings.append(Finding(
"FAKE_RUNNING_NO_HEARTBEAT",
Severity.BLOCKER,
"RUNNING task has no fresh worker heartbeat.",
{"claimed_by": task.claimed_by},
"Transition to recovery only through lease/fencing rules; never report RUNNING from a stale projection.",
))
if lease is None or (lease.expires_at and lease.expires_at <= now):
findings.append(Finding("RUNNING_WITHOUT_VALID_LEASE", Severity.BLOCKER, "RUNNING without valid lease."))
return TruthState.RUNNING, findings
if task.status in {"SUCCEEDED", "COMPLETE"}:
ev = evidence or EvidenceSnapshot()
commit = ev.commit_receipt or {}
before = git.expected_oid if git else None
post = (git.post_oid if git else None) or commit.get("post_oid")
commit_oid = (git.commit_oid if git else None) or commit.get("commit_oid")
if not commit_oid or not post or (before and post == before):
findings.append(Finding(
"TERMINAL_SUCCESS_UNPROVEN",
Severity.BLOCKER,
"Task reports success without a proven branch mutation receipt.",
{"commit_oid": commit_oid, "post_oid": post, "pre_oid": before},
"SUCCEEDED requires lease id, commit receipt, pre/post OID change, output hashes, and runtime/projection readback.",
))
return TruthState.TERMINAL_UNPROVEN, findings
if not ev.runtime_readback:
findings.append(Finding("SUCCESS_WITHOUT_RUNTIME_READBACK", Severity.MAJOR, "Commit exists but runtime readback is absent."))
return TruthState.COMPLETE if not findings else TruthState.TERMINAL_UNPROVEN, findings
if task.status == "EXPIRED":
if task.attempt in (None, 0) and not task.claimed_by:
findings.append(Finding(
"EXPIRED_WITHOUT_ANY_CLAIM",
Severity.MAJOR,
"Task expired without a single execution attempt.",
{"attempt": task.attempt, "claimed_by": task.claimed_by},
"Treat as schedulability failure, not execution failure; fix worker/principal supply before recovery.",
))
return TruthState.EXPIRED, findings
return TruthState.UNKNOWN, findings
def inspect_artifact(path: Path, expected_sha256: str | None = None, min_bytes: int = 8) -> list[Finding]:
findings: list[Finding] = []
if not path.is_file():
return [Finding("ARTIFACT_MISSING", Severity.BLOCKER, f"Artifact missing: {path}")]
data = path.read_bytes()
if len(data) < min_bytes:
findings.append(Finding("ARTIFACT_TOO_SMALL", Severity.BLOCKER, f"Artifact is too small: {path}", {"bytes": len(data)}))
digest = hashlib.sha256(data).hexdigest()
if expected_sha256 and digest != expected_sha256:
findings.append(Finding("ARTIFACT_HASH_MISMATCH", Severity.BLOCKER, f"Artifact hash mismatch: {path}", {"expected": expected_sha256, "actual": digest}))
try:
text = data.decode("utf-8")
except UnicodeDecodeError:
text = ""
if text and any(p.search(text) for p in PLACEHOLDER_PATTERNS):
findings.append(Finding(
"PLACEHOLDER_ARTIFACT",
Severity.BLOCKER,
f"Artifact contains placeholder/stub markers: {path}",
remediation="Reject plan admission until actual bytes exist and match the declared digest.",
))
return findings
def operation_ids_from_openapi(path: Path) -> dict[str, str]:
try:
import yaml # type: ignore
except ImportError as exc:
raise RuntimeError("PyYAML is required for OpenAPI parity checks") from exc
doc = yaml.safe_load(path.read_text(encoding="utf-8-sig"))
result: dict[str, str] = {}
for route, methods in (doc.get("paths") or {}).items():
if not isinstance(methods, dict):
continue
for _, spec in methods.items():
if isinstance(spec, dict) and spec.get("operationId"):
result[str(spec["operationId"])] = str(route)
return result
ROUTE_PAIR_RE = re.compile(r'["\'](?P<route>/v\d+/[^"\']+)["\']\s*:\s*["\'](?P<op>[A-Za-z0-9_]+)["\']')
def operation_ids_from_server_source(path: Path) -> dict[str, str]:
text = path.read_text(encoding="utf-8")
out: dict[str, str] = {}
for m in ROUTE_PAIR_RE.finditer(text):
out[m.group("op")] = m.group("route")
return out
def check_openapi_backend_parity(openapi_path: Path, server_path: Path, required_ops: Iterable[str] = ()) -> list[Finding]:
oa = operation_ids_from_openapi(openapi_path)
be = operation_ids_from_server_source(server_path)
findings: list[Finding] = []
for op in sorted(set(required_ops) | set(oa)):
if op in oa and op not in be:
findings.append(Finding(
"OPENAPI_DEAD_ENDPOINT",
Severity.BLOCKER,
f"OpenAPI advertises {op} ({oa[op]}) but backend route mapping is absent.",
remediation="Deploy projection and backend from the same source commit; reject schema-only rollout.",
))
for op in sorted(set(required_ops) | set(be)):
if op in be and op not in oa:
findings.append(Finding(
"BACKEND_UNPROJECTED_OPERATION",
Severity.MAJOR,
f"Backend implements {op} ({be[op]}) but OpenAPI projection is absent.",
remediation="Generate/validate external projection from the canonical backend route manifest.",
))
return findings
def check_capability_aliases(required: Sequence[str], workers: Sequence[WorkerSnapshot]) -> list[Finding]:
findings: list[Finding] = []
req = set(required)
if "governed.write.files" in req:
for w in workers:
caps = set(w.capabilities)
if "governed.write.file" in caps and "governed.write.files" not in caps:
findings.append(Finding(
"CAPABILITY_SINGULAR_PLURAL_ASYMMETRY",
Severity.MAJOR,
f"Worker {w.worker_id} exposes governed.write.file but task requires governed.write.files.",
{"worker_capabilities": sorted(caps)},
"Do not treat names as aliases implicitly; define one canonical capability or an explicit server-side compatibility mapping with tests.",
))
return findings
def git_ref(git_dir: Path, ref: str) -> str | None:
cp = subprocess.run(
["git", f"--git-dir={git_dir}", "rev-parse", ref],
text=True, capture_output=True
)
if cp.returncode != 0:
return None
value = cp.stdout.strip()
return value if re.fullmatch(r"[0-9a-f]{40}", value) else None
def sqlite_rows(path: Path, query: str, params: Sequence[Any] = ()) -> list[sqlite3.Row]:
uri = f"file:{path}?mode=ro&immutable=1"
con = sqlite3.connect(uri, uri=True)
con.row_factory = sqlite3.Row
try:
return list(con.execute(query, params))
finally:
con.close()
def dump_report(path: Path, report: Mapping[str, Any]) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
path.write_text(json.dumps(report, ensure_ascii=False, indent=2, default=str) + "\n", encoding="utf-8")
@@ -1,27 +1,7 @@
"""Non-sensitive control-plane snapshot assembly.
The base snapshot is a pure field-copy of each plane's reported state. It does
NOT itself assert whether that state is backed by evidence — a drift_status=PASS
or an ADMITTED status is a *claim*, not a proof alert/status change.
`attach_control_plane_truth` (see below) injects the read-only derived `SystemTruth`
(states + findings) from `aios_control_plane_consistency.derive_truth` into the
snapshot so that explain/status surfaces can expose *derived* truth alongside
the reported state, without ever mutating the persistence planes.
"""
"""Non-sensitive control-plane snapshot assembly."""
from __future__ import annotations
from typing import Any, Mapping, Sequence
from aios_control_plane_consistency import (
AdmissionSnapshot,
EvidenceSnapshot,
GitSnapshot,
LeaseSnapshot,
TaskSnapshot,
WorkerSnapshot,
derive_truth,
)
from typing import Any, Mapping
def _copy(value: Mapping[str, Any] | None, keys: tuple[str, ...]) -> dict[str, Any]:
@@ -48,48 +28,3 @@ def build_control_plane_snapshot(*, target_version: str, runtime: Mapping[str, A
def getControlPlaneSnapshot(**kwargs: Any) -> dict[str, Any]:
return build_control_plane_snapshot(**kwargs)
def attach_control_plane_truth(
snapshot: Mapping[str, Any],
*,
admission: AdmissionSnapshot | None = None,
task: TaskSnapshot | None = None,
workers: Sequence[WorkerSnapshot] = (),
lease: LeaseSnapshot | None = None,
git: GitSnapshot | None = None,
evidence: EvidenceSnapshot | None = None,
now: int | None = None,
queue_claim_threshold: int = 600,
heartbeat_ttl: int = 180,
) -> dict[str, Any]:
"""Attach read-only derived SystemTruth to a control-plane snapshot.
This never writes to any persistence plane. It derives a ``system_truth``
block from the already-loaded plane snapshots and adds it under
``snapshot["derived"]``, leaving all existing keys intact.
The block includes the derived truth state, any findings, and a compact
"equivalence" guard (see config/gitops/system_truth_contract.json
``forbidden_equivalences``) so surfaces cannot silently present a local
plane claim as end-to-end evidence.
"""
from dataclasses import asdict
truth_state, findings = derive_truth(
admission, task, workers, lease, git, evidence, now,
queue_claim_threshold=queue_claim_threshold,
heartbeat_ttl=heartbeat_ttl,
)
out = dict(snapshot)
if "derived" not in out:
out["derived"] = {}
blockers = [f for f in findings if f.severity.value == "BLOCKER"]
out["derived"]["system_truth"] = {
"truth_state": truth_state.value,
"findings": [asdict(f) for f in findings],
"blocker_count": sum(1 for f in findings if f.severity.value == "BLOCKER"),
"major_count": sum(1 for f in findings if f.severity.value == "MAJOR"),
# Derivation is read-only; it never confirms completion on its own.
"evidence_flow_confirmed": truth_state.value == "COMPLETE" and not blockers,
}
return out
@@ -664,6 +664,23 @@ class GitOpsJobController:
raise JobError("LEASE_DURATION_EXCEEDED")
return _iso_utc(deadline)
@staticmethod
def internal_job_id_for_idempotency_key(client_idempotency_key: str) -> str:
"""Return the durable Job ID for a client idempotency key."""
if (
not isinstance(client_idempotency_key, str)
or not client_idempotency_key
or len(client_idempotency_key) > 128
or not all(
character.isalnum() or character in "._:-"
for character in client_idempotency_key
)
):
raise JobError("IDEMPOTENCY_KEY_INVALID")
return "job-" + sha256_json({
"client_idempotency_key": client_idempotency_key,
})[:32]
@staticmethod
def _binding(request: Mapping[str, Any]) -> dict[str, str]:
if request.get("schema") == "aios.operation-job.v1":
@@ -767,7 +784,12 @@ class GitOpsJobController:
key = sha256_json({"client_idempotency_key": client_idempotency_key})
else:
key = sha256_json(binding)
job_id = "job-" + key[:32]
job_id = (
self.internal_job_id_for_idempotency_key(client_idempotency_key)
if binding.get("schema") != "aios.operation-job.v1"
and client_idempotency_key is not None
else "job-" + key[:32]
)
durable_request = dict(binding)
if payload_ref is not None:
durable_request.update({"payload_ref": str(payload_ref), "payload_size": int(payload_size)})
+32 -4
View File
@@ -57,10 +57,36 @@ def seed_base_models():
if not one('SELECT * FROM gitops_versions WHERE version=?', ('V1R1C0',)):
exec_sql('INSERT INTO gitops_versions(version,tag,level,summary,status,created_by,created_at) VALUES(?,?,?,?,?,?,?)', ('V1R1C0','V1R1C0','R','MultiUser runtime selector + GitOps + custom LLM priority registry','ready','system',now()))
_ROLE_WHITELIST = {'owner', 'admin', 'user'}
def token_identity(token: str):
"""Resolve the server-authoritative identity bound to a token.
Format of each TOKEN_FILE line (one per line, ':'-separated):
<token> -> authenticated principal, role 'user'
<token>:<user>:<role> -> mapped principal; role restricted to owner/admin/user
Identity NEVER comes from a client-supplied header. In local dev mode the
DEV_ALLOW flag maps a fixed dev principal (must be explicitly enabled).
"""
if DEV_ALLOW:
return {'username': 'dev-owner', 'role': 'owner'}
if not token or not TOKEN_FILE.exists():
return None
for line in TOKEN_FILE.read_text().splitlines():
line = line.strip()
if not line:
continue
parts = [x.strip() for x in line.split(':')]
if not parts or parts[0] != token:
continue
if len(parts) >= 3:
role = parts[2] if parts[2] in _ROLE_WHITELIST else 'user'
return {'username': safe_name(parts[1]), 'role': role}
return {'username': 'user-%s' % token[:8], 'role': 'user'}
return None
def valid_token(token: str) -> bool:
if DEV_ALLOW: return True
if not token or not TOKEN_FILE.exists(): return False
return token.strip() in [x.strip() for x in TOKEN_FILE.read_text().splitlines() if x.strip()]
return token_identity(token) is not None
def git_cmd(args, check=False):
WORKSPACE.mkdir(parents=True, exist_ok=True); p = subprocess.run(['git'] + args, cwd=str(WORKSPACE), text=True, capture_output=True)
if check and p.returncode != 0: raise RuntimeError(p.stderr or p.stdout or 'git command failed')
@@ -90,8 +116,10 @@ class Handler(BaseHTTPRequestHandler):
server_version = 'AIOSGitOpsModelAPI/V1R1C0'
def actor(self):
auth = self.headers.get('Authorization',''); token = auth.replace('Bearer ', '').strip() if auth.startswith('Bearer ') else self.headers.get('X-AIOS-Token','')
ok = valid_token(token); user = self.headers.get('X-AIOS-User') or ('dev-owner' if DEV_ALLOW else 'anonymous'); role = self.headers.get('X-AIOS-Role') or ('owner' if DEV_ALLOW else 'user')
return {'username': safe_name(user), 'role': role, 'token_ok': ok}
ident = token_identity(token)
if ident is None:
return {'username': 'anonymous', 'role': 'user', 'token_ok': False}
return {'username': ident['username'], 'role': ident['role'], 'token_ok': True}
def require_write(self, roles=None):
a = self.actor()
if not a['token_ok']: self.send_json({'error':'unauthorized','hint':'write APIs require Bearer token or AIOS_GITOPS_DEV_ALLOW=1'}, 401); return None
@@ -0,0 +1,890 @@
"""Read-only four-authority evidence provider for one canonical C04 attachment.
This module never creates, claims, resumes, reconciles, or mutates a Task or
Job. The expired Owner epoch is verified as historical provenance only.
"""
from __future__ import annotations
import hashlib
import json
import os
import sqlite3
import stat
import time
import urllib.error
import urllib.parse
import urllib.request
from dataclasses import dataclass
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Callable, Mapping
from app.aios_gitops_job_identity import legacy_task_storage_key, task_storage_key
from app.aios_single_host_authority import (
DEFAULT_OWNER_SIGNER_ID,
OWNER_NAMESPACE,
RegistryOwnerApprovalVerifier,
SignerRegistry,
)
from app.task_plane.legacy_attachment import (
ATTACH_EVIDENCE_SCHEMA,
AuthorityReadbackRef,
SOURCE_KIND,
VerifiedLegacyAttachmentEvidence,
)
class LegacyAttachmentRuntimeError(RuntimeError):
"""Stable fail-closed readback error."""
def _fail(code: str) -> None:
raise LegacyAttachmentRuntimeError(code)
def _strict_object(pairs: list[tuple[str, Any]]) -> dict[str, Any]:
value: dict[str, Any] = {}
for key, item in pairs:
if key in value:
_fail("DUPLICATE_JSON_FIELD")
value[key] = item
return value
def _json_object(raw: bytes | str, code: str) -> dict[str, Any]:
try:
if isinstance(raw, bytes):
raw = raw.decode("utf-8")
value = json.loads(raw, object_pairs_hook=_strict_object)
except (UnicodeDecodeError, json.JSONDecodeError) as exc:
raise LegacyAttachmentRuntimeError(code) from exc
if not isinstance(value, dict):
_fail(code)
return value
def _canonical(value: object) -> bytes:
try:
return json.dumps(
value,
ensure_ascii=False,
sort_keys=True,
separators=(",", ":"),
allow_nan=False,
).encode("utf-8")
except (TypeError, ValueError) as exc:
raise LegacyAttachmentRuntimeError("AUTHORITY_JSON_INVALID") from exc
def _sha(value: object) -> str:
return hashlib.sha256(_canonical(value)).hexdigest()
def _hex(value: object, code: str) -> str:
if (
not isinstance(value, str)
or len(value) != 64
or any(ch not in "0123456789abcdef" for ch in value)
):
_fail(code)
return value
def _text(value: object, code: str) -> str:
if not isinstance(value, str) or not value or value != value.strip():
_fail(code)
return value
def _integer(value: object, code: str) -> int:
if type(value) is not int or value < 0:
_fail(code)
return value
def _iso(value: object, code: str) -> str:
text = _text(value, code)
try:
parsed = datetime.fromisoformat(text.replace("Z", "+00:00"))
except ValueError as exc:
raise LegacyAttachmentRuntimeError(code) from exc
if parsed.tzinfo is None or parsed.utcoffset() is None:
_fail(code)
return parsed.astimezone(timezone.utc).isoformat().replace("+00:00", "Z")
def _observed(clock: Callable[[], float]) -> str:
return datetime.fromtimestamp(float(clock()), timezone.utc).isoformat().replace(
"+00:00", "Z"
)
def _mapping(value: object, code: str) -> dict[str, Any]:
if not isinstance(value, Mapping):
_fail(code)
return dict(value)
def _secure_file(
path: Path, *, require_root_owner: bool, writable_authority: bool = False
) -> os.stat_result:
try:
info = path.lstat()
except OSError as exc:
raise LegacyAttachmentRuntimeError("AUTHORITY_FILE_UNAVAILABLE") from exc
if path.is_symlink() or not stat.S_ISREG(info.st_mode):
_fail("AUTHORITY_FILE_INVALID")
if os.name == "posix":
if not writable_authority and stat.S_IMODE(info.st_mode) & 0o022:
_fail("AUTHORITY_FILE_WRITABLE")
if require_root_owner and info.st_uid != 0:
_fail("AUTHORITY_FILE_OWNER_INVALID")
return info
def _secure_directory(path: Path) -> os.stat_result:
try:
info = path.lstat()
except OSError as exc:
raise LegacyAttachmentRuntimeError("AUTHORITY_DIRECTORY_UNAVAILABLE") from exc
if path.is_symlink() or not stat.S_ISDIR(info.st_mode):
_fail("AUTHORITY_DIRECTORY_INVALID")
return info
def _stable_read(path: Path, *, require_root_owner: bool, maximum: int = 2_000_000) -> bytes:
before = _secure_file(path, require_root_owner=require_root_owner)
if before.st_size > maximum:
_fail("AUTHORITY_FILE_TOO_LARGE")
try:
raw = path.read_bytes()
except OSError as exc:
raise LegacyAttachmentRuntimeError("AUTHORITY_FILE_UNAVAILABLE") from exc
after = _secure_file(path, require_root_owner=require_root_owner)
if (before.st_dev, before.st_ino, before.st_size, before.st_mtime_ns) != (
after.st_dev,
after.st_ino,
after.st_size,
after.st_mtime_ns,
):
_fail("AUTHORITY_FILE_CHANGED")
return raw
def _ro_connection(path: Path, *, require_root_owner: bool) -> tuple[sqlite3.Connection, os.stat_result]:
before = _secure_file(
path, require_root_owner=require_root_owner, writable_authority=True
)
try:
resolved = path.resolve(strict=True)
connection = sqlite3.connect(
resolved.as_uri() + "?mode=ro", uri=True, timeout=5.0
)
connection.row_factory = sqlite3.Row
connection.execute("PRAGMA query_only=ON")
connection.execute("PRAGMA foreign_keys=ON")
connection.execute("BEGIN")
return connection, before
except (OSError, sqlite3.Error) as exc:
raise LegacyAttachmentRuntimeError("AUTHORITY_DB_UNAVAILABLE") from exc
def _finish_ro(connection: sqlite3.Connection, path: Path, before: os.stat_result) -> None:
connection.close()
after = _secure_file(
path, require_root_owner=False, writable_authority=True
)
if (before.st_dev, before.st_ino) != (after.st_dev, after.st_ino):
_fail("AUTHORITY_DB_CHANGED")
def _columns(connection: sqlite3.Connection, table: str) -> tuple[str, ...]:
return tuple(str(row[1]) for row in connection.execute(f"PRAGMA table_info({table})"))
@dataclass(frozen=True, slots=True)
class LegacyC04Binding:
mission_id: str
root_task_id: str
anchor_task_id: str
owner_principal_id: str
original_acceptance_digest: str
source_task_id: str
external_job_id: str
def validate(self) -> None:
for value in (
self.mission_id,
self.root_task_id,
self.anchor_task_id,
self.owner_principal_id,
self.source_task_id,
self.external_job_id,
):
_text(value, "C04_BINDING_INVALID")
_hex(self.original_acceptance_digest, "C04_ACCEPTANCE_DIGEST_INVALID")
@dataclass(frozen=True, slots=True)
class CredentialFileBearerSupplier:
"""Read a systemd credential for one call without caching or displaying it."""
path: Path
def __post_init__(self) -> None:
if not self.path.is_absolute():
_fail("AGENT_TASK_CREDENTIAL_PATH_INVALID")
if os.name == "posix":
try:
self.path.resolve(strict=False).relative_to(Path("/run/credentials"))
except ValueError as exc:
raise LegacyAttachmentRuntimeError(
"AGENT_TASK_CREDENTIAL_PATH_INVALID"
) from exc
def __call__(self) -> str:
raw = _stable_read(self.path, require_root_owner=False, maximum=8192)
try:
value = raw.decode("utf-8").strip()
except UnicodeDecodeError as exc:
raise LegacyAttachmentRuntimeError(
"AGENT_TASK_CREDENTIAL_INVALID"
) from exc
if len(value) < 32 or len(value) > 4096:
_fail("AGENT_TASK_CREDENTIAL_INVALID")
return value
class _NoRedirect(urllib.request.HTTPRedirectHandler):
def redirect_request(self, req, fp, code, msg, headers, newurl): # noqa: ANN001
return None
_AGENT_TASK_COLUMNS = (
"task_id",
"goal",
"instructions",
"success_criteria_json",
"required_capabilities_json",
"constraints_json",
"metadata_json",
"deadline",
"priority",
"max_attempts",
"attempt",
"status",
"claimed_by",
"lease_token_hash",
"lease_expires_at",
"result_json",
"result_digest",
"result_idempotency_key",
"verification_json",
"revision",
"created_at",
"updated_at",
)
@dataclass(slots=True)
class AuthenticatedAgentTaskReader:
"""Authenticated non-mutating proof + read-only authority-row snapshot.
The authenticated HTTP request is intentionally sent to the task events
endpoint because the historical GET /v1/tasks/{id} path calls the global
lease-expiry mutator. The task row itself is read from the authoritative
SQLite database using query_only mode and an exact schema allowlist.
"""
base_url: str
bearer_supplier: Callable[[], str]
task_db_path: Path
timeout_seconds: float = 10.0
clock: Callable[[], float] = time.time
require_db_root_owner: bool = False
def __post_init__(self) -> None:
parsed = urllib.parse.urlsplit(self.base_url)
if (
parsed.scheme != "http"
or parsed.hostname not in {"127.0.0.1", "localhost", "::1"}
or parsed.username is not None
or parsed.password is not None
or parsed.query
or parsed.fragment
):
_fail("AGENT_TASK_URL_INVALID")
if not self.task_db_path.is_absolute():
_fail("AGENT_TASK_DB_PATH_INVALID")
self.base_url = self.base_url.rstrip("/")
def _authenticated_events(self, task_id: str) -> tuple[list[dict[str, Any]], str]:
token = self.bearer_supplier()
if not isinstance(token, str) or len(token.strip()) < 32:
_fail("AGENT_TASK_CREDENTIAL_INVALID")
url = (
self.base_url
+ "/v1/tasks/"
+ urllib.parse.quote(task_id, safe="._:-")
+ "/events"
)
request = urllib.request.Request(
url,
method="GET",
headers={
"Authorization": "Bearer " + token.strip(),
"Accept": "application/json",
},
)
try:
with urllib.request.build_opener(_NoRedirect).open(
request, timeout=self.timeout_seconds
) as response:
if response.status != 200 or response.geturl() != url:
_fail("AGENT_TASK_EVENTS_REJECTED")
raw = response.read(2_000_001)
except (urllib.error.URLError, TimeoutError, OSError) as exc:
raise LegacyAttachmentRuntimeError(
"AGENT_TASK_EVENTS_UNAVAILABLE"
) from exc
finally:
token = ""
if len(raw) > 2_000_000:
_fail("AGENT_TASK_EVENTS_TOO_LARGE")
try:
value = json.loads(raw.decode("utf-8"), object_pairs_hook=_strict_object)
except (UnicodeDecodeError, json.JSONDecodeError, ValueError) as exc:
raise LegacyAttachmentRuntimeError("AGENT_TASK_EVENTS_INVALID") from exc
if not isinstance(value, list):
_fail("AGENT_TASK_EVENTS_INVALID")
events: list[dict[str, Any]] = []
for event in value:
if not isinstance(event, Mapping) or event.get("task_id") != task_id:
_fail("AGENT_TASK_EVENTS_INVALID")
events.append(dict(event))
return events, url
@staticmethod
def _task_projection(row: sqlite3.Row) -> dict[str, Any]:
try:
body = {
"task_id": row["task_id"],
"goal": row["goal"],
"instructions": row["instructions"],
"success_criteria": json.loads(row["success_criteria_json"]),
"required_capabilities": json.loads(row["required_capabilities_json"]),
"constraints": json.loads(row["constraints_json"]),
"metadata": json.loads(row["metadata_json"]),
"deadline": row["deadline"],
"priority": row["priority"],
"max_attempts": row["max_attempts"],
"attempt": row["attempt"],
"status": row["status"],
"claimed_by": row["claimed_by"],
"lease_expires_at": row["lease_expires_at"],
"revision": row["revision"],
"created_at": row["created_at"],
"updated_at": row["updated_at"],
"result": json.loads(row["result_json"]) if row["result_json"] else None,
"verification": (
json.loads(row["verification_json"])
if row["verification_json"]
else None
),
}
except (KeyError, TypeError, json.JSONDecodeError) as exc:
raise LegacyAttachmentRuntimeError(
"AGENT_TASK_ROW_INVALID"
) from exc
_mapping(body["constraints"], "AGENT_TASK_CONSTRAINTS_INVALID")
_mapping(body["metadata"], "AGENT_TASK_METADATA_INVALID")
_integer(body["revision"], "AGENT_TASK_REVISION_INVALID")
if body["status"] not in {"SUCCEEDED", "FAILED", "EXPIRED", "CANCELLED"}:
_fail("AGENT_TASK_NOT_TERMINAL")
return body
def read(self, task_id: str) -> dict[str, Any]:
events, events_url = self._authenticated_events(task_id)
connection, before = _ro_connection(
self.task_db_path, require_root_owner=self.require_db_root_owner
)
try:
if _columns(connection, "tasks") != _AGENT_TASK_COLUMNS:
_fail("AGENT_TASK_DB_SCHEMA_INVALID")
row = connection.execute(
"SELECT * FROM tasks WHERE task_id=? LIMIT 1", (task_id,)
).fetchone()
if row is None:
_fail("AGENT_TASK_NOT_FOUND")
body = self._task_projection(row)
finally:
_finish_ro(connection, self.task_db_path, before)
event_digest = _sha(events)
return {
"body": body,
"observed_at": _observed(self.clock),
"readback_ref": (
events_url
+ "#authenticated-events-sha256="
+ event_digest
+ ";task-db-revision="
+ str(body["revision"])
),
"canonical_sha256": _sha(body),
"authenticated_events_sha256": event_digest,
}
_ADMISSION_COLUMNS = {
"task_admission_envelopes": (
"idempotency_key", "envelope_id", "request_digest", "envelope_json", "status", "created_at"
),
"task_admission_grants": (
"envelope_id", "grant_id", "grant_digest", "grant_json", "authorization_digest", "created_at"
),
"task_admission_receipts": ("envelope_id", "task_id", "receipt_json", "created_at"),
}
@dataclass(slots=True)
class AdmissionEpochRegistryReader:
db_path: Path
epoch_root: Path
revocation_file: Path
registry_path: Path
require_db_root_owner: bool = False
require_static_root_owner: bool = True
clock: Callable[[], float] = time.time
def __post_init__(self) -> None:
for path in (self.db_path, self.epoch_root, self.revocation_file, self.registry_path):
if not path.is_absolute():
_fail("AUTHORITY_PATH_NOT_ABSOLUTE")
def read(self, task_id: str) -> dict[str, Any]:
connection, before = _ro_connection(
self.db_path, require_root_owner=self.require_db_root_owner
)
try:
if any(_columns(connection, table) != columns for table, columns in _ADMISSION_COLUMNS.items()):
_fail("ADMISSION_SCHEMA_INVALID")
rows = connection.execute(
"SELECT e.*,g.grant_id,g.grant_digest,g.grant_json,g.authorization_digest,"
"g.created_at AS grant_created_at,r.task_id,r.receipt_json,"
"r.created_at AS receipt_created_at FROM task_admission_receipts r "
"JOIN task_admission_envelopes e ON e.envelope_id=r.envelope_id "
"JOIN task_admission_grants g ON g.envelope_id=r.envelope_id WHERE r.task_id=?",
(task_id,),
).fetchall()
if len(rows) != 1:
_fail("ADMISSION_READBACK_AMBIGUOUS")
row = dict(rows[0])
except sqlite3.Error as exc:
raise LegacyAttachmentRuntimeError("ADMISSION_READBACK_FAILED") from exc
finally:
_finish_ro(connection, self.db_path, before)
envelope = _json_object(row["envelope_json"], "ADMISSION_ENVELOPE_INVALID")
grant = _json_object(row["grant_json"], "ADMISSION_GRANT_INVALID")
receipt = _json_object(row["receipt_json"], "ADMISSION_RECEIPT_INVALID")
envelope_id = _text(row["envelope_id"], "ADMISSION_ENVELOPE_INVALID")
grant_id = _text(row["grant_id"], "ADMISSION_GRANT_INVALID")
epoch_id = _text(grant.get("parent_epoch_id"), "ADMISSION_EPOCH_ID_INVALID")
if (
row.get("status") != "ADMITTED"
or envelope.get("schema") != "aios.task-admission-envelope.v1"
or grant.get("schema") != "aios.task-admission-grant.v1"
or receipt.get("schema") != "aios.task-admission-receipt.v1"
or envelope.get("envelope_id") != envelope_id
or grant.get("envelope_id") != envelope_id
or receipt.get("envelope_id") != envelope_id
or envelope.get("task_id") != task_id
or grant.get("task_id") != task_id
or receipt.get("task_id") != task_id
or grant.get("grant_id") != grant_id
or receipt.get("grant_id") != grant_id
or grant.get("authorization_kind") != "delegation_epoch"
or receipt.get("authorization_kind") != "delegation_epoch"
or receipt.get("parent_epoch_id") != epoch_id
):
_fail("ADMISSION_BINDING_INVALID")
request_value = {
key: envelope.get(key)
for key in (
"task_id", "repository", "authority_base_oid", "target_branch",
"generation", "execution_digest", "requested_capabilities",
"file_allowlist", "risk", "trust_domain", "policy_digest",
)
}
if envelope.get("request_digest") != _sha(request_value) or row.get("request_digest") != _sha(request_value):
_fail("ADMISSION_REQUEST_DIGEST_INVALID")
grant_digest = _sha(grant)
if row.get("grant_digest") != grant_digest or receipt.get("grant_digest_sha256") != grant_digest:
_fail("ADMISSION_GRANT_DIGEST_INVALID")
auth_value = {
"envelope_id": envelope_id,
"authorization_kind": "delegation_epoch",
"delegation_epoch_id": epoch_id,
}
if row.get("authorization_digest") != _sha(auth_value):
_fail("ADMISSION_AUTHORIZATION_DIGEST_INVALID")
linked = (
"repository", "authority_base_oid", "target_branch", "generation",
"execution_digest", "policy_digest", "trust_domain", "risk",
)
if any(grant.get(key) != envelope.get(key) for key in linked):
_fail("ADMISSION_GRANT_SCOPE_INVALID")
if grant.get("capabilities") != envelope.get("requested_capabilities") or grant.get("file_allowlist") != envelope.get("file_allowlist"):
_fail("ADMISSION_GRANT_SCOPE_INVALID")
if receipt.get("execution_digest") != grant.get("execution_digest"):
_fail("ADMISSION_RECEIPT_SCOPE_INVALID")
_secure_directory(self.epoch_root)
canonical_path = self.epoch_root / "owner-delegation-epoch.canonical.json"
signature_path = self.epoch_root / "owner-delegation-epoch.sig"
key_path = self.epoch_root / "owner-delegation-epoch.key-id"
canonical = _stable_read(canonical_path, require_root_owner=self.require_static_root_owner)
epoch = _json_object(canonical, "OWNER_EPOCH_INVALID")
if _canonical(epoch) != canonical:
_fail("OWNER_EPOCH_NOT_CANONICAL")
signature = _stable_read(signature_path, require_root_owner=self.require_static_root_owner).decode("utf-8")
key_id = _stable_read(key_path, require_root_owner=self.require_static_root_owner).decode("utf-8").strip()
registry_raw = _stable_read(self.registry_path, require_root_owner=self.require_static_root_owner)
registry_json = _json_object(registry_raw, "OWNER_REGISTRY_INVALID")
revocations = _json_object(
_stable_read(self.revocation_file, require_root_owner=self.require_static_root_owner),
"OWNER_REVOCATIONS_INVALID",
)
if (
set(revocations) != {"schema", "revoked_epoch_ids"}
or revocations.get("schema") != "aios.owner-delegation-revocations.v1"
or not isinstance(revocations.get("revoked_epoch_ids"), list)
or epoch_id in revocations["revoked_epoch_ids"]
):
_fail("OWNER_EPOCH_REVOKED")
try:
registry = SignerRegistry(
self.registry_path, require_root_owner=self.require_static_root_owner
)
record = registry.require(DEFAULT_OWNER_SIGNER_ID, role="OWNER_AUTHORITY")
verified = RegistryOwnerApprovalVerifier(
self.registry_path,
require_root_owner=self.require_static_root_owner,
signer_id=DEFAULT_OWNER_SIGNER_ID,
).verify(
namespace=OWNER_NAMESPACE,
payload=canonical,
signature=signature,
key_id=key_id,
)
except Exception as exc:
raise LegacyAttachmentRuntimeError("OWNER_REGISTRY_VERIFICATION_FAILED") from exc
now = int(self.clock())
provenance_checks = {
"signature": verified is True,
"fingerprint": record.fingerprint == key_id,
"schema": epoch.get("schema") == "aios.owner-delegation-epoch.v1",
"epoch_id": epoch.get("epoch_id") == epoch_id,
"owner": epoch.get("owner_identity") == grant.get("owner_identity"),
"repository": epoch.get("repository") == grant.get("repository"),
"trust_domain": epoch.get("trust_domain") == grant.get("trust_domain"),
"policy": epoch.get("policy_digest") == grant.get("policy_digest"),
"revocation_epoch": epoch.get("revocation_epoch") == 0,
"expires_type": type(epoch.get("expires_at")) is int,
"not_before_type": type(epoch.get("not_before")) is int,
}
failed = [name for name, passed in provenance_checks.items() if not passed]
if failed:
_fail("OWNER_EPOCH_PROVENANCE_INVALID:" + ",".join(sorted(failed)))
if (
now < epoch["expires_at"]
or type(epoch.get("issued_at")) is not int
or epoch["issued_at"] > grant.get("issued_at", 0)
or grant.get("issued_at", 0) < epoch["not_before"]
or grant.get("expires_at", 0) > epoch["expires_at"]
):
_fail("OWNER_EPOCH_PROVENANCE_INVALID:temporal")
if not any(str(grant["target_branch"]).startswith(str(prefix)) for prefix in epoch.get("allowed_branch_prefixes", [])):
_fail("OWNER_EPOCH_SCOPE_INVALID")
if not set(grant.get("capabilities", [])).issubset(set(epoch.get("allowed_capabilities", []))):
_fail("OWNER_EPOCH_SCOPE_INVALID")
risk_rank = {"low": 0, "medium": 1, "high": 2}
if (
grant.get("risk") not in risk_rank
or epoch.get("risk_ceiling") not in risk_rank
or risk_rank[grant["risk"]] > risk_rank[epoch["risk_ceiling"]]
):
_fail("OWNER_EPOCH_SCOPE_INVALID")
if any(
not any(path == prefix.rstrip("/") or path.startswith(prefix.rstrip("/") + "/") for prefix in epoch.get("allowed_file_prefixes", []))
for path in grant.get("file_allowlist", [])
):
_fail("OWNER_EPOCH_SCOPE_INVALID")
admission_projection = {
"envelope": envelope,
"grant": grant,
"receipt": receipt,
"request_digest": row["request_digest"],
"grant_digest": row["grant_digest"],
"authorization_digest": row["authorization_digest"],
"envelope_status": row["status"],
}
owner_projection = {
"epoch": epoch,
"epoch_sha256": hashlib.sha256(canonical).hexdigest(),
"epoch_key_id": key_id,
"signature_verified": True,
"epoch_currently_valid": False,
"revocations": revocations,
"registry_schema": registry_json.get("schema"),
"registry_generation": registry.authority_generation,
"signer": record.to_dict(),
}
observed = _observed(self.clock)
return {
"envelope": envelope,
"grant": grant,
"receipt": receipt,
"epoch": epoch,
"epoch_digest": hashlib.sha256(canonical).hexdigest(),
"signer_id": record.signer_id,
"signer_fingerprint": record.fingerprint,
"registry_generation": registry.authority_generation,
"observed_at": observed,
"admission_ref": f"sqlite:{self.db_path}#task={task_id};envelope={envelope_id}",
"admission_sha256": _sha(admission_projection),
"registry_ref": f"file:{self.registry_path}#generation={registry.authority_generation};epoch={epoch_id}",
"registry_sha256": _sha(owner_projection),
}
_JOB_COLUMNS = {
"job_aliases": (
"external_job_id", "internal_job_id", "task_id", "repository", "target_branch",
"branch_mode", "generation", "commit_idempotency_key", "expected_branch_oid",
"request_digest", "adopted_legacy", "created_at",
),
"jobs": (
"job_id", "idempotency_key", "campaign_id", "repository", "operation", "branch",
"expected_branch_oid", "payload_digest", "payload_ref", "payload_size", "state",
"side_effect_state", "revision", "request_json", "receipt_json", "created_at", "updated_at",
),
"idempotency_keys": ("key", "job_id", "binding_json", "created_at"),
}
@dataclass(slots=True)
class BridgeJobSQLiteReader:
jobs_root: Path
require_root_owner: bool = False
clock: Callable[[], float] = time.time
def __post_init__(self) -> None:
if not self.jobs_root.is_absolute():
_fail("BRIDGE_JOB_ROOT_NOT_ABSOLUTE")
def read(self, task_id: str, external_job_id: str) -> dict[str, Any]:
exact_root = self.jobs_root / task_storage_key(task_id)
legacy_root = self.jobs_root / legacy_task_storage_key(task_id)
db_path = exact_root / "state.sqlite3"
if legacy_root != exact_root and legacy_root.exists():
_fail("BRIDGE_JOB_STORE_AMBIGUOUS")
_secure_directory(self.jobs_root)
_secure_directory(exact_root)
try:
if db_path.resolve(strict=True).parent != exact_root.resolve(strict=True):
_fail("BRIDGE_JOB_STORE_INVALID")
except OSError as exc:
raise LegacyAttachmentRuntimeError("BRIDGE_JOB_STORE_INVALID") from exc
connection, before = _ro_connection(
db_path, require_root_owner=self.require_root_owner
)
try:
if any(_columns(connection, table) != columns for table, columns in _JOB_COLUMNS.items()):
_fail("BRIDGE_JOB_SCHEMA_INVALID")
aliases = connection.execute(
"SELECT * FROM job_aliases WHERE external_job_id=? AND task_id=?",
(external_job_id, task_id),
).fetchall()
if len(aliases) != 1:
_fail("BRIDGE_JOB_ALIAS_AMBIGUOUS")
alias = dict(aliases[0])
job_rows = connection.execute(
"SELECT * FROM jobs WHERE job_id=?", (alias["internal_job_id"],)
).fetchall()
key_rows = connection.execute(
"SELECT * FROM idempotency_keys WHERE job_id=?", (alias["internal_job_id"],)
).fetchall()
if len(job_rows) != 1 or len(key_rows) != 1:
_fail("BRIDGE_JOB_BINDING_MISSING")
job, key = dict(job_rows[0]), dict(key_rows[0])
except sqlite3.Error as exc:
raise LegacyAttachmentRuntimeError("BRIDGE_JOB_READBACK_FAILED") from exc
finally:
_finish_ro(connection, db_path, before)
request = _json_object(job.get("request_json") or "{}", "BRIDGE_JOB_REQUEST_INVALID")
binding = _mapping(request.get("binding", request), "BRIDGE_JOB_REQUEST_INVALID")
key_binding = _json_object(key.get("binding_json") or "{}", "BRIDGE_JOB_KEY_BINDING_INVALID")
external_commit_key = _text(alias.get("commit_idempotency_key"), "BRIDGE_EXTERNAL_IDEMPOTENCY_INVALID")
internal_key = _text(job.get("idempotency_key"), "BRIDGE_INTERNAL_IDEMPOTENCY_INVALID")
if (
alias.get("external_job_id") != external_job_id
or alias.get("task_id") != task_id
or job.get("job_id") != alias.get("internal_job_id")
or job.get("campaign_id") != task_id
or key.get("job_id") != job.get("job_id")
or key.get("key") != internal_key
or binding.get("task_id", binding.get("campaign_id")) != task_id
or binding.get("external_job_id") != external_job_id
or binding.get("idempotency_key") != external_commit_key
or key_binding.get("task_id", key_binding.get("campaign_id")) != task_id
or key_binding.get("external_job_id") != external_job_id
or key_binding.get("idempotency_key") != external_commit_key
or alias.get("repository") != job.get("repository")
or alias.get("repository") != binding.get("repository")
or alias.get("target_branch") != job.get("branch")
or alias.get("target_branch") != binding.get("branch", binding.get("target_branch"))
):
_fail("BRIDGE_JOB_BINDING_INVALID")
try:
if int(alias.get("generation")) != int(binding.get("generation")):
_fail("BRIDGE_JOB_BINDING_INVALID")
except (TypeError, ValueError) as exc:
raise LegacyAttachmentRuntimeError("BRIDGE_JOB_BINDING_INVALID") from exc
projection = {
"alias": alias,
"job": job,
"idempotency": key,
"external_commit_idempotency_key": external_commit_key,
"internal_idempotency_key": internal_key,
}
return {
**projection,
"observed_at": _observed(self.clock),
"readback_ref": f"sqlite:{db_path}#external={external_job_id};internal={job['job_id']}",
"canonical_sha256": _sha(projection),
}
@dataclass(slots=True)
class LegacyAttachmentRuntimeProvider:
binding: LegacyC04Binding
agent_task: AuthenticatedAgentTaskReader
admission: AdmissionEpochRegistryReader
bridge_job: BridgeJobSQLiteReader
def __post_init__(self) -> None:
self.binding.validate()
def read_verified(
self,
*,
mission_id: str,
root_task_id: str,
anchor_task_id: str,
source_kind: str,
source_task_id: str,
external_job_id: str,
original_acceptance_digest: str,
) -> VerifiedLegacyAttachmentEvidence:
expected = self.binding
if (
mission_id != expected.mission_id
or root_task_id != expected.root_task_id
or anchor_task_id != expected.anchor_task_id
or source_kind != SOURCE_KIND
or source_task_id != expected.source_task_id
or external_job_id != expected.external_job_id
or original_acceptance_digest != expected.original_acceptance_digest
):
_fail("C04_AUTHORITY_BINDING_MISMATCH")
task_read = self.agent_task.read(source_task_id)
admission_read = self.admission.read(source_task_id)
job_read = self.bridge_job.read(source_task_id, external_job_id)
task = task_read["body"]
metadata = _mapping(task["metadata"], "AGENT_TASK_METADATA_INVALID")
constraints = _mapping(task["constraints"], "AGENT_TASK_CONSTRAINTS_INVALID")
execution = _mapping(metadata.get("execution"), "AGENT_TASK_EXECUTION_INVALID")
grant = admission_read["grant"]
envelope = admission_read["envelope"]
alias, job = job_read["alias"], job_read["job"]
try:
immutable_metadata = _json_object(
constraints.get("immutable_metadata", ""), "AGENT_TASK_IMMUTABLE_METADATA_INVALID"
)
except LegacyAttachmentRuntimeError:
raise
if (
immutable_metadata != metadata
or metadata.get("grant_id") != grant.get("grant_id")
or metadata.get("grant_digest_sha256") != admission_read["receipt"].get("grant_digest_sha256")
or metadata.get("execution_digest") != grant.get("execution_digest")
or execution.get("execution_digest") != grant.get("execution_digest")
or execution.get("external_job_id") != external_job_id
or execution.get("task_id") != source_task_id
or metadata.get("repository") != grant.get("repository")
or metadata.get("target_branch") != grant.get("target_branch")
or metadata.get("generation") != grant.get("generation")
or constraints.get("repository") != grant.get("repository")
or constraints.get("target_branch") != grant.get("target_branch")
or constraints.get("generation") != grant.get("generation")
or envelope.get("task_id") != source_task_id
or alias.get("repository") != grant.get("repository")
or alias.get("target_branch") != grant.get("target_branch")
or alias.get("generation") != grant.get("generation")
or admission_read["signer_id"] != expected.owner_principal_id
):
_fail("FOUR_AUTHORITY_BINDING_INVALID")
return VerifiedLegacyAttachmentEvidence(
schema=ATTACH_EVIDENCE_SCHEMA,
source_kind=SOURCE_KIND,
source_task_id=source_task_id,
source_task_created_at=_iso(task.get("created_at"), "AGENT_TASK_CREATED_AT_INVALID"),
source_task_status=_text(task.get("status"), "AGENT_TASK_STATUS_INVALID"),
source_task_revision=_integer(task.get("revision"), "AGENT_TASK_REVISION_INVALID"),
source_task_attempt=_integer(task.get("attempt"), "AGENT_TASK_ATTEMPT_INVALID"),
source_task_max_attempts=_integer(task.get("max_attempts"), "AGENT_TASK_MAX_ATTEMPTS_INVALID"),
external_job_id=external_job_id,
internal_job_id=_text(job.get("job_id"), "BRIDGE_INTERNAL_JOB_INVALID"),
internal_idempotency_key=_text(job.get("idempotency_key"), "BRIDGE_INTERNAL_IDEMPOTENCY_INVALID"),
job_branch=_text(job.get("branch"), "BRIDGE_JOB_BRANCH_INVALID"),
job_state=_text(job.get("state"), "BRIDGE_JOB_STATE_INVALID"),
job_revision=_integer(job.get("revision"), "BRIDGE_JOB_REVISION_INVALID"),
side_effect_state=_text(job.get("side_effect_state"), "BRIDGE_SIDE_EFFECT_INVALID"),
grant_id=_text(grant.get("grant_id"), "ADMISSION_GRANT_INVALID"),
grant_digest=_hex(admission_read["receipt"].get("grant_digest_sha256"), "ADMISSION_GRANT_DIGEST_INVALID"),
execution_digest=_hex(grant.get("execution_digest"), "ADMISSION_EXECUTION_DIGEST_INVALID"),
epoch_id=_text(admission_read["epoch"].get("epoch_id"), "OWNER_EPOCH_ID_INVALID"),
epoch_digest=_hex(admission_read["epoch_digest"], "OWNER_EPOCH_DIGEST_INVALID"),
historical_signer_id=admission_read["signer_id"],
historical_signer_fingerprint=admission_read["signer_fingerprint"],
registry_generation=_integer(admission_read["registry_generation"], "OWNER_REGISTRY_GENERATION_INVALID"),
signature_verified=True,
epoch_currently_valid=False,
readbacks=(
AuthorityReadbackRef(
role="agent_task", authority="aios-agent-task-api",
readback_ref=task_read["readback_ref"], canonical_sha256=task_read["canonical_sha256"],
observed_at=task_read["observed_at"], revision=task["revision"],
),
AuthorityReadbackRef(
role="admission", authority="aios-task-admission-store",
readback_ref=admission_read["admission_ref"], canonical_sha256=admission_read["admission_sha256"],
observed_at=admission_read["observed_at"], revision=None,
),
AuthorityReadbackRef(
role="owner_registry", authority="aios-single-host-owner-authority",
readback_ref=admission_read["registry_ref"], canonical_sha256=admission_read["registry_sha256"],
observed_at=admission_read["observed_at"], revision=admission_read["registry_generation"],
),
AuthorityReadbackRef(
role="bridge_job", authority="aios-gitops-job-controller",
readback_ref=job_read["readback_ref"], canonical_sha256=job_read["canonical_sha256"],
observed_at=job_read["observed_at"], revision=job["revision"],
),
),
)
+34 -19
View File
@@ -135,6 +135,11 @@ SECRET_VAL_RE = re.compile(
r'(sk-[A-Za-z0-9_-]{20,}|HPUA[A-Z0-9]{10,}|SDC[A-Za-z0-9]{20,})'
)
# Control-plane file read: enforce auth, strict identifier, containment and a
# size cap on the command-detail read exposed by /admin/ai-evolution/api/data.
MAX_COMMAND_READ_BYTES = 1024 * 1024
_COMMAND_FILENAME_RE = re.compile(r"^[A-Za-z0-9][A-Za-z0-9._-]*\.md$")
READ_ONLY_GET_PATHS = frozenset({
'/', '/health', '/api/health', '/api/openapi.yaml',
'/api/openapi-chatgpt-actions.yaml', '/api/schema', '/api', '/api/status',
@@ -690,20 +695,6 @@ class H(BaseHTTPRequestHandler):
return self.sendj(body, exc.code)
except (urllib.error.URLError, TimeoutError) as exc:
return self.sendj({'error': 'GOVERNED_BRIDGE_UNAVAILABLE', 'detail': type(exc).__name__}, 503)
def do_POST(self):
path = urllib.parse.urlparse(self.path).path
if path in {
'/createTagApprovalRequest', '/requestOrCreateTag', '/getTagPublicationStatus',
'/buildProductionArtifact', '/executeApprovedProductionRelease',
'/getReleaseConsistency', '/runReleaseAcceptance',
}:
actor = self.need('tag:create')
if not actor:
return
return self.bridge_release(path, self.body())
if reject_non_read_http(self, 'POST', path):
return
return self.sendj({'error': 'not_found', 'path': path}, 404)
def do_GET(self):
u=urllib.parse.urlparse(self.path); path=u.path.rstrip('/') if u.path != '/' else u.path
if path not in READ_ONLY_GET_PATHS: return self.sendj({'error':'not_found','path':path},404)
@@ -949,6 +940,17 @@ class H(BaseHTTPRequestHandler):
path=urllib.parse.urlparse(self.path).path.rstrip('/')
if reject_non_read_http(self, 'POST', path): return
data=self.body()
# Governed V2 bridge release routes (previously shadowed by the
# redefinition of do_POST in this class).
if path in {
'/createTagApprovalRequest', '/requestOrCreateTag', '/getTagPublicationStatus',
'/buildProductionArtifact', '/executeApprovedProductionRelease',
'/getReleaseConsistency', '/runReleaseAcceptance',
}:
actor = self.need('tag:create')
if not actor:
return
return self.bridge_release(path, self.body())
try:
if path == '/createReleaseCandidate':
a=self.need('release:candidate')
@@ -1248,12 +1250,25 @@ return{vs,cmds,ar,cmd,cdet,tot,sok,sfail,runs,baseline,refreshMsg,loadCmd,selVer
commands.append(f.name)
command_detail=None
if cmd_filter:
actor = self.actor()
if not actor['ok']:
return self.sendj({'error':'unauthorized','hint':'Bearer token required'},401)
cmd = str(cmd_filter or '')
if not _COMMAND_FILENAME_RE.fullmatch(cmd) or '..' in cmd:
return self.sendj({'command':cmd_filter,'error':'unsafe_command'},400)
for d in cmd_dirs:
if d.exists():
fp=d/cmd_filter
if fp.exists():
command_detail={'command':cmd_filter,'content':fp.read_text(encoding='utf-8')}
break
if not d.exists():
continue
base = d.resolve()
fp = (base / cmd).resolve()
# Strict containment: must resolve directly inside one of the
# allowed command directories.
if fp.parent != base or not fp.is_file():
continue
if fp.stat().st_size > MAX_COMMAND_READ_BYTES:
return self.sendj({'command':cmd_filter,'error':'command_too_large'},413)
command_detail={'command':cmd_filter,'content':fp.read_text(encoding='utf-8')}
break
if not command_detail:
command_detail={'command':cmd_filter,'error':'not_found'}
return self.sendj(command_detail)
@@ -11,10 +11,12 @@ import base64
from dataclasses import dataclass
from datetime import datetime, timezone
import hashlib
import hmac
import importlib.util
import json
import os
import re
import sqlite3
from pathlib import Path
import stat
import time
@@ -29,6 +31,15 @@ from aios_task_admission_bootstrap import (
canonical_json_bytes,
sha256_json,
)
from aios_gitops_job_controller import (
GitOpsJobController as _FLGitOpsJobController,
JobState as _FLJobState,
sha256_json as _fl_job_sha256_json,
)
from aios_gitops_job_identity import (
legacy_task_storage_key as _fl_legacy_task_storage_key,
task_storage_key as _fl_task_storage_key,
)
OWNER_NAMESPACE = "aios-owner-approval-v1"
@@ -1069,6 +1080,8 @@ PLAN_FIELDS_V3=frozenset({
_SHA256=re.compile(r"^[0-9a-f]{64}$")
_OID=re.compile(r"^[0-9a-f]{40}$")
_JOB_ID=re.compile(r"^[A-Za-z0-9][A-Za-z0-9._:-]{0,199}$")
_TASK_ID=re.compile(r"^[A-Za-z0-9][A-Za-z0-9._:-]{2,179}$")
_TASK_BRANCH=re.compile(r"^agent/[A-Za-z0-9][A-Za-z0-9._/-]{2,179}$")
class ProducerReceiptError(RuntimeError):
@@ -1130,6 +1143,38 @@ def require_target_grant(grant:object)->None:
fail("TASK_ADMISSION_EXECUTION_DIGEST_INVALID")
def require_reservable_grant(grant:object)->None:
"""Validate the authenticated, persisted grant accepted by the producer.
The historical target keeps its exact tuple. Every other target is driven
by the grant persisted by TaskAdmissionBootstrap; no Mission metadata or
caller-supplied execution identity is consulted here.
"""
task_id=_gv(grant,"task_id")
if task_id==TARGET_TASK_ID:
require_target_grant(grant)
return
if not isinstance(task_id,str) or not _TASK_ID.fullmatch(task_id):
fail("TASK_ADMISSION_POST_OWNER_RESERVATION_TASK_INVALID")
if _gv(grant,"repository")!=TARGET_REPOSITORY:
fail("TASK_ADMISSION_POST_OWNER_RESERVATION_REPOSITORY_INVALID")
base=_gv(grant,"authority_base_oid")
if not isinstance(base,str) or not _OID.fullmatch(base):
fail("TASK_ADMISSION_POST_OWNER_RESERVATION_BASE_INVALID")
branch=_gv(grant,"target_branch")
if not isinstance(branch,str) or not _TASK_BRANCH.fullmatch(branch):
fail("TASK_ADMISSION_POST_OWNER_RESERVATION_BRANCH_INVALID")
generation=_gv(grant,"generation")
if isinstance(generation,bool) or not isinstance(generation,int) or generation<1:
fail("TASK_ADMISSION_POST_OWNER_RESERVATION_GENERATION_INVALID")
digest=_gv(grant,"execution_digest")
if not isinstance(digest,str) or not _SHA256.fullmatch(digest):
fail("TASK_ADMISSION_EXECUTION_DIGEST_INVALID")
grant_id=_gv(grant,"grant_id")
if not isinstance(grant_id,str) or not grant_id:
fail("TASK_ADMISSION_POST_OWNER_RESERVATION_GRANT_INVALID")
def _safe_path(value:object)->str:
if not isinstance(value,str) or not value or "\x00" in value:
fail("TASK_ADMISSION_EXECUTION_PLAN_FILE_PATH_INVALID")
@@ -1152,7 +1197,7 @@ def _unique_strings(value:object,code:str)->list[str]:
def validate_v3_plan(*,grant:object,raw:bytes)->dict[str,Any]:
require_target_grant(grant)
require_reservable_grant(grant)
plan=strict_json_object(raw)
if set(plan)!=PLAN_FIELDS_V3:
fail("TASK_ADMISSION_EXECUTION_PLAN_FIELDS_INVALID")
@@ -1195,14 +1240,17 @@ def validate_v3_plan(*,grant:object,raw:bytes)->dict[str,Any]:
def load_v3_plan(root:Path,grant:object)->dict[str,Any]:
require_target_grant(grant)
require_reservable_grant(grant)
try:
rst=root.lstat()
except OSError:
fail("TASK_ADMISSION_EXECUTION_PLAN_UNAVAILABLE")
if (
stat.S_ISLNK(rst.st_mode) or not stat.S_ISDIR(rst.st_mode)
or rst.st_uid!=0 or rst.st_mode&0o022
or (
_host_posix_metadata_enforced()
and (rst.st_uid!=0 or rst.st_mode&0o022)
)
):
fail("TASK_ADMISSION_EXECUTION_PLAN_DIRECTORY_INVALID")
path=root/(str(_gv(grant,"execution_digest"))+".json")
@@ -1212,7 +1260,10 @@ def load_v3_plan(root:Path,grant:object)->dict[str,Any]:
fail("TASK_ADMISSION_EXECUTION_PLAN_UNAVAILABLE")
if (
stat.S_ISLNK(st.st_mode) or not stat.S_ISREG(st.st_mode)
or st.st_uid!=0 or st.st_mode&0o022
or (
_host_posix_metadata_enforced()
and (st.st_uid!=0 or st.st_mode&0o022)
)
):
fail("TASK_ADMISSION_EXECUTION_PLAN_FILE_INVALID")
try:
@@ -1223,25 +1274,30 @@ def load_v3_plan(root:Path,grant:object)->dict[str,Any]:
def reservation_idempotency_key(grant:object)->str:
require_target_grant(grant)
require_reservable_grant(grant)
material=(
str(_gv(grant,"grant_id"))+"\0"+str(_gv(grant,"execution_digest"))
).encode("utf-8")
return "task-admission-execution-reservation-"+hashlib.sha256(material).hexdigest()[:32]
def build_reservation_request(grant:object)->dict[str,Any]:
require_target_grant(grant)
return {
"task_id":TARGET_TASK_ID,
"repository":TARGET_REPOSITORY,
"base_oid":TARGET_BASE_OID,
"target_branch":TARGET_BRANCH,
"generation":TARGET_GENERATION,
def build_reservation_request(
grant:object,*,expected_branch_oid:str|None=None,
)->dict[str,Any]:
require_reservable_grant(grant)
request={
"task_id":str(_gv(grant,"task_id")),
"repository":str(_gv(grant,"repository")),
"base_oid":str(_gv(grant,"authority_base_oid")),
"target_branch":str(_gv(grant,"target_branch")),
"generation":int(_gv(grant,"generation")),
"risk":str(_gv(grant,"risk")),
"description":"Reserve one inert durable GitOps Job identity after exact Owner TaskAdmission authorization; no Agent Task claim or execution.",
"idempotency_key":reservation_idempotency_key(grant),
}
if expected_branch_oid is not None:
request["expected_branch_oid"]=expected_branch_oid
return request
def _mapping(value:object,code:str)->Mapping[str,Any]:
@@ -1256,21 +1312,22 @@ def validate_bootstrap_response(
request:Mapping[str,Any],
response:object,
expected_side_effect_state:str,
expected_branch_mode:str="CONTINUE_EXISTING",
)->str:
require_target_grant(grant)
require_reservable_grant(grant)
result=_mapping(response,"TASK_ADMISSION_EXECUTION_RESERVATION_RESPONSE_INVALID")
job_id=result.get("job_id")
if not isinstance(job_id,str) or not _JOB_ID.fullmatch(job_id):
fail("TASK_ADMISSION_EXECUTION_RESERVATION_JOB_ID_INVALID")
expected={
"task_id":TARGET_TASK_ID,
"task_id":_gv(grant,"task_id"),
"job_id":job_id,
"repository":TARGET_REPOSITORY,
"base_oid":TARGET_BASE_OID,
"expected_branch_oid":TARGET_BASE_OID,
"target_branch":TARGET_BRANCH,
"branch_mode":"CONTINUE_EXISTING",
"generation":TARGET_GENERATION,
"repository":_gv(grant,"repository"),
"base_oid":_gv(grant,"authority_base_oid"),
"expected_branch_oid":request.get("expected_branch_oid",_gv(grant,"authority_base_oid")),
"target_branch":_gv(grant,"target_branch"),
"branch_mode":expected_branch_mode,
"generation":_gv(grant,"generation"),
"state":"BOUND",
}
for key,value in expected.items():
@@ -1333,13 +1390,20 @@ def validate_bootstrap_replay(
first:object,
replay:object,
replay_side_effect_state:str,
first_side_effect_state:str="QUEUED",
expected_branch_mode:str="CONTINUE_EXISTING",
)->str:
if first_side_effect_state not in {"QUEUED","SAME_JOB"}:
fail("TASK_ADMISSION_EXECUTION_RESERVATION_PHASE_INVALID")
first_id=validate_bootstrap_response(
grant=grant,request=request,response=first,expected_side_effect_state="QUEUED"
grant=grant,request=request,response=first,
expected_side_effect_state=first_side_effect_state,
expected_branch_mode=expected_branch_mode,
)
replay_id=validate_bootstrap_response(
grant=grant,request=request,response=replay,
expected_side_effect_state=replay_side_effect_state
expected_side_effect_state=replay_side_effect_state,
expected_branch_mode=expected_branch_mode,
)
if first_id!=replay_id:
fail("TASK_ADMISSION_EXECUTION_RESERVATION_REPLAY_JOB_CONFLICT")
@@ -1355,6 +1419,12 @@ import inspect as _fl_inspect
_FL_HARDENING_PROVIDER = ExecutionPlanDirectoryProvider
_FL_BUILD_FROM_ACTIVATION = build_task_admission_handlers_from_activation_file
_FL_BUILD_FROM_ENVIRONMENT = globals().get("build_task_admission_handlers_from_environment")
_FL_BUILD_RUNTIME_FROM_ACTIVATION = globals().get(
"build_task_admission_runtime_from_activation_file"
)
_FL_BUILD_RUNTIME_FROM_ENVIRONMENT = globals().get(
"build_task_admission_runtime_from_environment"
)
_FL_APPROVED_BOOTSTRAP_SHA256="758895a8407568890fd2f8eb2f232a3dbf2cfa369e5f65ae8c6975bf8d01e3b5"
REPLAY_SIDE_EFFECT_STATE="SAME_JOB"
@@ -1395,38 +1465,428 @@ def _fl_assert_bootstrap_source(bootstrap:object)->None:
if _fl_sha(path)!=_FL_APPROVED_BOOTSTRAP_SHA256:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_BOOTSTRAP_SOURCE_DRIFT",503)
def _fl_reservation_refs(runtime:object,grant:object)->tuple[str,str|None,str,str]:
require_reservable_grant(grant)
authority=getattr(runtime,"authority",None)
if authority is None or not callable(getattr(authority,"read_ref",None)):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_BACKEND_INVALID",503)
base=str(_gv(grant,"authority_base_oid"))
branch=str(_gv(grant,"target_branch"))
try:
main=authority.read_ref("refs/heads/main")
target=authority.read_ref("refs/heads/"+branch)
except Exception as exc:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_AUTHORITY_UNAVAILABLE",503) from exc
if main!=base:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_AUTHORITY_DRIFT",409)
if _gv(grant,"task_id")==TARGET_TASK_ID:
if target!=base:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_AUTHORITY_DRIFT",409)
return main,target,"CONTINUE_EXISTING",base
if target is None:
if _gv(grant,"generation")!=1:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_GENERATION_CONFLICT",409)
return main,target,"CREATE_NEW",base
if target==base:
return main,target,"CONTINUE_EXISTING",base
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_AUTHORITY_DRIFT",409)
def _fl_validate_durable_binding(
*,grant:object,request:Mapping[str,Any],durable:object,job_id:str,
expected_branch_mode:str,
)->None:
expected={
"task_id":_gv(grant,"task_id"),
"job_id":job_id,
"repository":_gv(grant,"repository"),
"base_oid":_gv(grant,"authority_base_oid"),
"expected_branch_oid":request.get("expected_branch_oid"),
"target_branch":_gv(grant,"target_branch"),
"branch_mode":expected_branch_mode,
"generation":_gv(grant,"generation"),
"risk":_gv(grant,"risk"),
"idempotency_key":request.get("idempotency_key"),
"description":request.get("description"),
}
if any(getattr(durable,field,None)!=value for field,value in expected.items()):
raise _fl_error("TASK_ADMISSION_PRODUCER_DURABLE_BINDING_CONFLICT",409)
def _fl_existing_reservation_readback(
*,bootstrap:object,grant:object,request:Mapping[str,Any],durable:object,
expected_branch_mode:str,
)->dict[str,Any]:
"""Verify an existing producer reservation without opening a mutating store.
``GovernedTaskBootstrap._store`` initializes missing directories and SQLite
state, while ``create`` may submit a new Job. Neither is valid after a
durable task binding already exists: missing producer state is corruption,
not crash recovery. Read the exact Job and idempotency rows through a
read-only SQLite URI and return an explicitly typed readback proof. This
is deliberately not a producer submit receipt: Job lifecycle state may
have advanced since the reservation was first created.
"""
task_id=str(_gv(grant,"task_id"))
job_id=getattr(durable,"job_id",None)
if not isinstance(job_id,str) or not _JOB_ID.fullmatch(job_id):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_JOB_ID_INVALID",409)
key=request.get("idempotency_key")
if not isinstance(key,str) or not key:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_IDEMPOTENCY_INVALID",409)
try:
expected_job_id=_FLGitOpsJobController.internal_job_id_for_idempotency_key(key)
producer_key=_fl_job_sha256_json({"client_idempotency_key":key})
except Exception as exc:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_IDEMPOTENCY_INVALID",409) from exc
if job_id!=expected_job_id or job_id!="job-"+producer_key[:32]:
raise _fl_error("TASK_ADMISSION_PRODUCER_DURABLE_BINDING_CONFLICT",409)
job_root=getattr(bootstrap,"job_root",None)
if not isinstance(job_root,(str,os.PathLike)):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_BACKEND_INVALID",503)
job_root=Path(job_root)
store_root=job_root/_fl_task_storage_key(task_id)
legacy_root=job_root/_fl_legacy_task_storage_key(task_id)
try:
legacy_stat=None
try:
legacy_stat=legacy_root.lstat()
except FileNotFoundError:
pass
if legacy_root!=store_root and legacy_stat is not None:
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STORE_CONFLICT",409)
store_stat=store_root.lstat()
db_path=store_root/"state.sqlite3"
db_stat=db_path.lstat()
except FileNotFoundError as exc:
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STORE_MISSING",409) from exc
except OSError as exc:
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STORE_READ_FAILED",503) from exc
if (
stat.S_ISLNK(store_stat.st_mode) or not stat.S_ISDIR(store_stat.st_mode)
or stat.S_ISLNK(db_stat.st_mode) or not stat.S_ISREG(db_stat.st_mode)
):
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STORE_INVALID",409)
values={
field:str(request[field]) for field in (
"task_id","repository","base_oid","target_branch","risk",
"description","idempotency_key",
)
}
payload=json.dumps(values,sort_keys=True,separators=(",",":")).encode()
payload_digest=hashlib.sha256(payload).hexdigest()
expected_branch_oid=str(
request.get("expected_branch_oid") or values["base_oid"]
)
producer_binding={
"campaign_id":task_id,
"repository":values["repository"],
"operation":"GOVERNED_TASK_CREATE",
"branch":values["target_branch"],
"expected_branch_oid":expected_branch_oid,
"payload_digest":payload_digest,
"idempotency_key":key,
"task_id":task_id,
"lineage_anchor_oid":values["base_oid"],
}
expected_request={
**producer_binding,
"binding":producer_binding,
"payload_ref":"sha256/"+payload_digest,
"payload_size":len(payload),
}
expected_job={
"job_id":job_id,
"idempotency_key":producer_key,
"campaign_id":task_id,
"repository":values["repository"],
"operation":"GOVERNED_TASK_CREATE",
"branch":values["target_branch"],
"expected_branch_oid":expected_branch_oid,
"payload_digest":payload_digest,
"payload_ref":"sha256/"+payload_digest,
"payload_size":len(payload),
}
# mode=ro prevents schema/data creation while still including committed WAL
# rows. query_only supplies a second SQLite-level write prohibition.
try:
resolved_store=store_root.resolve(strict=True)
resolved_db=db_path.resolve(strict=True)
if resolved_db.parent!=resolved_store:
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STORE_INVALID",409)
connection=sqlite3.connect(
resolved_db.as_uri()+"?mode=ro",
uri=True,timeout=5.0,
)
connection.row_factory=sqlite3.Row
try:
connection.execute("PRAGMA query_only=ON")
connection.execute("BEGIN")
job=connection.execute(
"SELECT job_id,idempotency_key,campaign_id,repository,operation,"
"branch,expected_branch_oid,payload_digest,payload_ref,payload_size,"
"state,side_effect_state,revision,request_json FROM jobs WHERE job_id=?",
(job_id,),
).fetchone()
idempotency=connection.execute(
"SELECT key,job_id,binding_json FROM idempotency_keys WHERE key=?",
(producer_key,),
).fetchone()
store_after=store_root.lstat()
db_after=db_path.lstat()
if (
(store_stat.st_dev,store_stat.st_ino)!=(store_after.st_dev,store_after.st_ino)
or (db_stat.st_dev,db_stat.st_ino)!=(db_after.st_dev,db_after.st_ino)
or stat.S_ISLNK(store_after.st_mode)
or not stat.S_ISDIR(store_after.st_mode)
or stat.S_ISLNK(db_after.st_mode)
or not stat.S_ISREG(db_after.st_mode)
):
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STORE_CHANGED",409)
finally:
connection.close()
del connection
except (OSError,sqlite3.Error) as exc:
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STORE_READ_FAILED",503) from exc
if job is None:
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_MISSING",409)
if idempotency is None:
raise _fl_error("TASK_ADMISSION_PRODUCER_IDEMPOTENCY_MISSING",409)
if any(job[field]!=value for field,value in expected_job.items()):
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_CONFLICT",409)
valid_states={item.value for item in _FLJobState}
if (
job["state"] not in valid_states
or not isinstance(job["side_effect_state"],str)
or not job["side_effect_state"]
or not isinstance(job["revision"],int)
or isinstance(job["revision"],bool)
or job["revision"]<0
):
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_STATE_INVALID",409)
try:
recorded_request=json.loads(job["request_json"])
recorded_binding=json.loads(idempotency["binding_json"])
except (TypeError,ValueError) as exc:
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_CONFLICT",409) from exc
if (
recorded_request!=expected_request
or idempotency["key"]!=producer_key
or idempotency["job_id"]!=job_id
or recorded_binding!=producer_binding
):
raise _fl_error("TASK_ADMISSION_PRODUCER_JOB_CONFLICT",409)
payload_path=store_root/"payloads"/payload_digest
descriptor=None
try:
descriptor=os.open(
payload_path,
os.O_RDONLY|getattr(os,"O_NOFOLLOW",0),
)
payload_stat=os.fstat(descriptor)
with os.fdopen(descriptor,"rb",closefd=False) as stream:
payload_bytes=stream.read(len(payload)+1)
payload_after=payload_path.lstat()
except FileNotFoundError as exc:
raise _fl_error("TASK_ADMISSION_PRODUCER_PAYLOAD_MISSING",409) from exc
except OSError as exc:
raise _fl_error("TASK_ADMISSION_PRODUCER_PAYLOAD_READ_FAILED",503) from exc
finally:
if descriptor is not None:
os.close(descriptor)
if (
not stat.S_ISREG(payload_stat.st_mode)
or stat.S_ISLNK(payload_after.st_mode)
or not stat.S_ISREG(payload_after.st_mode)
or (payload_stat.st_dev,payload_stat.st_ino)!=(payload_after.st_dev,payload_after.st_ino)
or len(payload_bytes)!=len(payload)
or not hmac.compare_digest(hashlib.sha256(payload_bytes).hexdigest(),payload_digest)
or payload_bytes!=payload
):
raise _fl_error("TASK_ADMISSION_PRODUCER_PAYLOAD_CONFLICT",409)
response=dict(durable.response(idempotent=True))
response["reservation_readback"]={
"schema":"aios.governed-task-reservation-readback.v1",
"job_id":job_id,
"state":job["state"],
"side_effect_state":job["side_effect_state"],
"revision":job["revision"],
"producer_idempotency_key":producer_key,
"request_idempotency_key":key,
}
return response
def _fl_validate_reservation_readback(
*,grant:object,request:Mapping[str,Any],durable:object,
response:object,expected_branch_mode:str,
)->str:
if not isinstance(response,Mapping):
raise _fl_error("TASK_ADMISSION_PRODUCER_READBACK_INVALID",409)
result=response
expected_response=dict(durable.response(idempotent=True))
if (
set(result)!=set(expected_response)|{"reservation_readback"}
or any(result.get(field)!=value for field,value in expected_response.items())
):
raise _fl_error("TASK_ADMISSION_PRODUCER_READBACK_CONFLICT",409)
proof=result.get("reservation_readback")
if not isinstance(proof,Mapping):
raise _fl_error("TASK_ADMISSION_PRODUCER_READBACK_INVALID",409)
expected_keys={
"schema","job_id","state","side_effect_state","revision",
"producer_idempotency_key","request_idempotency_key",
}
key=request.get("idempotency_key")
try:
producer_key=_fl_job_sha256_json({"client_idempotency_key":key})
except Exception as exc:
raise _fl_error("TASK_ADMISSION_PRODUCER_READBACK_INVALID",409) from exc
job_id=getattr(durable,"job_id",None)
if (
set(proof)!=expected_keys
or proof.get("schema")!="aios.governed-task-reservation-readback.v1"
or proof.get("job_id")!=job_id
or proof.get("producer_idempotency_key")!=producer_key
or proof.get("request_idempotency_key")!=key
or proof.get("state") not in {item.value for item in _FLJobState}
or not isinstance(proof.get("side_effect_state"),str)
or not proof.get("side_effect_state")
or not isinstance(proof.get("revision"),int)
or isinstance(proof.get("revision"),bool)
or proof.get("revision")<0
or getattr(durable,"branch_mode",None)!=expected_branch_mode
):
raise _fl_error("TASK_ADMISSION_PRODUCER_READBACK_CONFLICT",409)
return str(job_id)
class _FLProducerReceiptReserver:
def __init__(self,runtime:object):
self.runtime=runtime
self.readbacks:dict[str,dict[str,Any]]={}
def validate_v2_replay(self,grant:object,plan:Mapping[str,Any])->None:
"""Permit the old v2 contract only as an exact durable replay."""
require_reservable_grant(grant)
bootstrap=getattr(self.runtime,"bootstrap",None)
if bootstrap is None or not callable(getattr(bootstrap,"get_binding",None)):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_BACKEND_INVALID",503)
_fl_assert_bootstrap_source(bootstrap)
task_id=str(_gv(grant,"task_id"))
try:
durable=bootstrap.get_binding(task_id)
except Exception as exc:
raise _fl_error("TASK_ADMISSION_EXECUTION_PLAN_V2_REPLAY_BINDING_REQUIRED",409) from exc
expected={
"task_id":task_id,
"job_id":plan.get("external_job_id"),
"repository":_gv(grant,"repository"),
"base_oid":_gv(grant,"authority_base_oid"),
"expected_branch_oid":_gv(grant,"authority_base_oid"),
"target_branch":_gv(grant,"target_branch"),
"generation":_gv(grant,"generation"),
"risk":_gv(grant,"risk"),
"idempotency_key":"task-admission:"+task_id,
"description":"P0 governed execution projection for officially admitted Agent Task.",
}
if any(getattr(durable,field,None)!=value for field,value in expected.items()):
raise _fl_error("TASK_ADMISSION_EXECUTION_PLAN_V2_REPLAY_BINDING_CONFLICT",409)
if getattr(durable,"branch_mode",None) not in {"CREATE_NEW","CONTINUE_EXISTING"}:
raise _fl_error("TASK_ADMISSION_EXECUTION_PLAN_V2_REPLAY_BINDING_CONFLICT",409)
if not isinstance(plan.get("external_job_id"),str) or not _JOB_ID.fullmatch(plan["external_job_id"]):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_JOB_ID_INVALID",409)
def existing_v3_job(self,grant:object,plan:Mapping[str,Any])->str|None:
"""Reuse an exact durable reservation without reinterpreting live refs.
Once reserved, CREATE_NEW versus CONTINUE_EXISTING is immutable input
lineage. A later TaskPrincipal read must not turn legitimate branch
creation or forward progress into a fresh admission decision.
"""
try:
validate_v3_plan(grant=grant,raw=canonical_json_bytes(dict(plan)))
except ProducerReceiptError as exc:
raise _fl_error(exc.code,403) from exc
bootstrap=getattr(self.runtime,"bootstrap",None)
bindings=getattr(self.runtime,"bindings",None)
if (
bootstrap is None
or not callable(getattr(bootstrap,"get_binding",None))
or not isinstance(bindings,dict)
):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_BACKEND_INVALID",503)
_fl_assert_bootstrap_source(bootstrap)
task_id=str(_gv(grant,"task_id"))
try:
durable=bootstrap.get_binding(task_id)
except Exception as exc:
if str(exc)=="TASK_EXECUTION_PROJECTION_MISSING":
return None
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_READ_FAILED",503) from exc
request=build_reservation_request(
grant,expected_branch_oid=str(_gv(grant,"authority_base_oid")),
)
branch_mode=getattr(durable,"branch_mode",None)
if branch_mode not in {"CREATE_NEW","CONTINUE_EXISTING"}:
raise _fl_error("TASK_ADMISSION_PRODUCER_DURABLE_BINDING_CONFLICT",409)
job_id=getattr(durable,"job_id",None)
if not isinstance(job_id,str) or not _JOB_ID.fullmatch(job_id):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_JOB_ID_INVALID",409)
_fl_validate_durable_binding(
grant=grant,request=request,durable=durable,job_id=job_id,
expected_branch_mode=branch_mode,
)
readback=_fl_existing_reservation_readback(
bootstrap=bootstrap,grant=grant,request=request,durable=durable,
expected_branch_mode=branch_mode,
)
_fl_validate_reservation_readback(
grant=grant,request=request,durable=durable,response=readback,
expected_branch_mode=branch_mode,
)
# Keep readback evidence separate from producer submit receipts. In
# particular, a principal re-read must not overwrite the real receipt
# retained by a fresh admission in this runtime.
self.readbacks[task_id]=dict(readback)
return job_id
def __call__(self,grant:object,plan:Mapping[str,Any])->str:
try:
validate_v3_plan(grant=grant,raw=canonical_json_bytes(dict(plan)))
except ProducerReceiptError as exc:
raise _fl_error(exc.code,403) from exc
authority=getattr(self.runtime,"authority",None)
bootstrap=getattr(self.runtime,"bootstrap",None)
bindings=getattr(self.runtime,"bindings",None)
if (
authority is None or bootstrap is None
or not callable(getattr(bootstrap,"create",None))
or not callable(getattr(bootstrap,"get_binding",None))
or not isinstance(bindings,dict)
):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_BACKEND_INVALID",503)
_fl_assert_bootstrap_source(bootstrap)
try:
main_before=authority.read_ref("refs/heads/main")
branch_before=authority.read_ref("refs/heads/"+TARGET_BRANCH)
except Exception as exc:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_AUTHORITY_UNAVAILABLE",503) from exc
if main_before!=TARGET_BASE_OID or branch_before!=TARGET_BASE_OID:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_AUTHORITY_DRIFT",409)
main_before,branch_before,branch_mode,expected_branch_oid=_fl_reservation_refs(self.runtime,grant)
request=build_reservation_request(grant)
request=build_reservation_request(grant,expected_branch_oid=expected_branch_oid)
request_before=dict(request)
try:
first=bootstrap.create(request)
if request!=request_before:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_REQUEST_MUTATED",503)
validate_bootstrap_response(grant=grant,request=request,response=first,expected_side_effect_state="QUEUED")
first_phase=_mapping(first,"TASK_ADMISSION_EXECUTION_RESERVATION_RESPONSE_INVALID").get("reservation_receipt",{}).get("side_effect_state")
if first_phase not in {"QUEUED","SAME_JOB"}:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_PHASE_INVALID",409)
validate_bootstrap_response(
grant=grant,request=request,response=first,
expected_side_effect_state=first_phase,expected_branch_mode=branch_mode,
)
replay=bootstrap.create(request)
if request!=request_before:
@@ -1434,6 +1894,12 @@ class _FLProducerReceiptReserver:
job_id=validate_bootstrap_replay(
grant=grant,request=request,first=first,replay=replay,
replay_side_effect_state=REPLAY_SIDE_EFFECT_STATE,
first_side_effect_state=first_phase,expected_branch_mode=branch_mode,
)
durable=bootstrap.get_binding(str(_gv(grant,"task_id")))
_fl_validate_durable_binding(
grant=grant,request=request,durable=durable,job_id=job_id,
expected_branch_mode=branch_mode,
)
except ProducerReceiptError as exc:
raise _fl_error(exc.code,409) from exc
@@ -1448,11 +1914,12 @@ class _FLProducerReceiptReserver:
# Existing runtime's public createGovernedTask projection performs this
# in-memory binding update after bootstrap.create(). Mirror it only
# after the exact producer receipt + replay proof has passed.
bindings[TARGET_TASK_ID]=dict(replay)
bindings[str(_gv(grant,"task_id"))]=dict(replay)
self.readbacks.pop(str(_gv(grant,"task_id")),None)
try:
main_after=authority.read_ref("refs/heads/main")
branch_after=authority.read_ref("refs/heads/"+TARGET_BRANCH)
branch_after=authority.read_ref("refs/heads/"+str(_gv(grant,"target_branch")))
except Exception as exc:
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_AUTHORITY_UNAVAILABLE",503) from exc
if main_after!=main_before or branch_after!=branch_before:
@@ -1467,29 +1934,33 @@ class V3ProducerReceiptExecutionPlanDirectoryProvider(_FL_HARDENING_PROVIDER):
raise _fl_error("TASK_ADMISSION_EXECUTION_RESERVATION_BACKEND_INVALID",503)
try:
plan=load_v3_plan(_fl_provider_root(self),grant)
job_id=self._reserver(grant,plan)
job_id=self._reserver.existing_v3_job(grant,plan)
if job_id is None:
job_id=self._reserver(grant,plan)
except ProducerReceiptError as exc:
raise _fl_error(exc.code,403) from exc
return {**plan,"external_job_id":job_id}
return {
**plan,
"external_job_id":job_id,
"execution_digest":_gv(grant,"execution_digest"),
}
def __call__(self,grant:object)->Mapping[str,Any]:
if getattr(grant,"task_id",None)!=TARGET_TASK_ID:
try:
return self._target(grant)
except Exception as v3_error:
parent=getattr(super(),"__call__",None)
if callable(parent):
return parent(grant)
raise _fl_error("TASK_ADMISSION_EXECUTION_PLAN_PROVIDER_CONTRACT_INVALID",503)
return self._target(grant)
if not callable(parent) or self._reserver is None:
raise v3_error
try:
legacy=parent(grant)
except Exception:
raise v3_error
self._reserver.validate_v2_replay(grant,legacy)
return legacy
def load(self,grant:object)->Mapping[str,Any]:
if getattr(grant,"task_id",None)!=TARGET_TASK_ID:
parent=getattr(super(),"load",None)
if callable(parent):
return parent(grant)
parent_call=getattr(super(),"__call__",None)
if callable(parent_call):
return parent_call(grant)
raise _fl_error("TASK_ADMISSION_EXECUTION_PLAN_PROVIDER_CONTRACT_INVALID",503)
return self._target(grant)
return self(grant)
def _fl_runtime_from_post_admission(post_admission:object)->object:
runtime=getattr(post_admission,"__self__",None)
@@ -1509,31 +1980,93 @@ def _fl_call(builder,*args,post_admission=None,**kwargs):
previous=ExecutionPlanDirectoryProvider
ExecutionPlanDirectoryProvider=BoundProvider
try:
return builder(*args,post_admission=post_admission,**kwargs)
def exact_post_admission(grant,result):
task_id=str(_gv(grant,"task_id"))
durable=reserver.runtime.bootstrap.get_binding(task_id)
request=build_reservation_request(
grant,expected_branch_oid=str(_gv(grant,"authority_base_oid")),
)
if getattr(durable,"idempotency_key",None)!=request["idempotency_key"]:
return post_admission(grant,result)
branch_mode=getattr(durable,"branch_mode",None)
if branch_mode not in {"CREATE_NEW","CONTINUE_EXISTING"}:
raise _fl_error("TASK_ADMISSION_PRODUCER_DURABLE_BINDING_CONFLICT",409)
binding=reserver.runtime.bindings.get(task_id)
if isinstance(binding,Mapping) and "reservation_receipt" in binding:
# Fresh admission must retain the actual producer submit replay
# receipt produced by GovernedTaskBootstrap.create().
job_id=validate_bootstrap_response(
grant=grant,request=request,response=binding,
expected_side_effect_state="SAME_JOB",expected_branch_mode=branch_mode,
)
else:
# A process restart may leave only durable producer state. Its
# separately typed readback proves immutable identity without
# claiming that a producer submit happened in this request.
binding=reserver.readbacks.get(task_id)
job_id=_fl_validate_reservation_readback(
grant=grant,request=request,durable=durable,response=binding,
expected_branch_mode=branch_mode,
)
if result.get("task_id")!=task_id or result.get("external_job_id")!=job_id or result.get("state",result.get("status")) not in {"QUEUED","CLAIMED","RUNNING"}:
raise _fl_error("TASK_ADMISSION_PRODUCER_TASK_BINDING_CONFLICT",409)
_fl_validate_durable_binding(
grant=grant,request=request,durable=durable,job_id=job_id,
expected_branch_mode=branch_mode,
)
return dict(binding)
return builder(*args,post_admission=exact_post_admission,**kwargs)
finally:
ExecutionPlanDirectoryProvider=previous
def build_task_admission_runtime_from_activation_file(
activation_file:Path,*,
caller_service:str="aios-chatgpt-bridge-v2-staging.service",
post_admission=None,
):
if not callable(_FL_BUILD_RUNTIME_FROM_ACTIVATION):
raise RuntimeError("TASK_ADMISSION_RUNTIME_ACTIVATION_BUILDER_MISSING")
return _fl_call(
_FL_BUILD_RUNTIME_FROM_ACTIVATION,activation_file,
caller_service=caller_service,post_admission=post_admission,
)
def build_task_admission_handlers_from_activation_file(
activation_file:Path,*,
caller_service:str="aios-chatgpt-bridge-v2-staging.service",
post_admission=None,
):
return build_task_admission_runtime_from_activation_file(
activation_file,
caller_service=caller_service,post_admission=post_admission,
).handlers
def build_task_admission_runtime_from_environment(
*,
caller_service:str="aios-chatgpt-bridge-v2-staging.service",
post_admission=None,
):
if not callable(_FL_BUILD_RUNTIME_FROM_ENVIRONMENT):
raise RuntimeError("TASK_ADMISSION_RUNTIME_ENVIRONMENT_BUILDER_MISSING")
return _fl_call(
_FL_BUILD_FROM_ACTIVATION,activation_file,
_FL_BUILD_RUNTIME_FROM_ENVIRONMENT,
caller_service=caller_service,post_admission=post_admission,
)
def build_task_admission_handlers_from_environment(
*,
caller_service:str="aios-chatgpt-bridge-v2-staging.service",
post_admission=None,
):
if not callable(_FL_BUILD_FROM_ENVIRONMENT):
raise RuntimeError("TASK_ADMISSION_ENVIRONMENT_BUILDER_MISSING")
return _fl_call(
_FL_BUILD_FROM_ENVIRONMENT,
return build_task_admission_runtime_from_environment(
caller_service=caller_service,post_admission=post_admission,
)
).handlers
ExecutionPlanDirectoryProvider=V3ProducerReceiptExecutionPlanDirectoryProvider
# Keep the standalone legacy provider contract. The exported admission builders
# install a runtime-bound v3 subclass only while composing their trusted port;
# an unbound provider cannot prove a durable producer receipt.
ExecutionPlanDirectoryProvider=_FL_HARDENING_PROVIDER
# ===== AIOS FORMAL LINEAGE POST-OWNER PRODUCER NATIVE V12 — END =====
+779
View File
@@ -0,0 +1,779 @@
from __future__ import annotations
from dataclasses import dataclass, field, replace
from datetime import datetime, timezone
from enum import Enum
from hashlib import sha256
import re
from typing import Callable, Iterable, Mapping, Protocol, Sequence
class LineageError(ValueError):
pass
class RelationType(str, Enum):
ROOT = "ROOT"
CHILD = "CHILD"
SUCCESSOR = "SUCCESSOR"
RECOVERY = "RECOVERY"
class NextLegalAction(str, Enum):
ROOT_COMPLETE = "ROOT_COMPLETE"
CREATE_OR_RESUME_NEXT_CHILD = "CREATE_OR_RESUME_NEXT_CHILD"
WAIT_DEPENDENCY = "WAIT_DEPENDENCY"
RECOVER_EXISTING_CHILD = "RECOVER_EXISTING_CHILD"
OWNER_BOUNDARY = "OWNER_BOUNDARY"
TERMINAL_STATUSES = frozenset(
{"COMPLETE", "COMPLETED", "SUCCEEDED", "FAILED", "FAILED_FINAL", "CANCELLED"}
)
RECOVERY_STATUSES = frozenset(
{"RECONCILIATION_REQUIRED", "RECOVERY_REQUIRED", "FAILED_RECOVERABLE"}
)
BLOCKED_STATUSES = frozenset({"BLOCKED", "WAITING_DEPENDENCY", "WAITING_CAPACITY"})
_ALIAS_RE = re.compile(r"^[A-Z0-9][A-Z0-9_-]{1,31}$")
_SLUG_RE = re.compile(r"^[A-Z0-9][A-Z0-9-]{1,63}$")
def _require_utc(ts: datetime) -> datetime:
if ts.tzinfo is None or ts.utcoffset() is None:
raise LineageError("created_at/reviewed_at must be timezone-aware")
return ts.astimezone(timezone.utc)
def derive_safe_owner_alias(principal_id: str, preferred_alias: str | None = None) -> str:
"""Return a non-secret display alias.
A trusted identity adapter may supply a safe human alias. Otherwise we use a
deterministic pseudonym derived from the principal identifier. We never put
raw email addresses, tokens, or principal strings into task names.
"""
if preferred_alias and "@" not in preferred_alias and ":" not in preferred_alias:
candidate = re.sub(r"[^A-Za-z0-9_-]+", "-", preferred_alias.strip()).strip("-_").upper()
if 2 <= len(candidate) <= 32 and _ALIAS_RE.fullmatch(candidate):
return candidate
if not principal_id:
raise LineageError("principal_id is required")
return "U" + sha256(principal_id.encode("utf-8")).hexdigest()[:9].upper()
def normalize_goal_slug(value: str) -> str:
value = re.sub(r"[^A-Za-z0-9]+", "-", value.strip()).strip("-").upper()
value = re.sub(r"-+", "-", value)
if len(value) > 64:
value = value[:64].rstrip("-")
if not _SLUG_RE.fullmatch(value):
raise LineageError("goal_slug must normalize to 2-64 [A-Z0-9-] characters")
return value
def short_ref(task_id: str) -> str:
if not task_id:
raise LineageError("task_id is required")
compact = re.sub(r"[^A-Za-z0-9]", "", task_id).upper()
if len(compact) >= 8:
return compact[-8:]
return sha256(task_id.encode("utf-8")).hexdigest()[:8].upper()
@dataclass(frozen=True)
class TaskLineage:
task_id: str
display_name: str
owner_principal_id: str
owner_alias: str
created_at: datetime
mission_id: str
root_task_id: str
parent_task_id: str | None
lineage_depth: int
relation_type: RelationType
sibling_sequence: int
goal_slug: str
root_acceptance_ref: str
parent_revision: int | None = None
@dataclass(frozen=True)
class TaskRecord:
task_id: str
revision: int
status: str
lineage: TaskLineage
evidence_refs: tuple[str, ...] = ()
terminal_evidence_proven: bool = False
@dataclass(frozen=True)
class AcceptanceCondition:
key: str
satisfied: bool
evidence_refs: tuple[str, ...] = ()
owner_boundary: bool = False
@dataclass(frozen=True)
class RootAcceptanceSnapshot:
conditions: tuple[AcceptanceCondition, ...]
@property
def remaining(self) -> tuple[str, ...]:
return tuple(c.key for c in self.conditions if not c.satisfied)
@property
def complete(self) -> bool:
return all(c.satisfied for c in self.conditions)
@property
def owner_boundary_required(self) -> bool:
return any((not c.satisfied) and c.owner_boundary for c in self.conditions)
@property
def evidence_refs(self) -> tuple[str, ...]:
values: list[str] = []
for condition in self.conditions:
if condition.satisfied:
values.extend(condition.evidence_refs)
return tuple(dict.fromkeys(values))
def canonicalize_acceptance(
acceptance: RootAcceptanceSnapshot,
) -> RootAcceptanceSnapshot:
if not isinstance(acceptance, RootAcceptanceSnapshot):
raise LineageError("acceptance evaluator returned an invalid snapshot")
normalized: list[AcceptanceCondition] = []
keys: set[str] = set()
for condition in acceptance.conditions:
if not isinstance(condition, AcceptanceCondition):
raise LineageError("acceptance snapshot contains an invalid condition")
if not isinstance(condition.key, str) or not condition.key:
raise LineageError("acceptance condition key must be a non-empty string")
if condition.key in keys:
raise LineageError("acceptance condition keys must be unique")
if not isinstance(condition.satisfied, bool) or not isinstance(
condition.owner_boundary, bool
):
raise LineageError("acceptance condition flags must be booleans")
if not all(isinstance(ref, str) and ref for ref in condition.evidence_refs):
raise LineageError("acceptance evidence refs must be non-empty strings")
keys.add(condition.key)
normalized.append(
replace(
condition,
evidence_refs=tuple(sorted(set(condition.evidence_refs))),
)
)
return RootAcceptanceSnapshot(tuple(sorted(normalized, key=lambda c: c.key)))
@dataclass(frozen=True)
class RootReviewRecord:
review_id: str
root_task_id: str
reviewed_at: datetime
trigger_child_task_id: str
root_revision_before: int
root_revision_after: int
completed_children: tuple[str, ...]
active_children: tuple[str, ...]
blocked_children: tuple[str, ...]
remaining_acceptance_conditions: tuple[str, ...]
next_legal_action: NextLegalAction
root_complete: bool
evidence_refs: tuple[str, ...]
class RootReviewStore(Protocol):
def get_task(self, task_id: str) -> TaskRecord: ...
def list_descendants(self, root_task_id: str) -> Sequence[TaskRecord]: ...
def get_root_review(self, review_id: str) -> RootReviewRecord | None: ...
def compare_and_swap_root_review(
self,
root_task_id: str,
expected_root_revision: int,
review: RootReviewRecord,
) -> bool: ...
AcceptanceEvaluator = Callable[[TaskRecord, Sequence[TaskRecord]], RootAcceptanceSnapshot]
def _timestamp(ts: datetime) -> str:
return _require_utc(ts).strftime("%Y%m%dT%H%M%SZ")
def _relation_token(relation: RelationType, sequence: int) -> str:
if sequence < 1 or sequence > 99:
raise LineageError("non-root sibling_sequence must be 1..99")
prefix = {
RelationType.CHILD: "C",
RelationType.SUCCESSOR: "S",
RelationType.RECOVERY: "X",
}.get(relation)
if not prefix:
raise LineageError("ROOT does not use a child relation token")
return f"{prefix}{sequence:02d}"
def build_display_name(lineage: TaskLineage) -> str:
alias = lineage.owner_alias.upper()
if not _ALIAS_RE.fullmatch(alias):
raise LineageError("owner_alias must be a non-secret [A-Z0-9_-] alias")
goal = normalize_goal_slug(lineage.goal_slug)
ts = _timestamp(lineage.created_at)
if lineage.relation_type == RelationType.ROOT:
return f"TASK-{alias}-{ts}-ROOT-{goal}-{short_ref(lineage.task_id)[:4]}"
relation = _relation_token(lineage.relation_type, lineage.sibling_sequence)
return (
f"TASK-{alias}-{ts}"
f"-R{short_ref(lineage.root_task_id)}"
f"-P{short_ref(lineage.parent_task_id or '')}"
f"-D{lineage.lineage_depth:02d}"
f"-{relation}-{goal}"
)
def make_root_lineage(
*,
task_id: str,
owner_principal_id: str,
owner_alias: str,
created_at: datetime,
mission_id: str,
goal_slug: str,
root_acceptance_ref: str,
) -> TaskLineage:
lineage = TaskLineage(
task_id=task_id,
display_name="",
owner_principal_id=owner_principal_id,
owner_alias=owner_alias.upper(),
created_at=_require_utc(created_at),
mission_id=mission_id,
root_task_id=task_id,
parent_task_id=None,
lineage_depth=0,
relation_type=RelationType.ROOT,
sibling_sequence=0,
goal_slug=normalize_goal_slug(goal_slug),
root_acceptance_ref=root_acceptance_ref,
parent_revision=None,
)
lineage = replace(lineage, display_name=build_display_name(lineage))
validate_lineage(lineage)
return lineage
def make_child_lineage(
*,
task_id: str,
created_at: datetime,
parent: TaskRecord,
root: TaskRecord,
relation_type: RelationType,
sibling_sequence: int,
goal_slug: str,
) -> TaskLineage:
if relation_type == RelationType.ROOT:
raise LineageError("child relation_type cannot be ROOT")
lineage = TaskLineage(
task_id=task_id,
display_name="",
owner_principal_id=root.lineage.owner_principal_id,
owner_alias=root.lineage.owner_alias,
created_at=_require_utc(created_at),
mission_id=root.lineage.mission_id,
root_task_id=root.task_id,
parent_task_id=parent.task_id,
lineage_depth=parent.lineage.lineage_depth + 1,
relation_type=relation_type,
sibling_sequence=sibling_sequence,
goal_slug=normalize_goal_slug(goal_slug),
root_acceptance_ref=root.lineage.root_acceptance_ref,
parent_revision=parent.revision,
)
lineage = replace(lineage, display_name=build_display_name(lineage))
validate_lineage(lineage, parent=parent, root=root)
return lineage
def validate_lineage(
lineage: TaskLineage,
*,
parent: TaskRecord | None = None,
root: TaskRecord | None = None,
) -> None:
_validate_lineage_shape(lineage)
if lineage.relation_type == RelationType.ROOT:
return
if parent is None or root is None:
raise LineageError("non-root validation requires parent and root authority records")
if root.lineage.relation_type != RelationType.ROOT:
raise LineageError("root authority record must have ROOT lineage")
if parent.lineage.root_task_id != root.task_id:
raise LineageError("parent is not attached to the selected root")
if parent.lineage.owner_principal_id != root.lineage.owner_principal_id:
raise LineageError("parent owner does not match root owner")
if parent.lineage.mission_id != root.lineage.mission_id:
raise LineageError("parent mission does not match root mission")
if parent.lineage.root_acceptance_ref != root.lineage.root_acceptance_ref:
raise LineageError("parent acceptance reference does not match root")
if lineage.root_task_id != root.task_id:
raise LineageError("root_task_id mismatch")
if lineage.parent_task_id != parent.task_id:
raise LineageError("parent_task_id mismatch")
if lineage.lineage_depth != parent.lineage.lineage_depth + 1:
raise LineageError("lineage_depth must equal parent depth + 1")
if lineage.owner_principal_id != root.lineage.owner_principal_id:
raise LineageError("cross-owner lineage attachment rejected")
if lineage.owner_alias != root.lineage.owner_alias:
raise LineageError("owner_alias must match root")
if lineage.mission_id != root.lineage.mission_id:
raise LineageError("mission_id must match root")
if lineage.root_acceptance_ref != root.lineage.root_acceptance_ref:
raise LineageError("root_acceptance_ref must match root")
if lineage.parent_revision != parent.revision:
raise LineageError("stale or mismatched parent revision")
def _validate_lineage_shape(lineage: TaskLineage) -> None:
if not lineage.task_id or not lineage.owner_principal_id or not lineage.mission_id:
raise LineageError("task_id, owner_principal_id and mission_id are required")
if not lineage.root_acceptance_ref:
raise LineageError("root_acceptance_ref is required")
if not _ALIAS_RE.fullmatch(lineage.owner_alias):
raise LineageError("invalid owner_alias")
if normalize_goal_slug(lineage.goal_slug) != lineage.goal_slug:
raise LineageError("goal_slug must already be normalized")
if not lineage.display_name or len(lineage.display_name) > 200:
raise LineageError("display_name must be non-empty and <= 200 characters")
if lineage.relation_type == RelationType.ROOT:
if lineage.root_task_id != lineage.task_id:
raise LineageError("root_task_id must equal task_id for ROOT")
if lineage.parent_task_id is not None:
raise LineageError("ROOT parent_task_id must be null")
if lineage.lineage_depth != 0:
raise LineageError("ROOT lineage_depth must be 0")
if lineage.sibling_sequence != 0:
raise LineageError("ROOT sibling_sequence must be 0")
if lineage.parent_revision is not None:
raise LineageError("ROOT parent_revision must be null")
return
if not lineage.root_task_id or not lineage.parent_task_id:
raise LineageError("non-root root_task_id and parent_task_id are required")
if lineage.lineage_depth < 1:
raise LineageError("non-root lineage_depth must be >= 1")
if lineage.parent_revision is None or lineage.parent_revision < 0:
raise LineageError("non-root parent_revision must be >= 0")
_relation_token(lineage.relation_type, lineage.sibling_sequence)
def attach_legacy_lineage(
*,
existing_task_id: str,
created_at: datetime,
parent: TaskRecord,
root: TaskRecord,
relation_type: RelationType,
sibling_sequence: int,
goal_slug: str,
) -> TaskLineage:
"""Return lineage metadata for an existing durable task without changing its ID."""
return make_child_lineage(
task_id=existing_task_id,
created_at=created_at,
parent=parent,
root=root,
relation_type=relation_type,
sibling_sequence=sibling_sequence,
goal_slug=goal_slug,
)
def rename_display_only(lineage: TaskLineage, new_display_name: str) -> TaskLineage:
if not new_display_name or len(new_display_name) > 200:
raise LineageError("display name must be non-empty and <= 200 characters")
return replace(lineage, display_name=new_display_name)
LINEAGE_SCHEMA = "aios.task-lineage.v1"
ROOT_REVIEW_SCHEMA = "aios.root-review.v1"
_LINEAGE_FIELDS = frozenset(
{
"schema",
"task_id",
"display_name",
"owner_principal_id",
"owner_alias",
"created_at",
"mission_id",
"root_task_id",
"parent_task_id",
"lineage_depth",
"relation_type",
"sibling_sequence",
"goal_slug",
"root_acceptance_ref",
"parent_revision",
}
)
_ROOT_REVIEW_FIELDS = frozenset(
{
"schema",
"review_id",
"root_task_id",
"reviewed_at",
"trigger_child_task_id",
"root_revision_before",
"root_revision_after",
"completed_children",
"active_children",
"blocked_children",
"remaining_acceptance_conditions",
"next_legal_action",
"root_complete",
"evidence_refs",
}
)
def _require_exact_fields(metadata: Mapping[str, object], expected: frozenset[str]) -> None:
if not all(isinstance(key, str) for key in metadata):
raise LineageError("metadata field names must be strings")
actual = frozenset(metadata)
if actual != expected:
missing = sorted(expected - actual)
extra = sorted(actual - expected)
raise LineageError(f"metadata fields mismatch: missing={missing} extra={extra}")
def _require_string(metadata: Mapping[str, object], key: str) -> str:
value = metadata[key]
if not isinstance(value, str) or not value:
raise LineageError(f"{key} must be a non-empty string")
return value
def _require_int(metadata: Mapping[str, object], key: str) -> int:
value = metadata[key]
if isinstance(value, bool) or not isinstance(value, int):
raise LineageError(f"{key} must be an integer")
return value
def _require_string_list(metadata: Mapping[str, object], key: str) -> tuple[str, ...]:
value = metadata[key]
if not isinstance(value, list) or not all(isinstance(item, str) and item for item in value):
raise LineageError(f"{key} must be a list of non-empty strings")
items = tuple(value)
if len(items) != len(set(items)):
raise LineageError(f"{key} must not contain duplicates")
return items
def lineage_to_metadata(lineage: TaskLineage) -> dict[str, object]:
return {
"schema": LINEAGE_SCHEMA,
"task_id": lineage.task_id,
"display_name": lineage.display_name,
"owner_principal_id": lineage.owner_principal_id,
"owner_alias": lineage.owner_alias,
"created_at": _require_utc(lineage.created_at).isoformat().replace("+00:00", "Z"),
"mission_id": lineage.mission_id,
"root_task_id": lineage.root_task_id,
"parent_task_id": lineage.parent_task_id,
"lineage_depth": lineage.lineage_depth,
"relation_type": lineage.relation_type.value,
"sibling_sequence": lineage.sibling_sequence,
"goal_slug": lineage.goal_slug,
"root_acceptance_ref": lineage.root_acceptance_ref,
"parent_revision": lineage.parent_revision,
}
def lineage_from_metadata(metadata: Mapping[str, object]) -> TaskLineage:
_require_exact_fields(metadata, _LINEAGE_FIELDS)
if metadata.get("schema") != LINEAGE_SCHEMA:
raise LineageError("unsupported lineage metadata schema")
try:
created_raw = _require_string(metadata, "created_at")
if created_raw.endswith("Z"):
created_raw = created_raw[:-1] + "+00:00"
parent_task_id = metadata["parent_task_id"]
if parent_task_id is not None and (
not isinstance(parent_task_id, str) or not parent_task_id
):
raise LineageError("parent_task_id must be null or a non-empty string")
parent_revision = metadata["parent_revision"]
if parent_revision is not None and (
isinstance(parent_revision, bool) or not isinstance(parent_revision, int)
):
raise LineageError("parent_revision must be null or an integer")
lineage = TaskLineage(
task_id=_require_string(metadata, "task_id"),
display_name=_require_string(metadata, "display_name"),
owner_principal_id=_require_string(metadata, "owner_principal_id"),
owner_alias=_require_string(metadata, "owner_alias"),
created_at=datetime.fromisoformat(created_raw),
mission_id=_require_string(metadata, "mission_id"),
root_task_id=_require_string(metadata, "root_task_id"),
parent_task_id=parent_task_id,
lineage_depth=_require_int(metadata, "lineage_depth"),
relation_type=RelationType(_require_string(metadata, "relation_type")),
sibling_sequence=_require_int(metadata, "sibling_sequence"),
goal_slug=_require_string(metadata, "goal_slug"),
root_acceptance_ref=_require_string(metadata, "root_acceptance_ref"),
parent_revision=parent_revision,
)
except (KeyError, TypeError, ValueError, LineageError) as exc:
raise LineageError(f"invalid lineage metadata: {exc}") from exc
_validate_lineage_shape(lineage)
return lineage
def root_review_to_record(review: RootReviewRecord) -> dict[str, object]:
return {
"schema": ROOT_REVIEW_SCHEMA,
"review_id": review.review_id,
"root_task_id": review.root_task_id,
"reviewed_at": _require_utc(review.reviewed_at).isoformat().replace("+00:00", "Z"),
"trigger_child_task_id": review.trigger_child_task_id,
"root_revision_before": review.root_revision_before,
"root_revision_after": review.root_revision_after,
"completed_children": list(review.completed_children),
"active_children": list(review.active_children),
"blocked_children": list(review.blocked_children),
"remaining_acceptance_conditions": list(review.remaining_acceptance_conditions),
"next_legal_action": review.next_legal_action.value,
"root_complete": review.root_complete,
"evidence_refs": list(review.evidence_refs),
}
def root_review_from_record(
record: Mapping[str, object], *, expected_review_id: str | None = None
) -> RootReviewRecord:
_require_exact_fields(record, _ROOT_REVIEW_FIELDS)
if record.get("schema") != ROOT_REVIEW_SCHEMA:
raise LineageError("unsupported root-review metadata schema")
try:
reviewed_raw = _require_string(record, "reviewed_at")
if reviewed_raw.endswith("Z"):
reviewed_raw = reviewed_raw[:-1] + "+00:00"
root_complete = record["root_complete"]
if not isinstance(root_complete, bool):
raise LineageError("root_complete must be a boolean")
review = RootReviewRecord(
review_id=_require_string(record, "review_id"),
root_task_id=_require_string(record, "root_task_id"),
reviewed_at=_require_utc(datetime.fromisoformat(reviewed_raw)),
trigger_child_task_id=_require_string(record, "trigger_child_task_id"),
root_revision_before=_require_int(record, "root_revision_before"),
root_revision_after=_require_int(record, "root_revision_after"),
completed_children=_require_string_list(record, "completed_children"),
active_children=_require_string_list(record, "active_children"),
blocked_children=_require_string_list(record, "blocked_children"),
remaining_acceptance_conditions=_require_string_list(
record, "remaining_acceptance_conditions"
),
next_legal_action=NextLegalAction(
_require_string(record, "next_legal_action")
),
root_complete=root_complete,
evidence_refs=_require_string_list(record, "evidence_refs"),
)
except (KeyError, TypeError, ValueError, LineageError) as exc:
raise LineageError(f"invalid root-review metadata: {exc}") from exc
if expected_review_id is not None and review.review_id != expected_review_id:
raise LineageError("root-review id does not match lookup key")
if review.root_revision_before < 0 or (
review.root_revision_after != review.root_revision_before + 1
):
raise LineageError("root-review revision transition is invalid")
if review.root_complete != (
review.next_legal_action == NextLegalAction.ROOT_COMPLETE
):
raise LineageError("root_complete does not match next_legal_action")
child_sets = (
set(review.completed_children),
set(review.active_children),
set(review.blocked_children),
)
if any(child_sets[i] & child_sets[j] for i in range(3) for j in range(i + 1, 3)):
raise LineageError("root-review child classifications overlap")
return review
def _review_id(
root_task_id: str,
trigger_child_task_id: str,
trigger_revision: int,
acceptance: RootAcceptanceSnapshot,
) -> str:
acceptance_digest = sha256(
"\n".join(
f"{c.key}:{int(c.satisfied)}:{int(c.owner_boundary)}:"
f"{','.join(sorted(set(c.evidence_refs)))}"
for c in acceptance.conditions
).encode("utf-8")
).hexdigest()
payload = f"{root_task_id}|{trigger_child_task_id}|{trigger_revision}|{acceptance_digest}"
return "RR-" + sha256(payload.encode("utf-8")).hexdigest()[:24]
def _next_action(
descendants: Sequence[TaskRecord],
acceptance: RootAcceptanceSnapshot,
) -> NextLegalAction:
if acceptance.complete:
return NextLegalAction.ROOT_COMPLETE
if acceptance.owner_boundary_required:
return NextLegalAction.OWNER_BOUNDARY
if any(d.status in RECOVERY_STATUSES for d in descendants):
return NextLegalAction.RECOVER_EXISTING_CHILD
if any(d.status in BLOCKED_STATUSES for d in descendants):
return NextLegalAction.WAIT_DEPENDENCY
return NextLegalAction.CREATE_OR_RESUME_NEXT_CHILD
def review_root_after_child_terminal(
*,
store: RootReviewStore,
trigger_child_task_id: str,
acceptance_evaluator: AcceptanceEvaluator,
reviewed_at: datetime,
max_cas_retries: int = 4,
) -> RootReviewRecord:
"""Persist a root review after a proven child terminal transition.
This function is deliberately storage-agnostic. TaskPlane remains the
authority and supplies the CAS implementation. The function never creates a
new root or mutates a child identity.
"""
if max_cas_retries < 1:
raise LineageError("max_cas_retries must be >= 1")
reviewed_at = _require_utc(reviewed_at)
for _ in range(max_cas_retries):
child = store.get_task(trigger_child_task_id)
if child.lineage.relation_type == RelationType.ROOT:
raise LineageError("root review hook requires a child task")
if child.status not in TERMINAL_STATUSES:
raise LineageError("child must be terminal before root review")
if not child.terminal_evidence_proven:
raise LineageError("child terminal evidence is not proven")
root = store.get_task(child.lineage.root_task_id)
if root.lineage.relation_type != RelationType.ROOT:
raise LineageError("root_task_id does not resolve to a ROOT task")
if child.lineage.owner_principal_id != root.lineage.owner_principal_id:
raise LineageError("trigger child owner does not match root owner")
if child.lineage.mission_id != root.lineage.mission_id:
raise LineageError("trigger child mission does not match root mission")
if child.lineage.root_acceptance_ref != root.lineage.root_acceptance_ref:
raise LineageError("trigger child acceptance reference does not match root")
descendants = tuple(
sorted(store.list_descendants(root.task_id), key=lambda item: item.task_id)
)
descendant_ids = tuple(item.task_id for item in descendants)
if len(descendant_ids) != len(set(descendant_ids)):
raise LineageError("descendant authority returned duplicate task ids")
if child.task_id not in descendant_ids:
raise LineageError("trigger child missing from authoritative descendants")
for descendant in descendants:
if descendant.lineage.root_task_id != root.task_id:
raise LineageError("descendant is attached to a different root")
if descendant.lineage.owner_principal_id != root.lineage.owner_principal_id:
raise LineageError("descendant owner does not match root owner")
if descendant.lineage.mission_id != root.lineage.mission_id:
raise LineageError("descendant mission does not match root mission")
if (
descendant.lineage.root_acceptance_ref
!= root.lineage.root_acceptance_ref
):
raise LineageError("descendant acceptance reference does not match root")
acceptance = canonicalize_acceptance(acceptance_evaluator(root, descendants))
review_id = _review_id(root.task_id, child.task_id, child.revision, acceptance)
completed = tuple(
d.task_id for d in descendants
if d.status in TERMINAL_STATUSES and d.terminal_evidence_proven
)
active = tuple(
d.task_id for d in descendants
if d.status not in TERMINAL_STATUSES and d.status not in BLOCKED_STATUSES
)
blocked = tuple(
d.task_id for d in descendants
if d.status in BLOCKED_STATUSES
)
evidence = tuple(sorted(set(child.evidence_refs + acceptance.evidence_refs)))
action = _next_action(descendants, acceptance)
existing = store.get_root_review(review_id)
if existing is not None:
if (
existing.root_task_id != root.task_id
or existing.trigger_child_task_id != child.task_id
):
raise LineageError("persisted root review identity conflicts with trigger")
if existing.root_revision_after > root.revision:
raise LineageError("persisted root review revision is ahead of authority")
expected_semantics = (
completed,
active,
blocked,
acceptance.remaining,
action,
action == NextLegalAction.ROOT_COMPLETE,
evidence,
)
actual_semantics = (
existing.completed_children,
existing.active_children,
existing.blocked_children,
existing.remaining_acceptance_conditions,
existing.next_legal_action,
existing.root_complete,
existing.evidence_refs,
)
if actual_semantics != expected_semantics:
raise LineageError("persisted root review conflicts with current semantics")
return existing
review = RootReviewRecord(
review_id=review_id,
root_task_id=root.task_id,
reviewed_at=reviewed_at,
trigger_child_task_id=child.task_id,
root_revision_before=root.revision,
root_revision_after=root.revision + 1,
completed_children=completed,
active_children=active,
blocked_children=blocked,
remaining_acceptance_conditions=acceptance.remaining,
next_legal_action=action,
root_complete=(action == NextLegalAction.ROOT_COMPLETE),
evidence_refs=evidence,
)
if store.compare_and_swap_root_review(root.task_id, root.revision, review):
return review
# Another sibling may have updated root progress. Re-read authority and
# recompute; never overwrite a newer root review.
raise LineageError("root review CAS retry budget exhausted")
@@ -0,0 +1,285 @@
from __future__ import annotations
from dataclasses import dataclass, replace
from datetime import datetime
from typing import Mapping
if not __package__:
from aios_task_lineage import AcceptanceEvaluator, RelationType, TaskLineage, lineage_to_metadata, lineage_from_metadata, root_review_to_record
from aios_task_lineage_integration import (
DISPLAY_NAME_METADATA_KEY,
IntegrationError,
LINEAGE_METADATA_KEY,
NewTaskIdentityRequest,
PreparedTaskIdentity,
TaskIdentityLineageService,
TerminalReviewOutcome,
)
from aios_task_lineage_taskplane_adapter import (
AuthenticatedTaskContext,
InProcessTaskPlaneLineageAdapter,
require_context_matches_task,
)
else:
from .aios_task_lineage import AcceptanceEvaluator, RelationType, TaskLineage, lineage_to_metadata, lineage_from_metadata, root_review_to_record
from .aios_task_lineage_integration import (
DISPLAY_NAME_METADATA_KEY,
IntegrationError,
LINEAGE_METADATA_KEY,
NewTaskIdentityRequest,
PreparedTaskIdentity,
TaskIdentityLineageService,
TerminalReviewOutcome,
)
from .aios_task_lineage_taskplane_adapter import (
AuthenticatedTaskContext,
InProcessTaskPlaneLineageAdapter,
require_context_matches_task,
)
class LifecycleHookError(IntegrationError):
pass
@dataclass(frozen=True)
class IdentityPersistenceOutcome:
prepared: PreparedTaskIdentity
persisted_revision_at_least: int
independently_verified: bool
@dataclass(frozen=True)
class LegacyAttachOutcome:
lineage: TaskLineage
independently_verified: bool
@dataclass(frozen=True)
class TerminalHookOutcome:
outcome: TerminalReviewOutcome
root_review_persisted: bool
dispatch_persisted: bool
independently_verified: bool
class TaskLineageLifecycleHooks:
"""Concrete lifecycle hook coordinator.
It does not own TaskPlane state. It orchestrates existing TaskPlane CAS,
journal/continuation and optional independent authority readback.
"""
def __init__(self, adapter: InProcessTaskPlaneLineageAdapter):
self._adapter = adapter
self._service = TaskIdentityLineageService(adapter)
def persist_new_task_identity(
self,
*,
request: NewTaskIdentityRequest,
context: AuthenticatedTaskContext,
max_cas_retries: int = 4,
require_independent_readback: bool = True,
) -> IdentityPersistenceOutcome:
if max_cas_retries < 1:
raise LifecycleHookError("max_cas_retries must be >= 1")
if require_independent_readback and not self._adapter.has_independent_readback:
raise LifecycleHookError("independent TaskPlane readback is not configured")
if request.owner_principal_id != context.principal_id:
raise LifecycleHookError("request principal does not match authenticated context")
if request.mission_id != context.mission_id:
raise LifecycleHookError("request mission does not match authenticated context")
for _ in range(max_cas_retries):
current = self._adapter.get_task_snapshot(request.task_id)
require_context_matches_task(context, current)
if current.created_at != request.created_at:
raise LifecycleHookError("request created_at does not match authority")
if (request.relation_type == RelationType.ROOT
and request.preferred_owner_alias not in (None, context.preferred_alias)):
raise LifecycleHookError("request alias does not match authenticated context")
request = replace(request, preferred_owner_alias=context.preferred_alias)
prepared = self._service.prepare_new_task_identity(request)
# Idempotent replay and legacy display-name repair are fail-closed:
# never overwrite an existing conflicting lineage record.
raw = current.metadata.get(LINEAGE_METADATA_KEY)
display = current.metadata.get(DISPLAY_NAME_METADATA_KEY)
if raw is not None and not isinstance(raw, Mapping):
raise LifecycleHookError("existing lineage metadata is malformed")
if isinstance(raw, Mapping):
# parent_revision records the creation observation, not the
# parent's current progress revision on an idempotent replay.
expected = dict(prepared.metadata_patch[LINEAGE_METADATA_KEY])
if request.parent_task_id is not None:
stored = lineage_from_metadata(dict(raw))
if (stored.parent_revision is None
or stored.parent_revision > prepared.lineage.parent_revision):
raise LifecycleHookError("existing parent revision conflicts with authority")
expected["parent_revision"] = stored.parent_revision
if raw != expected:
raise LifecycleHookError(
"existing lineage metadata conflicts with requested identity"
)
prepared = replace(prepared, lineage=lineage_from_metadata(dict(raw)),
metadata_patch={LINEAGE_METADATA_KEY: dict(raw),
DISPLAY_NAME_METADATA_KEY: prepared.display_name})
if display == prepared.display_name:
verified = False
if require_independent_readback:
self._adapter.verify_task_readback(
task_id=request.task_id,
expected_revision_at_least=current.revision,
expected_metadata=prepared.metadata_patch,
)
verified = True
return IdentityPersistenceOutcome(
prepared=prepared,
persisted_revision_at_least=current.revision,
independently_verified=verified,
)
if display not in (None, ""):
raise LifecycleHookError(
"existing display name conflicts with requested identity"
)
metadata_patch = {
DISPLAY_NAME_METADATA_KEY: prepared.display_name,
}
else:
if display not in (None, ""):
raise LifecycleHookError(
"display name exists without authoritative lineage"
)
metadata_patch = prepared.metadata_patch
ok = self._adapter.compare_and_swap_task_metadata(
task_id=request.task_id,
expected_revision=current.revision,
metadata_patch=metadata_patch,
)
if not ok:
continue
verified = False
if require_independent_readback:
self._adapter.verify_task_readback(
task_id=request.task_id,
expected_revision_at_least=current.revision + 1,
expected_metadata=prepared.metadata_patch,
)
verified = True
return IdentityPersistenceOutcome(
prepared=prepared,
persisted_revision_at_least=current.revision + 1,
independently_verified=verified,
)
raise LifecycleHookError("identity persistence CAS retry budget exhausted")
def attach_legacy_identity(
self,
*,
task_id: str,
parent_task_id: str,
relation_type: RelationType,
goal_slug: str,
context: AuthenticatedTaskContext,
max_cas_retries: int = 4,
require_independent_readback: bool = True,
) -> LegacyAttachOutcome:
if require_independent_readback and not self._adapter.has_independent_readback:
raise LifecycleHookError("independent TaskPlane readback is not configured")
before = self._adapter.get_task_snapshot(task_id)
require_context_matches_task(context, before)
lineage = self._service.attach_legacy_task(
task_id=task_id,
parent_task_id=parent_task_id,
relation_type=relation_type,
goal_slug=goal_slug,
max_cas_retries=max_cas_retries,
expected_owner_principal_id=context.principal_id,
expected_mission_id=context.mission_id,
)
verified = False
if require_independent_readback:
self._adapter.verify_task_readback(
task_id=task_id,
expected_revision_at_least=before.revision,
expected_metadata={
LINEAGE_METADATA_KEY: lineage_to_metadata(lineage),
DISPLAY_NAME_METADATA_KEY: lineage.display_name,
},
)
# Domain serialization compatibility check below avoids requiring
# TaskLineage to expose a method in older versions.
verified = True
return LegacyAttachOutcome(lineage=lineage, independently_verified=verified)
def on_proven_terminal_child(
self,
*,
child_task_id: str,
context: AuthenticatedTaskContext,
acceptance_evaluator: AcceptanceEvaluator,
reviewed_at: datetime,
max_cas_retries: int = 4,
require_independent_readback: bool = True,
) -> TerminalHookOutcome:
if require_independent_readback and not self._adapter.has_independent_readback:
raise LifecycleHookError("independent TaskPlane readback is not configured")
child = self._adapter.get_task_snapshot(child_task_id)
require_context_matches_task(context, child)
if not child.terminal_evidence_proven:
raise LifecycleHookError(
"terminal child hook requires proven terminal evidence"
)
raw = child.metadata.get(LINEAGE_METADATA_KEY)
if not isinstance(raw, Mapping):
raise LifecycleHookError("terminal child lineage missing")
lineage = lineage_from_metadata(dict(raw))
root = self._adapter.get_task_snapshot(lineage.root_task_id)
require_context_matches_task(context, root)
def authorized_acceptance(root_record, descendants):
if (root_record.lineage.owner_principal_id != context.principal_id
or root_record.lineage.mission_id != context.mission_id):
raise LifecycleHookError("root authority changed during terminal review")
return acceptance_evaluator(root_record, descendants)
outcome = self._service.on_proven_terminal_child(
child_task_id=child_task_id,
acceptance_evaluator=authorized_acceptance,
reviewed_at=reviewed_at,
max_cas_retries=max_cas_retries,
)
persisted_review = self._adapter.get_root_review_record(
outcome.review.review_id
)
if persisted_review is None:
raise LifecycleHookError("root review was not persisted")
if persisted_review != root_review_to_record(outcome.review):
raise LifecycleHookError("persisted review does not match returned review")
dispatch = self._adapter.get_root_action_dispatch(outcome.review.review_id)
if dispatch is None:
raise LifecycleHookError("root action dispatch was not persisted")
if dispatch != outcome.dispatch:
raise LifecycleHookError("persisted dispatch does not match returned receipt")
verified = False
if require_independent_readback:
self._adapter.verify_root_readback(
root_task_id=outcome.review.root_task_id,
expected_review_id=outcome.review.review_id,
)
verified = True
return TerminalHookOutcome(
outcome=outcome,
root_review_persisted=True,
dispatch_persisted=True,
independently_verified=verified,
)
@@ -0,0 +1,539 @@
from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime
from enum import Enum
from typing import Mapping, MutableMapping, Protocol, Sequence
if not __package__:
from aios_task_lineage import (
AcceptanceEvaluator,
LineageError,
NextLegalAction,
RelationType,
RootReviewRecord,
RootReviewStore,
TaskLineage,
TaskRecord,
attach_legacy_lineage,
derive_safe_owner_alias,
lineage_from_metadata,
lineage_to_metadata,
make_child_lineage,
make_root_lineage,
normalize_goal_slug,
review_root_after_child_terminal,
root_review_from_record,
root_review_to_record,
)
else:
from .aios_task_lineage import (
AcceptanceEvaluator,
LineageError,
NextLegalAction,
RelationType,
RootReviewRecord,
RootReviewStore,
TaskLineage,
TaskRecord,
attach_legacy_lineage,
derive_safe_owner_alias,
lineage_from_metadata,
lineage_to_metadata,
make_child_lineage,
make_root_lineage,
normalize_goal_slug,
review_root_after_child_terminal,
root_review_from_record,
root_review_to_record,
)
LINEAGE_METADATA_KEY = "aios_task_lineage"
LINEAGE_REQUEST_DIGEST_KEY = "aios_lineage_request_digest"
ROOT_REVIEW_RECORD_KEY = "aios_root_review"
DISPLAY_NAME_METADATA_KEY = "aios_task_display_name"
class IntegrationError(RuntimeError):
pass
@dataclass(frozen=True)
class TaskAuthoritySnapshot:
task_id: str
revision: int
status: str
owner_principal_id: str
mission_id: str
created_at: datetime
metadata: Mapping[str, object]
evidence_refs: tuple[str, ...] = ()
terminal_evidence_proven: bool = False
@dataclass(frozen=True)
class NewTaskIdentityRequest:
task_id: str
owner_principal_id: str
preferred_owner_alias: str | None
created_at: datetime
mission_id: str
goal_slug: str
root_acceptance_ref: str
relation_type: RelationType = RelationType.ROOT
parent_task_id: str | None = None
@dataclass(frozen=True)
class PreparedTaskIdentity:
task_id: str
display_name: str
lineage: TaskLineage
metadata_patch: Mapping[str, object]
@dataclass(frozen=True)
class TerminalReviewOutcome:
review: RootReviewRecord
dispatch: "RootActionDispatchReceipt"
@dataclass(frozen=True)
class RootActionDispatchReceipt:
review_id: str
root_task_id: str
action: NextLegalAction
idempotency_key: str
status: str
mission_id: str = ""
root_revision: int = 0
plan_digest: str | None = None
execution_key: str | None = None
PERSISTED_DISPATCH_STATUSES = frozenset(
{
"PENDING",
"CLAIMED",
"EFFECT_UNKNOWN",
"EXECUTED_PENDING_REREVIEW",
"WAITING_DEPENDENCY",
"WAITING_OWNER",
"CLOSED",
"SUPERSEDED",
"BLOCKED",
# Read compatibility only. A consumer must reconcile this legacy
# state unless a complete durable receipt is present.
"DISPATCHED",
}
)
class TaskPlaneLineagePort(Protocol):
"""Narrow authority port required by the lineage integration service.
The live adapter must implement these methods over the existing TaskPlane /
mission / journal / CAS mechanisms. This protocol deliberately avoids
guessing concrete DB tables, HTTP endpoints, or handler names.
"""
def get_task_snapshot(self, task_id: str) -> TaskAuthoritySnapshot: ...
def list_descendant_snapshots(
self, root_task_id: str
) -> Sequence[TaskAuthoritySnapshot]: ...
def reserve_sibling_sequence(
self,
*,
parent_task_id: str,
relation_type: RelationType,
child_task_id: str,
) -> int:
"""Idempotently reserve one sibling sequence for child_task_id."""
def compare_and_swap_task_metadata(
self,
*,
task_id: str,
expected_revision: int,
metadata_patch: Mapping[str, object],
) -> bool: ...
def get_root_review_record(
self, review_id: str
) -> Mapping[str, object] | None: ...
def compare_and_swap_root_review_record(
self,
*,
root_task_id: str,
expected_root_revision: int,
review_record: Mapping[str, object],
) -> bool: ...
def get_root_action_dispatch(
self, review_id: str
) -> RootActionDispatchReceipt | None: ...
def dispatch_root_action(
self,
*,
review: RootReviewRecord,
idempotency_key: str,
) -> RootActionDispatchReceipt: ...
def _extract_lineage(snapshot: TaskAuthoritySnapshot) -> TaskLineage:
raw = snapshot.metadata.get(LINEAGE_METADATA_KEY)
if not isinstance(raw, dict):
raise IntegrationError(
f"task {snapshot.task_id} does not have authoritative lineage metadata"
)
lineage = lineage_from_metadata(raw)
if lineage.task_id != snapshot.task_id:
raise IntegrationError("lineage task_id does not match authority task")
if lineage.owner_principal_id != snapshot.owner_principal_id:
raise IntegrationError("lineage owner does not match authority owner")
if lineage.mission_id != snapshot.mission_id:
raise IntegrationError("lineage mission does not match authority mission")
if snapshot.created_at.tzinfo is None or snapshot.created_at.utcoffset() is None:
raise IntegrationError("authority created_at must be timezone-aware")
if lineage.created_at != snapshot.created_at:
raise IntegrationError("lineage created_at does not match authority task")
display = snapshot.metadata.get(DISPLAY_NAME_METADATA_KEY)
if display not in (None, "", lineage.display_name):
raise IntegrationError("lineage display name conflicts with authority metadata")
return lineage
def _as_task_record(snapshot: TaskAuthoritySnapshot) -> TaskRecord:
return TaskRecord(
task_id=snapshot.task_id,
revision=snapshot.revision,
status=snapshot.status,
lineage=_extract_lineage(snapshot),
evidence_refs=snapshot.evidence_refs,
terminal_evidence_proven=snapshot.terminal_evidence_proven,
)
def _validate_record_root_binding(record: TaskRecord, root: TaskRecord) -> None:
if root.lineage.relation_type != RelationType.ROOT:
raise IntegrationError("root_task_id does not resolve to a ROOT task")
if record.lineage.root_task_id != root.task_id:
raise IntegrationError("task lineage points to a different root")
if record.lineage.owner_principal_id != root.lineage.owner_principal_id:
raise IntegrationError("task lineage owner does not match root owner")
if record.lineage.mission_id != root.lineage.mission_id:
raise IntegrationError("task lineage mission does not match root mission")
if record.lineage.root_acceptance_ref != root.lineage.root_acceptance_ref:
raise IntegrationError("task lineage acceptance reference does not match root")
def _load_authority_record(
port: TaskPlaneLineagePort, task_id: str
) -> tuple[TaskRecord, TaskRecord]:
record = _as_task_record(port.get_task_snapshot(task_id))
if record.lineage.relation_type == RelationType.ROOT:
return record, record
root = _as_task_record(port.get_task_snapshot(record.lineage.root_task_id))
current = record
seen = {record.task_id}
while current.lineage.relation_type != RelationType.ROOT:
_validate_record_root_binding(current, root)
parent_id = current.lineage.parent_task_id
if not parent_id or parent_id in seen:
raise IntegrationError("task lineage contains a parent cycle")
seen.add(parent_id)
parent = _as_task_record(port.get_task_snapshot(parent_id))
_validate_record_root_binding(parent, root)
if current.lineage.parent_task_id != parent.task_id:
raise IntegrationError("task lineage parent does not match authority parent")
if current.lineage.lineage_depth != parent.lineage.lineage_depth + 1:
raise IntegrationError("task lineage depth does not match authority parent")
if current.lineage.parent_revision is None or (
current.lineage.parent_revision > parent.revision
):
raise IntegrationError("task lineage parent revision is not authoritative")
current = parent
if current.task_id != root.task_id:
raise IntegrationError("task lineage parent chain ends at a different root")
return record, root
def _validate_dispatch_receipt(
receipt: RootActionDispatchReceipt,
*,
review: RootReviewRecord,
idempotency_key: str,
existing: bool,
) -> None:
if (
receipt.review_id != review.review_id
or receipt.root_task_id != review.root_task_id
or receipt.action != review.next_legal_action
or receipt.idempotency_key != idempotency_key
or receipt.status not in PERSISTED_DISPATCH_STATUSES
):
noun = "existing root-action dispatch" if existing else "root-action dispatch receipt"
raise IntegrationError(f"{noun} conflicts with persisted review")
class _RootReviewStoreAdapter(RootReviewStore):
def __init__(self, port: TaskPlaneLineagePort):
self._port = port
def get_task(self, task_id: str) -> TaskRecord:
record, _ = _load_authority_record(self._port, task_id)
return record
def list_descendants(self, root_task_id: str) -> Sequence[TaskRecord]:
root, _ = _load_authority_record(self._port, root_task_id)
if root.lineage.relation_type != RelationType.ROOT:
raise IntegrationError("descendant lookup key is not a ROOT task")
records: list[TaskRecord] = []
for snapshot in self._port.list_descendant_snapshots(root_task_id):
record, resolved_root = _load_authority_record(self._port, snapshot.task_id)
if resolved_root.task_id != root_task_id:
raise IntegrationError("descendant authority returned a cross-root task")
_validate_record_root_binding(record, root)
records.append(record)
return tuple(records)
def get_root_review(self, review_id: str) -> RootReviewRecord | None:
raw = self._port.get_root_review_record(review_id)
if raw is None:
return None
if not isinstance(raw, Mapping):
raise IntegrationError("persisted root-review record must be a mapping")
try:
return root_review_from_record(
raw,
expected_review_id=review_id,
)
except LineageError as exc:
raise IntegrationError(
f"invalid persisted root-review record: {exc}"
) from exc
def compare_and_swap_root_review(
self,
root_task_id: str,
expected_root_revision: int,
review: RootReviewRecord,
) -> bool:
return self._port.compare_and_swap_root_review_record(
root_task_id=root_task_id,
expected_root_revision=expected_root_revision,
review_record=root_review_to_record(review),
)
class TaskIdentityLineageService:
"""Application service that wires the lineage domain to TaskPlane authority."""
def __init__(self, port: TaskPlaneLineagePort):
self._port = port
def prepare_new_task_identity(
self, request: NewTaskIdentityRequest
) -> PreparedTaskIdentity:
alias = derive_safe_owner_alias(
request.owner_principal_id, request.preferred_owner_alias
)
if request.relation_type == RelationType.ROOT:
if request.parent_task_id is not None:
raise IntegrationError("ROOT task cannot have parent_task_id")
lineage = make_root_lineage(
task_id=request.task_id,
owner_principal_id=request.owner_principal_id,
owner_alias=alias,
created_at=request.created_at,
mission_id=request.mission_id,
goal_slug=request.goal_slug,
root_acceptance_ref=request.root_acceptance_ref,
)
else:
if not request.parent_task_id:
raise IntegrationError("non-root task requires parent_task_id")
parent, root = _load_authority_record(
self._port, request.parent_task_id
)
if request.owner_principal_id != root.lineage.owner_principal_id:
raise IntegrationError("cross-owner child creation rejected")
if request.mission_id != root.lineage.mission_id:
raise IntegrationError("cross-mission child creation rejected")
if request.root_acceptance_ref != root.lineage.root_acceptance_ref:
raise IntegrationError("root acceptance reference mismatch")
sequence = self._port.reserve_sibling_sequence(
parent_task_id=parent.task_id,
relation_type=request.relation_type,
child_task_id=request.task_id,
)
lineage = make_child_lineage(
task_id=request.task_id,
created_at=request.created_at,
parent=parent,
root=root,
relation_type=request.relation_type,
sibling_sequence=sequence,
goal_slug=request.goal_slug,
)
return PreparedTaskIdentity(
task_id=request.task_id,
display_name=lineage.display_name,
lineage=lineage,
metadata_patch={
LINEAGE_METADATA_KEY: lineage_to_metadata(lineage),
DISPLAY_NAME_METADATA_KEY: lineage.display_name,
},
)
def attach_legacy_task(
self,
*,
task_id: str,
parent_task_id: str,
relation_type: RelationType,
goal_slug: str,
expected_owner_principal_id: str | None = None,
expected_mission_id: str | None = None,
max_cas_retries: int = 4,
) -> TaskLineage:
if max_cas_retries < 1:
raise IntegrationError("max_cas_retries must be >= 1")
for _ in range(max_cas_retries):
current = self._port.get_task_snapshot(task_id)
if (
expected_owner_principal_id is not None
and current.owner_principal_id != expected_owner_principal_id
):
raise IntegrationError("legacy task owner changed after authorization")
if (
expected_mission_id is not None
and current.mission_id != expected_mission_id
):
raise IntegrationError("legacy task mission changed after authorization")
raw = current.metadata.get(LINEAGE_METADATA_KEY)
if isinstance(raw, Mapping):
record, _ = _load_authority_record(self._port, task_id)
lineage = record.lineage
expected_goal_slug = normalize_goal_slug(goal_slug)
if (
lineage.parent_task_id != parent_task_id
or lineage.relation_type != relation_type
or lineage.goal_slug != expected_goal_slug
):
raise IntegrationError(
"existing legacy lineage conflicts with requested binding"
)
display = current.metadata.get(DISPLAY_NAME_METADATA_KEY)
if display == lineage.display_name:
return lineage
if display not in (None, ""):
raise IntegrationError(
"legacy display name conflicts with authoritative lineage"
)
if self._port.compare_and_swap_task_metadata(
task_id=current.task_id,
expected_revision=current.revision,
metadata_patch={
DISPLAY_NAME_METADATA_KEY: lineage.display_name,
},
):
return lineage
continue
parent, root = _load_authority_record(self._port, parent_task_id)
if root.status == "COMPLETED":
raise IntegrationError("cannot attach lineage below a completed root")
if (
expected_owner_principal_id is not None
and root.lineage.owner_principal_id != expected_owner_principal_id
):
raise IntegrationError("legacy root owner does not match authorization")
if (
expected_mission_id is not None
and root.lineage.mission_id != expected_mission_id
):
raise IntegrationError("legacy root mission does not match authorization")
if current.owner_principal_id != root.lineage.owner_principal_id:
raise IntegrationError("legacy task owner does not match root owner")
if current.mission_id != root.lineage.mission_id:
raise IntegrationError("legacy task mission does not match root mission")
sequence = self._port.reserve_sibling_sequence(
parent_task_id=parent.task_id,
relation_type=relation_type,
child_task_id=current.task_id,
)
lineage = attach_legacy_lineage(
existing_task_id=current.task_id,
created_at=current.created_at,
parent=parent,
root=root,
relation_type=relation_type,
sibling_sequence=sequence,
goal_slug=goal_slug,
)
if self._port.compare_and_swap_task_metadata(
task_id=current.task_id,
expected_revision=current.revision,
metadata_patch={
LINEAGE_METADATA_KEY: lineage_to_metadata(lineage),
DISPLAY_NAME_METADATA_KEY: lineage.display_name,
},
):
return lineage
raise IntegrationError("legacy lineage CAS retry budget exhausted")
def on_proven_terminal_child(
self,
*,
child_task_id: str,
acceptance_evaluator: AcceptanceEvaluator,
reviewed_at: datetime,
max_cas_retries: int = 4,
) -> TerminalReviewOutcome:
store = _RootReviewStoreAdapter(self._port)
review = review_root_after_child_terminal(
store=store,
trigger_child_task_id=child_task_id,
acceptance_evaluator=acceptance_evaluator,
reviewed_at=reviewed_at,
max_cas_retries=max_cas_retries,
)
existing = self._port.get_root_action_dispatch(review.review_id)
if existing is not None:
idempotency_key = (
f"root-review:{review.review_id}:{review.next_legal_action.value}"
)
_validate_dispatch_receipt(
existing,
review=review,
idempotency_key=idempotency_key,
existing=True,
)
return TerminalReviewOutcome(review=review, dispatch=existing)
idempotency_key = f"root-review:{review.review_id}:{review.next_legal_action.value}"
receipt = self._port.dispatch_root_action(
review=review,
idempotency_key=idempotency_key,
)
_validate_dispatch_receipt(
receipt,
review=review,
idempotency_key=idempotency_key,
existing=False,
)
return TerminalReviewOutcome(review=review, dispatch=receipt)
@@ -0,0 +1,464 @@
from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime, timezone
import json
from pathlib import Path
import sqlite3
from typing import Callable, Mapping, Protocol, Sequence, runtime_checkable
if not __package__:
from aios_task_lineage import RelationType, RootReviewRecord, lineage_from_metadata
from aios_task_lineage_integration import (
IntegrationError,
LINEAGE_METADATA_KEY,
RootActionDispatchReceipt,
TaskAuthoritySnapshot,
TaskPlaneLineagePort,
)
else:
from .aios_task_lineage import RelationType, RootReviewRecord, lineage_from_metadata
from .aios_task_lineage_integration import (
IntegrationError,
LINEAGE_METADATA_KEY,
RootActionDispatchReceipt,
TaskAuthoritySnapshot,
TaskPlaneLineagePort,
)
class AuthorityAdapterError(IntegrationError):
pass
def _parse_datetime(value: object) -> datetime:
if isinstance(value, datetime):
if value.tzinfo is None or value.utcoffset() is None:
raise AuthorityAdapterError("authority datetime must be timezone-aware")
return value.astimezone(timezone.utc)
if not isinstance(value, str) or not value:
raise AuthorityAdapterError("authority datetime missing")
try:
dt = datetime.fromisoformat(value.replace("Z", "+00:00"))
except ValueError as exc:
raise AuthorityAdapterError("authority datetime invalid") from exc
if dt.tzinfo is None or dt.utcoffset() is None:
raise AuthorityAdapterError("authority datetime must include timezone")
return dt.astimezone(timezone.utc)
def _load_json_object(value: object, *, field: str) -> dict[str, object]:
if value in (None, ""):
return {}
if isinstance(value, Mapping):
return dict(value)
if not isinstance(value, str):
raise AuthorityAdapterError(f"{field} must be a JSON object")
try:
obj = json.loads(value)
except json.JSONDecodeError as exc:
raise AuthorityAdapterError(f"{field} is invalid JSON") from exc
if not isinstance(obj, dict):
raise AuthorityAdapterError(f"{field} must decode to an object")
return obj
def _metadata_identity(metadata: Mapping[str, object]) -> tuple[str, str]:
lineage = metadata.get(LINEAGE_METADATA_KEY)
if isinstance(lineage, Mapping):
owner = lineage.get("owner_principal_id")
mission = lineage.get("mission_id")
if isinstance(owner, str) and owner and isinstance(mission, str) and mission:
return owner, mission
# Pre-lineage/legacy compatibility. These keys are accepted only as
# authority-provided metadata; missing values fail closed rather than being
# invented by this adapter.
owner = metadata.get("owner_principal_id")
mission = metadata.get("mission_id")
if not isinstance(owner, str) or not owner:
raise AuthorityAdapterError("task owner_principal_id is not discoverable")
if not isinstance(mission, str) or not mission:
raise AuthorityAdapterError("task mission_id is not discoverable")
return owner, mission
def _metadata_evidence(metadata: Mapping[str, object]) -> tuple[str, ...]:
raw = metadata.get("evidence_refs")
if raw is None:
return ()
if not isinstance(raw, list) or not all(isinstance(x, str) for x in raw):
raise AuthorityAdapterError("evidence_refs must be a string list")
return tuple(dict.fromkeys(raw))
@runtime_checkable
class TaskPlaneReadback(Protocol):
"""Independent authority readback surface.
This may be backed by the TaskPlane query service, a read replica, or a
read-only SQLite authority. It must not mutate task state.
"""
def get_task_snapshot(self, task_id: str) -> TaskAuthoritySnapshot: ...
def list_descendant_snapshots(
self, root_task_id: str
) -> Sequence[TaskAuthoritySnapshot]: ...
def get_root_review_record(self, review_id: str) -> Mapping[str, object] | None: ...
def get_root_action_dispatch(self, review_id: str) -> RootActionDispatchReceipt | None: ...
@dataclass(frozen=True)
class CallableTaskPlaneReadback:
"""Readback adapter for an existing official TaskPlane query/service."""
get_task: Callable[[str], TaskAuthoritySnapshot]
list_descendants: Callable[[str], Sequence[TaskAuthoritySnapshot]]
get_review: Callable[[str], Mapping[str, object] | None] | None = None
get_dispatch: Callable[[str], RootActionDispatchReceipt | None] | None = None
def get_task_snapshot(self, task_id: str) -> TaskAuthoritySnapshot:
snapshot = self.get_task(task_id)
if not isinstance(snapshot, TaskAuthoritySnapshot):
raise AuthorityAdapterError("readback returned invalid task snapshot")
if snapshot.task_id != task_id:
raise AuthorityAdapterError("readback returned wrong task identity")
return snapshot
def list_descendant_snapshots(
self, root_task_id: str
) -> Sequence[TaskAuthoritySnapshot]:
values = tuple(self.list_descendants(root_task_id))
if not all(isinstance(v, TaskAuthoritySnapshot) for v in values):
raise AuthorityAdapterError("readback returned invalid descendant snapshot")
return values
def get_root_review_record(self, review_id: str) -> Mapping[str, object] | None:
if self.get_review is None:
raise AuthorityAdapterError("independent review readback is not configured")
return self.get_review(review_id)
def get_root_action_dispatch(self, review_id: str) -> RootActionDispatchReceipt | None:
if self.get_dispatch is None:
raise AuthorityAdapterError("independent dispatch readback is not configured")
return self.get_dispatch(review_id)
@dataclass(frozen=True)
class AuthoritySchemaReport:
task_table: str
columns: tuple[str, ...]
required_columns_present: bool
missing_columns: tuple[str, ...]
task_count: int
class ReadOnlySQLiteTaskPlane:
"""Read-only discovery/readback helper for the current TaskPlane SQLite authority.
This class never writes. It opens SQLite read-only, including committed WAL
data, and exists only for discovery, diagnostics and independent readback.
Execution writes must go through the live TaskPlane repository/service and
are supplied separately through ``TaskPlaneServiceBindings``.
"""
REQUIRED_TASK_COLUMNS = (
"task_id",
"status",
"revision",
"metadata_json",
"created_at",
)
def __init__(self, db_path: Path):
self.db_path = Path(db_path)
def _connect(self) -> sqlite3.Connection:
if not self.db_path.is_file():
raise AuthorityAdapterError(f"TaskPlane DB missing: {self.db_path}")
# A live WAL database is not immutable: ignoring its WAL can return
# stale state. as_uri also quotes URI metacharacters in file names.
uri = self.db_path.resolve().as_uri() + "?mode=ro"
con = sqlite3.connect(uri, uri=True)
con.row_factory = sqlite3.Row
return con
def inspect_schema(self) -> AuthoritySchemaReport:
con = self._connect()
try:
rows = list(con.execute("PRAGMA table_info(tasks)"))
columns = tuple(str(r["name"]) for r in rows)
missing = tuple(c for c in self.REQUIRED_TASK_COLUMNS if c not in columns)
count = int(con.execute("SELECT COUNT(*) FROM tasks").fetchone()[0]) if rows else 0
return AuthoritySchemaReport(
task_table="tasks",
columns=columns,
required_columns_present=not missing,
missing_columns=missing,
task_count=count,
)
finally:
con.close()
def _row_to_snapshot(self, row: sqlite3.Row) -> TaskAuthoritySnapshot:
keys = set(row.keys())
metadata = _load_json_object(
row["metadata_json"] if "metadata_json" in keys else None,
field="metadata_json",
)
owner, mission = _metadata_identity(metadata)
created_value = row["created_at"] if "created_at" in keys else None
return TaskAuthoritySnapshot(
task_id=str(row["task_id"]),
revision=int(row["revision"]),
status=str(row["status"]),
owner_principal_id=owner,
mission_id=mission,
created_at=_parse_datetime(created_value),
metadata=metadata,
evidence_refs=_metadata_evidence(metadata),
# A metadata flag cannot stand in for execution-integrity proof.
# The official service adapter must resolve verified receipts.
terminal_evidence_proven=False,
)
def get_task_snapshot(self, task_id: str) -> TaskAuthoritySnapshot:
con = self._connect()
try:
row = con.execute("SELECT * FROM tasks WHERE task_id=? LIMIT 1", (task_id,)).fetchone()
if row is None:
raise AuthorityAdapterError(f"task not found: {task_id}")
return self._row_to_snapshot(row)
finally:
con.close()
def list_descendant_snapshots(self, root_task_id: str) -> tuple[TaskAuthoritySnapshot, ...]:
con = self._connect()
try:
rows = list(con.execute("SELECT * FROM tasks ORDER BY task_id"))
finally:
con.close()
out: list[TaskAuthoritySnapshot] = []
for row in rows:
try:
snap = self._row_to_snapshot(row)
except AuthorityAdapterError:
# Legacy tasks without discoverable identity cannot safely be
# attached to a root by this read-only adapter.
continue
raw = snap.metadata.get(LINEAGE_METADATA_KEY)
if (
isinstance(raw, Mapping)
and raw.get("root_task_id") == root_task_id
and snap.task_id != root_task_id
):
out.append(snap)
return tuple(out)
@dataclass(frozen=True)
class TaskPlaneServiceBindings:
"""In-process bindings to the existing TaskPlane authority.
These callables must be backed by the real TaskPlane repository/service,
including its existing transactions, revision/CAS, journal and continuation
mechanisms. They must NOT be implemented as out-of-band SQLite mutations.
"""
get_task_snapshot: Callable[[str], TaskAuthoritySnapshot]
list_descendant_snapshots: Callable[[str], Sequence[TaskAuthoritySnapshot]]
reserve_sibling_sequence: Callable[[str, RelationType, str], int]
compare_and_swap_task_metadata: Callable[[str, int, Mapping[str, object]], bool]
get_root_review_record: Callable[[str], Mapping[str, object] | None]
compare_and_swap_root_review_record: Callable[
[str, int, Mapping[str, object]], bool
]
get_root_action_dispatch: Callable[[str], RootActionDispatchReceipt | None]
dispatch_root_action: Callable[
[RootReviewRecord, str], RootActionDispatchReceipt
]
class InProcessTaskPlaneLineageAdapter(TaskPlaneLineagePort):
"""Concrete TaskPlaneLineagePort backed by existing TaskPlane service calls."""
def __init__(
self,
bindings: TaskPlaneServiceBindings,
*,
readback: TaskPlaneReadback | None = None,
):
self._b = bindings
self._readback = readback
def get_task_snapshot(self, task_id: str) -> TaskAuthoritySnapshot:
return self._b.get_task_snapshot(task_id)
def list_descendant_snapshots(self, root_task_id: str) -> Sequence[TaskAuthoritySnapshot]:
return tuple(self._b.list_descendant_snapshots(root_task_id))
def reserve_sibling_sequence(
self,
*,
parent_task_id: str,
relation_type: RelationType,
child_task_id: str,
) -> int:
seq = self._b.reserve_sibling_sequence(
parent_task_id, relation_type, child_task_id
)
if not isinstance(seq, int) or seq < 1:
raise AuthorityAdapterError("TaskPlane returned invalid sibling sequence")
return seq
def compare_and_swap_task_metadata(
self,
*,
task_id: str,
expected_revision: int,
metadata_patch: Mapping[str, object],
) -> bool:
return bool(
self._b.compare_and_swap_task_metadata(
task_id, expected_revision, metadata_patch
)
)
def get_root_review_record(self, review_id: str) -> Mapping[str, object] | None:
return self._b.get_root_review_record(review_id)
def compare_and_swap_root_review_record(
self,
*,
root_task_id: str,
expected_root_revision: int,
review_record: Mapping[str, object],
) -> bool:
return bool(
self._b.compare_and_swap_root_review_record(
root_task_id, expected_root_revision, review_record
)
)
def get_root_action_dispatch(
self, review_id: str
) -> RootActionDispatchReceipt | None:
return self._b.get_root_action_dispatch(review_id)
def dispatch_root_action(
self,
*,
review: RootReviewRecord,
idempotency_key: str,
) -> RootActionDispatchReceipt:
return self._b.dispatch_root_action(review, idempotency_key)
def verify_task_readback(
self,
*,
task_id: str,
expected_revision_at_least: int,
expected_metadata: Mapping[str, object],
) -> TaskAuthoritySnapshot:
if self._readback is None:
raise AuthorityAdapterError("independent TaskPlane readback is not configured")
snap = self._readback.get_task_snapshot(task_id)
if snap.task_id != task_id:
raise AuthorityAdapterError("TaskPlane readback task identity mismatch")
if snap.revision < expected_revision_at_least:
raise AuthorityAdapterError("TaskPlane readback revision did not advance")
for key, expected in expected_metadata.items():
if snap.metadata.get(key) != expected:
raise AuthorityAdapterError(
f"TaskPlane readback metadata mismatch for {key}"
)
return snap
def verify_root_readback(
self,
*,
root_task_id: str,
expected_review_id: str,
) -> tuple[TaskAuthoritySnapshot, ...]:
"""Verify that independent authority descendants remain bound to root.
The persisted ROOT_REVIEW record itself may live in a journal rather
than the task row. This method verifies the task/lineage projection
independently while the service binding verifies the journal receipt.
"""
if self._readback is None:
raise AuthorityAdapterError("independent TaskPlane readback is not configured")
review = self.get_root_review_record(expected_review_id)
dispatch = self.get_root_action_dispatch(expected_review_id)
if (review is None or review.get("review_id") != expected_review_id
or review.get("root_task_id") != root_task_id or dispatch is None):
raise AuthorityAdapterError("root review or dispatch identity mismatch")
read_review = getattr(self._readback, "get_root_review_record", None)
read_dispatch = getattr(self._readback, "get_root_action_dispatch", None)
if not callable(read_review) or not callable(read_dispatch):
raise AuthorityAdapterError("independent review/dispatch readback is not configured")
if read_review(expected_review_id) != review:
raise AuthorityAdapterError("independent root review readback mismatch")
if read_dispatch(expected_review_id) != dispatch:
raise AuthorityAdapterError("independent dispatch readback mismatch")
root = self._readback.get_task_snapshot(root_task_id)
raw_root = root.metadata.get(LINEAGE_METADATA_KEY)
if root.task_id != root_task_id or not isinstance(raw_root, Mapping):
raise AuthorityAdapterError("root readback identity or lineage missing")
root_lineage = lineage_from_metadata(dict(raw_root))
if (root_lineage.relation_type != RelationType.ROOT
or root_lineage.task_id != root_task_id
or root_lineage.owner_principal_id != root.owner_principal_id
or root_lineage.mission_id != root.mission_id):
raise AuthorityAdapterError("root readback authority mismatch")
revision = review.get("root_revision_after")
if type(revision) is not int or root.revision < revision:
raise AuthorityAdapterError("root readback revision is stale")
descendants = tuple(self._readback.list_descendant_snapshots(root_task_id))
seen = set()
for snap in descendants:
raw = snap.metadata.get(LINEAGE_METADATA_KEY)
if not isinstance(raw, Mapping):
raise AuthorityAdapterError("descendant lineage missing")
lineage = lineage_from_metadata(dict(raw))
if (snap.task_id in seen or snap.task_id == root_task_id
or lineage.task_id != snap.task_id
or lineage.root_task_id != root_task_id
or lineage.owner_principal_id != root.owner_principal_id
or lineage.mission_id != root.mission_id
or snap.owner_principal_id != root.owner_principal_id
or snap.mission_id != root.mission_id
or lineage.root_acceptance_ref != root_lineage.root_acceptance_ref):
raise AuthorityAdapterError(
f"descendant {snap.task_id} is projected under wrong root"
)
seen.add(snap.task_id)
if review.get("trigger_child_task_id") not in seen:
raise AuthorityAdapterError("trigger child missing from independent readback")
return descendants
@property
def has_independent_readback(self) -> bool:
return self._readback is not None
@dataclass(frozen=True)
class AuthenticatedTaskContext:
principal_id: str
preferred_alias: str | None
mission_id: str
def require_context_matches_task(
context: AuthenticatedTaskContext,
task: TaskAuthoritySnapshot,
) -> None:
if context.principal_id != task.owner_principal_id:
raise AuthorityAdapterError("authenticated principal does not own task")
if context.mission_id != task.mission_id:
raise AuthorityAdapterError("authenticated mission does not match task")
@@ -0,0 +1,242 @@
"""Narrow staging Mission ingress authenticated by the registered Owner signer.
The signed payload binds the method, resource, body and idempotency key. This
adapter neither signs requests nor grants Git/worker/production capabilities.
"""
from __future__ import annotations
import argparse
import importlib.util
import json
from pathlib import Path
import re
import stat
import time
from http.server import BaseHTTPRequestHandler, HTTPServer
from typing import Any
from .task_plane.errors import (
ForbiddenError, TaskPlaneHTTPError, UnauthorizedError, ValidationError,
)
AUDIENCE = "aios.task-mission-authority.staging.v1"
NAMESPACE = "aios-owner-approval-v1"
MAX_BODY = 131072
REQUEST_FIELDS = frozenset({
"schema", "audience", "method", "path", "body", "idempotency_key",
"issued_at", "expires_at", "nonce",
})
def canonical_bytes(value: Any) -> bytes:
return json.dumps(value, sort_keys=True, separators=(",", ":"),
ensure_ascii=False, allow_nan=False).encode("utf-8")
def strict_object(pairs):
result = {}
for key, value in pairs:
if key in result:
raise ValidationError("duplicate JSON key")
result[key] = value
return result
class SignedMissionIngress:
def __init__(
self, service, verifier, *, terminal_proof_importer=None, clock=time.time
):
if not callable(getattr(verifier, "verify", None)):
raise ValueError("registered Owner verifier required")
identity = getattr(verifier, "signer_id", None)
if not isinstance(identity, str) or not identity.strip():
raise ValueError("registered signer identity required")
self.service = service
self.verifier = verifier
self.identity = identity
self.terminal_proof_importer = terminal_proof_importer
self.clock = clock
def handle(self, envelope: dict) -> tuple[int, dict]:
if not isinstance(envelope, dict) or set(envelope) != {
"request", "signature", "key_id"
}:
raise ValidationError("invalid signed mission envelope")
request = envelope["request"]
if not isinstance(request, dict) or set(request) != REQUEST_FIELDS:
raise ValidationError("invalid mission request fields")
if (request["schema"] != "aios.signed-mission-request.v1"
or request["audience"] != AUDIENCE):
raise ForbiddenError("mission request audience mismatch")
issued, expires = request["issued_at"], request["expires_at"]
if (type(issued) is not int or type(expires) is not int
or not 0 < expires - issued <= 300
or not issued <= self.clock() < expires):
raise UnauthorizedError("mission request expired or not yet valid")
if (not isinstance(request["nonce"], str)
or re.fullmatch(r"[A-Za-z0-9_-]{16,128}", request["nonce"]) is None):
raise ValidationError("invalid mission request nonce")
if (not isinstance(envelope["signature"], str)
or not isinstance(envelope["key_id"], str)):
raise UnauthorizedError("Owner signature required")
try:
verified = self.verifier.verify(
namespace=NAMESPACE, payload=canonical_bytes(request),
signature=envelope["signature"], key_id=envelope["key_id"],
)
except Exception:
verified = False
if verified is not True:
raise UnauthorizedError("Owner signature invalid")
method, path, body = request["method"], request["path"], request["body"]
key = request["idempotency_key"]
if method == "POST" and path == "/missions":
if not isinstance(body, dict) or not isinstance(key, str) or not key.strip():
raise ValidationError("mission body and idempotency key required")
return 201, self.service.create_mission(
body, key, actor_identity=self.identity,
)
if (method == "GET" and isinstance(path, str)
and re.fullmatch(r"/missions/[A-Za-z0-9][A-Za-z0-9._:-]{2,179}", path)
and body is None and key is None):
return 200, self.service.get_mission(
path.split("/")[2], actor_identity=self.identity,
)
if (method == "POST" and isinstance(path, str)
and re.fullmatch(r"/tasks/[A-Za-z0-9][A-Za-z0-9._:-]{2,179}/delegate", path)):
if not isinstance(body, dict) or not isinstance(key, str) or not key.strip():
raise ValidationError("child body and idempotency key required")
return 201, self.service.delegate_mission(
path.split("/")[2], body, key, actor_identity=self.identity,
)
if (method == "POST" and isinstance(path, str)
and re.fullmatch(
r"/missions/[A-Za-z0-9][A-Za-z0-9._:-]{2,179}/legacy-attachments",
path,
)):
if not isinstance(body, dict) or not isinstance(key, str) or not key.strip():
raise ValidationError(
"legacy attachment body and idempotency key required"
)
result = self.service.attach_legacy_attachment(
path.split("/")[2], body, key, actor_identity=self.identity,
)
return (200 if result.get("idempotent") is True else 201), result
if (method == "GET" and isinstance(path, str)
and re.fullmatch(
r"/missions/[A-Za-z0-9][A-Za-z0-9._:-]{2,179}/"
r"legacy-attachments/[A-Za-z0-9][A-Za-z0-9._:-]{2,179}",
path,
) and body is None and key is None):
pieces = path.split("/")
return 200, self.service.get_legacy_attachment(
pieces[2], pieces[4], actor_identity=self.identity,
)
if (method == "POST" and isinstance(path, str)
and re.fullmatch(
r"/tasks/[A-Za-z0-9][A-Za-z0-9._:-]{2,179}/reconcile-terminal-proof",
path,
)):
if body != {} or not isinstance(key, str) or not key.strip():
raise ValidationError(
"terminal proof trigger requires an empty body and idempotency key"
)
if self.terminal_proof_importer is None:
raise ForbiddenError("terminal proof importer is not configured")
return 200, self.terminal_proof_importer.reconcile(
path.split("/")[2], key, actor_identity=self.identity,
)
raise ForbiddenError("operation outside mission ingress scope")
def load_registered_verifier(path: Path):
metadata = path.lstat()
if (not stat.S_ISREG(metadata.st_mode) or metadata.st_uid != 0
or metadata.st_mode & 0o022):
raise ValueError("Owner verifier must be a root-owned immutable regular file")
spec = importlib.util.spec_from_file_location("aios_mission_owner_verifier", path)
if spec is None or spec.loader is None:
raise ValueError("Owner verifier module unavailable")
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module.GrantRootOwnerApprovalVerifier()
def make_handler(ingress: SignedMissionIngress):
class Handler(BaseHTTPRequestHandler):
def log_message(self, *_args):
pass # Do not log signed payloads or transport credentials.
def respond(self, status, value):
raw = canonical_bytes(value)
self.send_response(status)
self.send_header("Content-Type", "application/json")
self.send_header("Cache-Control", "no-store")
self.send_header("Content-Length", str(len(raw)))
self.end_headers()
self.wfile.write(raw)
def do_POST(self):
try:
if self.path != "/v1/signed-request":
raise ForbiddenError("unsupported mission ingress route")
if self.headers.get("Transfer-Encoding"):
raise ValidationError("transfer encoding is unsupported")
lengths = self.headers.get_all("Content-Length", [])
if len(lengths) != 1 or not lengths[0].isdigit():
raise ValidationError("one content length required")
length = int(lengths[0])
if not 0 < length <= MAX_BODY:
raise ValidationError("mission request too large or empty")
self.connection.settimeout(10)
raw = self.rfile.read(length)
if len(raw) != length:
raise ValidationError("truncated mission request")
envelope = json.loads(raw, object_pairs_hook=strict_object)
status, result = ingress.handle(envelope)
self.respond(status, result)
except TaskPlaneHTTPError as exc:
self.respond(exc.status_code, exc.to_payload())
except (ValueError, UnicodeError, RecursionError):
self.respond(400, {"error": "invalid_json"})
except Exception:
self.respond(500, {"error": "mission_authority_unavailable"})
return Handler
def main() -> None:
from .task_plane.api import TaskPlaneService
from .task_plane.mission import MissionAuthority, MissionPrincipalBinding
from .task_plane.policy import IdentityVerifier, PolicyEngine
from .task_plane.store import TaskPlaneStore
parser = argparse.ArgumentParser()
parser.add_argument("--db", type=Path, required=True)
parser.add_argument("--owner-verifier", type=Path, required=True)
parser.add_argument("--port", type=int, required=True)
args = parser.parse_args()
if not 1024 <= args.port <= 65535:
parser.error("unprivileged loopback port required")
verifier = load_registered_verifier(args.owner_verifier)
identity = verifier.signer_id
store = TaskPlaneStore(str(args.db), high_assurance=True)
authority = MissionAuthority(store, principal_bindings=[MissionPrincipalBinding(
authenticated_identity=identity, owner_principal_id=identity,
acceptance_verifier_id="identity-lineage-live-closure-v1",
allowed_task_capabilities=(
"code.read", "code.write", "test.execute",
"governed.write.file", "governed.write.branch",
),
max_risk="medium",
)])
service = TaskPlaneService(store, policy=PolicyEngine(
identity_verifier=IdentityVerifier({identity}), execution_enabled=False,
identity_capabilities={identity: {"task.creator", "task.read", "task.delegate"}},
), mission_authority=authority)
ingress = SignedMissionIngress(service, verifier)
# SQLite's existing connection and transaction lock share one request thread.
HTTPServer(("127.0.0.1", args.port), make_handler(ingress)).serve_forever()
if __name__ == "__main__":
main()
@@ -0,0 +1,511 @@
"""Authenticated, trigger-only import of authoritative terminal proof.
The caller authorizes reconciliation but supplies no proof fields. Proof is
read by a configured server-side provider and is accepted only after the
provider validates an authenticated authority document and its immutable
cross-system tuple.
"""
from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime, timezone
from hashlib import sha256
import json
import re
from typing import Any, Callable, Mapping, Protocol
from .aios_task_lineage_integration import LINEAGE_METADATA_KEY
from .task_plane.errors import ConflictError, ForbiddenError, ValidationError
from .task_plane.models import ResultEnvelope, TaskRecord
from .task_plane.policy import PolicyDecision
PROOF_SCHEMA = "aios.task-terminal-authority-proof.v1"
RECEIPT_SCHEMA = "aios.task-terminal-import-receipt.v1"
REQUIRED_SOURCES = frozenset(
{
"agent_task",
"admission",
"job_producer",
"execution_alias",
"governed_job",
"bridge",
"authority_git",
}
)
DOCUMENT_FIELDS = frozenset(
{
"schema",
"task_id",
"mission_id",
"root_task_id",
"parent_task_id",
"acceptance_condition",
"agent_task_status",
"admission_status",
"execution_outcome",
"external_job_id",
"internal_job_id",
"repository",
"base_revision",
"target_branch",
"generation",
"execution_plan_digest",
"task_principal_ref",
"task_lease_ref",
"writer_lease_ref",
"governed_commit_oid",
"authority_ref_oid",
"postimage_hashes",
"source_receipt_digests",
"attestation_ref",
"started_at",
"completed_at",
}
)
_HEX64 = re.compile(r"[0-9a-f]{64}")
_GIT_OID = re.compile(r"[0-9a-f]{40}(?:[0-9a-f]{24})?")
_IDENTIFIER = re.compile(r"[A-Za-z0-9][A-Za-z0-9._:/-]{2,255}")
def _canonical_bytes(value: Mapping[str, object]) -> bytes:
return json.dumps(
dict(value), ensure_ascii=False, sort_keys=True, separators=(",", ":")
).encode("utf-8")
def _nonempty_string(value: object, field: str) -> str:
if not isinstance(value, str) or not value or len(value) > 1024:
raise ForbiddenError(f"authoritative proof {field} invalid")
return value
def _identifier(value: object, field: str) -> str:
text = _nonempty_string(value, field)
if _IDENTIFIER.fullmatch(text) is None:
raise ForbiddenError(f"authoritative proof {field} invalid")
return text
def _utc_time(value: object, field: str) -> str:
text = _nonempty_string(value, field)
try:
parsed = datetime.fromisoformat(text.replace("Z", "+00:00"))
except ValueError as exc:
raise ForbiddenError(f"authoritative proof {field} invalid") from exc
if parsed.tzinfo is None or parsed.utcoffset() is None:
raise ForbiddenError(f"authoritative proof {field} lacks timezone")
return parsed.astimezone(timezone.utc).isoformat().replace("+00:00", "Z")
@dataclass(frozen=True)
class VerifiedTerminalProof:
document: dict[str, object]
@property
def task_id(self) -> str:
return str(self.document["task_id"])
@property
def attestation_ref(self) -> str:
return str(self.document["attestation_ref"])
@property
def started_at(self) -> str:
return str(self.document["started_at"])
@property
def completed_at(self) -> str:
return str(self.document["completed_at"])
def evidence(self) -> dict[str, object]:
return {
"kind": "acceptance-proof",
"proof_schema": PROOF_SCHEMA,
"authority_document_digest": sha256(
_canonical_bytes(self.document)
).hexdigest(),
"task_id": self.task_id,
"mission_id": self.document["mission_id"],
"root_task_id": self.document["root_task_id"],
"parent_task_id": self.document["parent_task_id"],
"acceptance_condition": self.document["acceptance_condition"],
"external_job_id": self.document["external_job_id"],
"internal_job_id": self.document["internal_job_id"],
"repository": self.document["repository"],
"base_revision": self.document["base_revision"],
"target_branch": self.document["target_branch"],
"generation": self.document["generation"],
"execution_plan_digest": self.document["execution_plan_digest"],
"task_principal_ref": self.document["task_principal_ref"],
"task_lease_ref": self.document["task_lease_ref"],
"writer_lease_ref": self.document["writer_lease_ref"],
"governed_commit_oid": self.document["governed_commit_oid"],
"authority_ref_oid": self.document["authority_ref_oid"],
"postimage_hashes": self.document["postimage_hashes"],
"source_receipt_digests": self.document["source_receipt_digests"],
"attestation_ref": self.attestation_ref,
}
def outputs(self) -> list[dict[str, object]]:
return [
{
"kind": "governed-commit-readback",
"repository": self.document["repository"],
"target_branch": self.document["target_branch"],
"governed_commit_oid": self.document["governed_commit_oid"],
"postimage_hashes": self.document["postimage_hashes"],
}
]
class AuthorityDocumentReader(Protocol):
def __call__(self, task_id: str) -> Mapping[str, object]: ...
class AuthorityDocumentVerifier(Protocol):
def __call__(self, document: Mapping[str, object]) -> bool: ...
class AuthenticatedTerminalProofProvider:
"""Read and validate one aggregate document backed by existing authorities.
``reader`` and ``verifier`` are mandatory runtime bindings. The verifier
must authenticate the referenced AgentTask/admission/Job/Bridge/Git
receipts using already configured trust roots. This class does not confer
trust on an arbitrary mapping.
"""
def __init__(
self,
*,
reader: AuthorityDocumentReader,
verifier: AuthorityDocumentVerifier,
task_reader: Callable[[str], TaskRecord],
worker_identity: str,
) -> None:
if not callable(reader) or not callable(verifier) or not callable(task_reader):
raise ValueError("authenticated proof reader, verifier and task reader required")
self._reader = reader
self._verifier = verifier
self._task_reader = task_reader
self.worker_identity = _identifier(worker_identity, "worker_identity")
@staticmethod
def _parse(raw: Mapping[str, object]) -> dict[str, object]:
if not isinstance(raw, Mapping) or set(raw) != DOCUMENT_FIELDS:
raise ForbiddenError("authoritative proof document shape invalid")
data = dict(raw)
if data["schema"] != PROOF_SCHEMA:
raise ForbiddenError("authoritative proof schema invalid")
for field in (
"task_id", "mission_id", "root_task_id", "parent_task_id",
"external_job_id", "internal_job_id", "task_principal_ref", "task_lease_ref",
"writer_lease_ref",
):
data[field] = _identifier(data[field], field)
for field in ("acceptance_condition", "repository", "base_revision", "target_branch"):
data[field] = _nonempty_string(data[field], field)
if data["agent_task_status"] != "SUCCEEDED":
raise ForbiddenError("AgentTask is not authoritatively successful")
if data["admission_status"] != "ADMITTED":
raise ForbiddenError("task admission is not authoritative")
if data["execution_outcome"] != "APPLIED_VERIFIED":
raise ForbiddenError("governed execution outcome is not verified")
if type(data["generation"]) is not int or int(data["generation"]) < 1:
raise ForbiddenError("authoritative proof generation invalid")
if not isinstance(data["execution_plan_digest"], str) or _HEX64.fullmatch(
data["execution_plan_digest"]
) is None:
raise ForbiddenError("authoritative execution plan digest invalid")
for field in ("governed_commit_oid", "authority_ref_oid"):
if not isinstance(data[field], str) or _GIT_OID.fullmatch(data[field]) is None:
raise ForbiddenError(f"authoritative proof {field} invalid")
if data["governed_commit_oid"] != data["authority_ref_oid"]:
raise ForbiddenError("authority ref does not read back governed commit")
images = data["postimage_hashes"]
if (
not isinstance(images, Mapping)
or not images
or len(images) > 256
or any(
not isinstance(path, str)
or not path
or path.startswith(("/", "\\"))
or ".." in path.replace("\\", "/").split("/")
or not isinstance(digest, str)
or _HEX64.fullmatch(digest) is None
for path, digest in images.items()
)
):
raise ForbiddenError("authoritative postimage hashes invalid")
data["postimage_hashes"] = dict(sorted(images.items()))
sources = data["source_receipt_digests"]
if (
not isinstance(sources, Mapping)
or set(sources) != REQUIRED_SOURCES
or any(
not isinstance(digest, str) or _HEX64.fullmatch(digest) is None
for digest in sources.values()
)
):
raise ForbiddenError("authoritative source receipt set invalid")
data["source_receipt_digests"] = dict(sorted(sources.items()))
data["attestation_ref"] = _nonempty_string(data["attestation_ref"], "attestation_ref")
data["started_at"] = _utc_time(data["started_at"], "started_at")
data["completed_at"] = _utc_time(data["completed_at"], "completed_at")
if datetime.fromisoformat(data["completed_at"].replace("Z", "+00:00")) < datetime.fromisoformat(
data["started_at"].replace("Z", "+00:00")
):
raise ForbiddenError("authoritative proof time order invalid")
return data
@staticmethod
def _assert_task_binding(task: TaskRecord, data: Mapping[str, object]) -> None:
lineage = task.task_plan.metadata.get(LINEAGE_METADATA_KEY)
if not isinstance(lineage, Mapping) or lineage.get("relation_type") == "ROOT":
raise ForbiddenError("terminal proof requires a canonical lineage child")
expected = {
"task_id": task.task_id,
"mission_id": task.mission_id,
"root_task_id": task.root_task_id,
"parent_task_id": task.task_plan.parent_task_id,
"acceptance_condition": task.task_plan.metadata.get("acceptance_condition"),
}
if any(data.get(field) != value or value in (None, "") for field, value in expected.items()):
raise ForbiddenError("authoritative proof TaskPlane binding mismatch")
if any(lineage.get(field) != expected[field] for field in expected if field != "acceptance_condition"):
raise ForbiddenError("authoritative proof lineage binding mismatch")
def collect(self, task: TaskRecord) -> VerifiedTerminalProof:
raw = self._reader(task.task_id)
if not isinstance(raw, Mapping) or self._verifier(raw) is not True:
raise ForbiddenError("authoritative terminal proof authentication failed")
data = self._parse(raw)
self._assert_task_binding(task, data)
return VerifiedTerminalProof(data)
def verify_evidence(self, task: TaskRecord, evidences: list[dict]) -> bool:
proof = self.collect(task)
return evidences == [proof.evidence()]
def verify_result(self, result: ResultEnvelope) -> bool:
if result.worker_identity != self.worker_identity or result.status != "COMPLETED":
return False
try:
task = self._task_reader(result.task_id)
proof = self.collect(task)
except Exception:
return False
return bool(
result.attestation_ref == proof.attestation_ref
and result.evidence == [proof.evidence()]
and result.outputs == proof.outputs()
and result.started_at == proof.started_at
and result.completed_at == proof.completed_at
and result.operation_digest == task.task_plan.operation_digest()
)
class TerminalProofEvidenceVerifier:
"""PolicyEngine evidence verifier backed by authenticated authority reads."""
def __init__(self, provider: AuthenticatedTerminalProofProvider) -> None:
self.provider = provider
def verify(self, task: TaskRecord, evidences: list[dict]) -> PolicyDecision:
try:
allowed = self.provider.verify_evidence(task, evidences)
except Exception:
allowed = False
return PolicyDecision(allowed, "" if allowed else "terminal proof evidence invalid")
class TerminalProofImporter:
"""Reconcile a server-verified proof through TaskPlane claim/result paths."""
def __init__(
self,
service: Any,
*,
provider: AuthenticatedTerminalProofProvider | None,
worker_id: str | None,
worker_identity: str | None,
lease_seconds: int = 300,
) -> None:
self.service = service
self.provider = provider
self.worker_id = worker_id
self.worker_identity = worker_identity
if provider is not None:
if not worker_id or not worker_identity:
raise ValueError("terminal proof worker binding required")
_identifier(worker_id, "worker_id")
_identifier(worker_identity, "worker_identity")
if provider.worker_identity != worker_identity:
raise ValueError("terminal proof provider identity mismatch")
if not 30 <= int(lease_seconds) <= 3600:
raise ValueError("terminal proof lease out of range")
self.lease_seconds = int(lease_seconds)
@staticmethod
def expected_idempotency_key(task: TaskRecord) -> str:
return f"terminal-proof:{task.mission_id}:{task.task_id}"
def _authorized_task(self, task_id: str, actor_identity: str) -> TaskRecord:
self.service.get_task(task_id, actor_identity)
task = self.service.store.get_task(task_id)
lineage = task.task_plan.metadata.get(LINEAGE_METADATA_KEY)
if (
not isinstance(lineage, Mapping)
or lineage.get("relation_type") == "ROOT"
or lineage.get("task_id") != task.task_id
or lineage.get("mission_id") != task.mission_id
or lineage.get("root_task_id") != task.root_task_id
or lineage.get("parent_task_id") != task.task_plan.parent_task_id
):
raise ForbiddenError("terminal proof target is not a canonical child")
return task
def _existing_receipt(
self, task: TaskRecord, actor_identity: str
) -> dict[str, object] | None:
if self.service.lineage is None:
raise ForbiddenError("TaskPlane lineage is not configured")
snapshot = self.service.lineage.binding.get_task_snapshot(task.task_id)
if not snapshot.terminal_evidence_proven:
return None
if snapshot.status != "COMPLETED":
raise ConflictError("terminal proof target did not complete successfully")
review = self.service.lineage.binding.get_root_review_for_trigger(task.task_id)
if review is None:
raise ConflictError("terminal proof lacks durable root review")
evidence_rows = self.service.store.list_evidence(task.task_id)
by_id = {str(row.get("evidence_id")): row for row in evidence_rows}
if not snapshot.evidence_refs or any(ref not in by_id for ref in snapshot.evidence_refs):
raise ConflictError("terminal proof evidence readback mismatch")
envelopes = []
for ref in snapshot.evidence_refs:
try:
envelope = json.loads(str(by_id[ref]["envelope_json"]))
except (KeyError, TypeError, ValueError, json.JSONDecodeError) as exc:
raise ConflictError("terminal proof evidence readback invalid") from exc
envelopes.append(envelope)
claim_ids = {item.get("claim_id") for item in envelopes}
evidence_payloads = [item.get("payload") for item in envelopes]
if (
len(claim_ids) != 1
or len(evidence_payloads) != 1
or any(item.get("worker_identity") != self.worker_identity for item in envelopes)
or envelopes[0].get("kind") != "acceptance-proof"
or not isinstance(evidence_payloads[0], dict)
or not self.provider.verify_evidence(task, evidence_payloads)
or not self.service.store.verify_evidence_integrity(task.task_id)
or not self.service.store.verify_event_chain(task.task_id)
):
raise ConflictError("terminal proof receipt integrity failed")
claim = self.service.store.get_claim(str(next(iter(claim_ids))))
if (
claim.task_id != task.task_id
or claim.worker_id != self.worker_id
or claim.worker_identity != self.worker_identity
or claim.operation_digest != task.task_plan.operation_digest()
or claim.status != "COMPLETED"
):
raise ConflictError("terminal proof claim binding mismatch")
dispatch = self.service.lineage.binding.get_root_action_dispatch(str(review["review_id"]))
if dispatch is None:
raise ConflictError("terminal proof lacks durable root action dispatch")
body: dict[str, object] = {
"schema": RECEIPT_SCHEMA,
"task_id": task.task_id,
"mission_id": task.mission_id,
"root_task_id": task.root_task_id,
"claim_id": next(iter(claim_ids)),
"task_status": snapshot.status,
"task_revision": snapshot.revision,
"evidence_refs": list(snapshot.evidence_refs),
"review_id": review["review_id"],
"root_revision_after": review["root_revision_after"],
"next_legal_action": review["next_legal_action"],
"dispatch_status": dispatch.status,
"dispatch_idempotency_key": dispatch.idempotency_key,
}
body["receipt_digest"] = sha256(_canonical_bytes(body)).hexdigest()
return body
def _exact_active_claim(self, task: TaskRecord):
active = self.service.store.active_claim(task.task_id)
if active is None:
return None
if (
active.worker_id != self.worker_id
or active.worker_identity != self.worker_identity
or active.operation_digest != task.task_plan.operation_digest()
):
raise ConflictError("active claim is owned by another execution binding")
return active
def reconcile(
self, task_id: str, idempotency_key: str | None, *, actor_identity: str
) -> dict[str, object]:
if self.provider is None:
raise ForbiddenError("terminal proof provider is not configured")
task = self._authorized_task(task_id, actor_identity)
if idempotency_key != self.expected_idempotency_key(task):
raise ValidationError("terminal proof idempotency key mismatch")
existing = self._existing_receipt(task, actor_identity)
if existing is not None:
return existing
proof = self.provider.collect(task)
claim = self._exact_active_claim(task)
if claim is None:
try:
claimed = self.service.claim(
task.task_id,
{
"worker_id": self.worker_id,
"lease_seconds": self.lease_seconds,
"actor_identity": self.worker_identity,
},
)
claim_id = str(claimed["claim_id"])
except Exception:
refreshed = self._authorized_task(task_id, actor_identity)
replay = self._existing_receipt(refreshed, actor_identity)
if replay is not None:
return replay
claim = self._exact_active_claim(refreshed)
if claim is None:
raise
claim_id = claim.claim_id
else:
claim_id = claim.claim_id
try:
self.service.complete(
task.task_id,
{
"claim_id": claim_id,
"actor_identity": self.worker_identity,
"status": "COMPLETED",
"outputs": proof.outputs(),
"evidence": [proof.evidence()],
"started_at": proof.started_at,
"completed_at": proof.completed_at,
"attestation_ref": proof.attestation_ref,
},
)
except Exception:
refreshed = self._authorized_task(task_id, actor_identity)
replay = self._existing_receipt(refreshed, actor_identity)
if replay is None:
raise
return replay
refreshed = self._authorized_task(task_id, actor_identity)
receipt = self._existing_receipt(refreshed, actor_identity)
if receipt is None:
raise ConflictError("terminal proof completion lacks authoritative readback")
return receipt
+40 -8
View File
@@ -4,6 +4,7 @@ import hashlib, html, json, os, sqlite3, time, uuid, urllib.parse
from http import cookies
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from pathlib import Path
import hmac # SEC-P0-BOOTSTRAP-CREDENTIAL
ROOT = Path(os.getenv('AIOS_WEBOPS_ROOT', Path(__file__).resolve().parents[1]))
DB_PATH = Path(os.getenv('AIOS_WEBOPS_DB', ROOT / 'data' / 'webops.sqlite3'))
@@ -13,7 +14,11 @@ PUBLIC_DOMAIN = os.getenv('AIOS_PUBLIC_DOMAIN','https://cnnetmkt.top')
WEBOPS_DOMAIN = os.getenv('AIOS_WEBOPS_DOMAIN','https://ops.cnnetmkt.top')
API_DOMAIN = os.getenv('AIOS_API_DOMAIN','https://api.cnnetmkt.top')
WECHAT_DOMAIN = os.getenv('AIOS_WECHAT_CALLBACK_DOMAIN','https://wechat.cnnetmkt.top')
SECRET = os.getenv('AIOS_WEBOPS_SECRET','change-me-before-production')
SECRET = os.getenv('AIOS_WEBOPS_SECRET') or ''
# SEC-P0: fail closed — never boot with a missing/insecure default secret.
if SECRET in ('', 'change-me-before-production'):
raise SystemExit('AIOS_WEBOPS_SECRET not set (or insecure default). Refusing to start: fail closed.')
SESSION_TTL = int(os.getenv('AIOS_SESSION_TTL','3600'))
ROLES = ['owner','admin','user']
NODES = ['server-hw','server-local']
PAGES = [('/dashboard','Dashboard'),('/nodes','Nodes'),('/roles','Roles'),('/routing','Routing'),('/failover','Failover'),('/approvals','Approvals'),('/operations','Operations'),('/users','Users'),('/quota','Quota'),('/models','Models'),('/memory','Memory'),('/channels/wechat','WeChat'),('/life','Life Core'),('/patches','Patch Studio'),('/optimization','Optimization'),('/simulation','Simulation'),('/logs','Logs'),('/config','Config'),('/uat','UAT'),('/versions','Versions'),('/evidence','Evidence')]
@@ -43,19 +48,46 @@ def audit(actor, action, target='', payload=None):
exec_sql('INSERT INTO audit_logs(id,actor,action,target,payload,created_at) VALUES(?,?,?,?,?,?)',(uid('audit'),actor,action,target,jdump(payload or {}),now()))
def notify(event_type,payload):
exec_sql('INSERT INTO owner_notifications(id,event_type,payload,status,created_at) VALUES(?,?,?,?,?)',(uid('notify'),event_type,jdump(payload),'queued',now()))
def hash_pw(p): return hashlib.sha256((SECRET+'|'+p).encode()).hexdigest()
def check_pw(stored,p): return stored in (p, hash_pw(p))
_PBKDF2_ITERS = int(os.getenv('AIOS_PBKDF2_ITERS','260000'))
def _is_pbkdf2(stored): return isinstance(stored,str) and stored.startswith('pbkdf2$') and stored.count('$')==2
def hash_pw(p):
salt=os.urandom(16).hex()
dk=hashlib.pbkdf2_hmac('sha256', p.encode(), bytes.fromhex(salt), _PBKDF2_ITERS).hex()
return f'pbkdf2${salt}${dk}'
def check_pw(stored,p):
if not stored or not isinstance(stored,str) or not stored: return False
if _is_pbkdf2(stored):
_,salt,dk=stored.split('$')
calc=hashlib.pbkdf2_hmac('sha256', p.encode(), bytes.fromhex(salt), _PBKDF2_ITERS).hex()
return hmac.compare_digest(dk,calc)
# Legacy record path: NEVER accept the plaintext form as a credential.
# A legacy (unsalted) sha256 is honored ONLY as a one-time owner migration
# credential and never for the known bootstrap default; anything else is
# rejected so the account must be re-seeded via a secure channel (fail closed).
if stored == p: return False
try:
old=hashlib.sha256((SECRET+'|'+p).encode()).hexdigest()
except Exception:
return False
return (hmac.compare_digest(old,stored)) and (p not in {'bootstrap-change-me'})
def token(username,role):
data=f'{username}|{role}|{int(time.time())}'; sig=hashlib.sha256((SECRET+'|'+data).encode()).hexdigest()[:16]; return data+'|'+sig
exp=int(time.time())+SESSION_TTL
data=f'{username}|{role}|{exp}'
sig=hmac.new(SECRET.encode(), data.encode(), hashlib.sha256).hexdigest()
return data+'|'+sig
def parse_session(header):
if not header: return None
c=cookies.SimpleCookie(); c.load(header)
if 'aios_session' not in c: return None
parts=c['aios_session'].value.split('|')
if len(parts)!=4: return None
data='|'.join(parts[:3]); sig=parts[3]
if hashlib.sha256((SECRET+'|'+data).encode()).hexdigest()[:16]!=sig: return None
return one('SELECT * FROM users WHERE username=? AND role=? AND status=?',(parts[0],parts[1],'approved'))
username,role,exp,sig=parts
data=f'{username}|{role}|{exp}'
if not hmac.compare_digest(hmac.new(SECRET.encode(), data.encode(), hashlib.sha256).hexdigest(), sig): return None
try:
if int(exp) <= int(time.time()): return None
except Exception: return None
return one('SELECT * FROM users WHERE username=? AND role=? AND status=?',(username,role,'approved'))
def require(user, roles): return bool(user and user['role'] in roles)
def table(data):
if not data: return '<p class="muted">No records.</p>'
@@ -215,7 +247,7 @@ class H(BaseHTTPRequestHandler):
if path=='/login':
n=int(self.headers.get('Content-Length','0') or 0); data=urllib.parse.parse_qs(self.rfile.read(n).decode())
u=one('SELECT * FROM users WHERE username=? AND status=?',(data.get('username',[''])[0],'approved'))
if u and check_pw(u.get('password_hash') or '',data.get('password',[''])[0]): audit(u['username'],'login','webops',{}); self.redirect('/dashboard',[f"aios_session={token(u['username'],u['role'])}; Path=/; HttpOnly; SameSite=Lax"]); return
if u and check_pw(u.get('password_hash') or '',data.get('password',[''])[0]): audit(u['username'],'login','webops',{}); self.redirect('/dashboard',[f"aios_session={token(u['username'],u['role'])}; Path=/; HttpOnly; SameSite=Strict"]); return
self.send_body(403,layout('Login Failed','<p>登录失败。</p>',None)); return
if path=='/register':
n=int(self.headers.get('Content-Length','0') or 0); data=urllib.parse.parse_qs(self.rfile.read(n).decode())
+40 -8
View File
@@ -4,6 +4,7 @@ import hashlib, html, json, os, sqlite3, time, uuid, urllib.parse
from http import cookies
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from pathlib import Path
import hmac # SEC-P0-BOOTSTRAP-CREDENTIAL
ROOT = Path(os.getenv('AIOS_WEBOPS_ROOT', Path(__file__).resolve().parents[1]))
DB_PATH = Path(os.getenv('AIOS_WEBOPS_DB', ROOT / 'data' / 'webops.sqlite3'))
@@ -13,7 +14,11 @@ PUBLIC_DOMAIN = os.getenv('AIOS_PUBLIC_DOMAIN','https://cnnetmkt.top')
WEBOPS_DOMAIN = os.getenv('AIOS_WEBOPS_DOMAIN','https://ops.cnnetmkt.top')
API_DOMAIN = os.getenv('AIOS_API_DOMAIN','https://api.cnnetmkt.top')
WECHAT_DOMAIN = os.getenv('AIOS_WECHAT_CALLBACK_DOMAIN','https://wechat.cnnetmkt.top')
SECRET = os.getenv('AIOS_WEBOPS_SECRET','change-me-before-production')
SECRET = os.getenv('AIOS_WEBOPS_SECRET') or ''
# SEC-P0: fail closed — never boot with a missing/insecure default secret.
if SECRET in ('', 'change-me-before-production'):
raise SystemExit('AIOS_WEBOPS_SECRET not set (or insecure default). Refusing to start: fail closed.')
SESSION_TTL = int(os.getenv('AIOS_SESSION_TTL','3600'))
ROLES = ['owner','admin','user']
NODES = ['server-hw','server-local']
PAGES = [('/dashboard','Dashboard'),('/nodes','Nodes'),('/roles','Roles'),('/routing','Routing'),('/failover','Failover'),('/approvals','Approvals'),('/operations','Operations'),('/users','Users'),('/quota','Quota'),('/models','Models'),('/memory','Memory'),('/channels/wechat','WeChat'),('/life','Life Core'),('/patches','Patch Studio'),('/optimization','Optimization'),('/simulation','Simulation'),('/logs','Logs'),('/config','Config'),('/uat','UAT'),('/versions','Versions'),('/evidence','Evidence')]
@@ -43,19 +48,46 @@ def audit(actor, action, target='', payload=None):
exec_sql('INSERT INTO audit_logs(id,actor,action,target,payload,created_at) VALUES(?,?,?,?,?,?)',(uid('audit'),actor,action,target,jdump(payload or {}),now()))
def notify(event_type,payload):
exec_sql('INSERT INTO owner_notifications(id,event_type,payload,status,created_at) VALUES(?,?,?,?,?)',(uid('notify'),event_type,jdump(payload),'queued',now()))
def hash_pw(p): return hashlib.sha256((SECRET+'|'+p).encode()).hexdigest()
def check_pw(stored,p): return stored in (p, hash_pw(p))
_PBKDF2_ITERS = int(os.getenv('AIOS_PBKDF2_ITERS','260000'))
def _is_pbkdf2(stored): return isinstance(stored,str) and stored.startswith('pbkdf2$') and stored.count('$')==2
def hash_pw(p):
salt=os.urandom(16).hex()
dk=hashlib.pbkdf2_hmac('sha256', p.encode(), bytes.fromhex(salt), _PBKDF2_ITERS).hex()
return f'pbkdf2${salt}${dk}'
def check_pw(stored,p):
if not stored or not isinstance(stored,str) or not stored: return False
if _is_pbkdf2(stored):
_,salt,dk=stored.split('$')
calc=hashlib.pbkdf2_hmac('sha256', p.encode(), bytes.fromhex(salt), _PBKDF2_ITERS).hex()
return hmac.compare_digest(dk,calc)
# Legacy record path: NEVER accept the plaintext form as a credential.
# A legacy (unsalted) sha256 is honored ONLY as a one-time owner migration
# credential and never for the known bootstrap default; anything else is
# rejected so the account must be re-seeded via a secure channel (fail closed).
if stored == p: return False
try:
old=hashlib.sha256((SECRET+'|'+p).encode()).hexdigest()
except Exception:
return False
return (hmac.compare_digest(old,stored)) and (p not in {'bootstrap-change-me'})
def token(username,role):
data=f'{username}|{role}|{int(time.time())}'; sig=hashlib.sha256((SECRET+'|'+data).encode()).hexdigest()[:16]; return data+'|'+sig
exp=int(time.time())+SESSION_TTL
data=f'{username}|{role}|{exp}'
sig=hmac.new(SECRET.encode(), data.encode(), hashlib.sha256).hexdigest()
return data+'|'+sig
def parse_session(header):
if not header: return None
c=cookies.SimpleCookie(); c.load(header)
if 'aios_session' not in c: return None
parts=c['aios_session'].value.split('|')
if len(parts)!=4: return None
data='|'.join(parts[:3]); sig=parts[3]
if hashlib.sha256((SECRET+'|'+data).encode()).hexdigest()[:16]!=sig: return None
return one('SELECT * FROM users WHERE username=? AND role=? AND status=?',(parts[0],parts[1],'approved'))
username,role,exp,sig=parts
data=f'{username}|{role}|{exp}'
if not hmac.compare_digest(hmac.new(SECRET.encode(), data.encode(), hashlib.sha256).hexdigest(), sig): return None
try:
if int(exp) <= int(time.time()): return None
except Exception: return None
return one('SELECT * FROM users WHERE username=? AND role=? AND status=?',(username,role,'approved'))
def require(user, roles): return bool(user and user['role'] in roles)
def table(data):
if not data: return '<p class="muted">No records.</p>'
@@ -215,7 +247,7 @@ class H(BaseHTTPRequestHandler):
if path=='/login':
n=int(self.headers.get('Content-Length','0') or 0); data=urllib.parse.parse_qs(self.rfile.read(n).decode())
u=one('SELECT * FROM users WHERE username=? AND status=?',(data.get('username',[''])[0],'approved'))
if u and check_pw(u.get('password_hash') or '',data.get('password',[''])[0]): audit(u['username'],'login','webops',{}); self.redirect('/dashboard',[f"aios_session={token(u['username'],u['role'])}; Path=/; HttpOnly; SameSite=Lax"]); return
if u and check_pw(u.get('password_hash') or '',data.get('password',[''])[0]): audit(u['username'],'login','webops',{}); self.redirect('/dashboard',[f"aios_session={token(u['username'],u['role'])}; Path=/; HttpOnly; SameSite=Strict"]); return
self.send_body(403,layout('Login Failed','<p>登录失败。</p>',None)); return
if path=='/register':
n=int(self.headers.get('Content-Length','0') or 0); data=urllib.parse.parse_qs(self.rfile.read(n).decode())
@@ -314,27 +314,6 @@ def _tool_argv(
return [str(path), *args]
def _provisioner_env_argv(
config: ProvisioningConfig,
root_config: Path,
path: Path,
*args: str,
) -> list[str]:
"""Build the provisioner argv with an explicit AIOS_GRANT_ROOT_CONFIG so it
survives sudo `Defaults env_reset` in production sudoers (same systemic fix
as the validator). Direct (non-sudo) runs keep the original form."""
if config.governed_use_sudo:
return [
str(config.sudo_path),
"-n",
"env",
f"AIOS_GRANT_ROOT_CONFIG={root_config}",
str(path),
*args,
]
return [str(path), *args]
def _verify_tool_contract(
path: Path,
name: str,
@@ -840,7 +819,7 @@ class AuthorizedTaskWriterProvisioner:
raise ProvisioningError(
"SIGNER_COMPATIBILITY_FAILED",
503,
details={"error": type(exc).__name__, "message": str(exc)},
details={"error": type(exc).__name__},
) from exc
signer_result = _read_json(signer_result_path)
signer = _verify_signer_result(
@@ -858,7 +837,7 @@ class AuthorizedTaskWriterProvisioner:
raise ProvisioningError(
"VALIDATOR_COMPATIBILITY_FAILED",
503,
details={"error": type(exc).__name__, "message": str(exc)},
details={"error": type(exc).__name__},
) from exc
validator_result = _read_json(validator_result_path)
_verify_validator_result(
@@ -872,9 +851,8 @@ class AuthorizedTaskWriterProvisioner:
validator_config_path
)
_run(
_provisioner_env_argv(
_tool_argv(
self.config,
validator_config_path,
self.config.provisioner,
"--grant-dir",
str(grant_dir),
@@ -77,87 +77,6 @@ def _read(path: Path) -> dict[str, Any]:
return value
def _read_json(path: Path) -> dict[str, Any]:
"""Read a JSON file that may be root-written (0600 root:root) by the
governed signer/validator. Mirrors `_read` but goes through
`_root_file_bytes` so root-only result files are read via sudo."""
if path.is_symlink() or not path.is_file():
raise SignerContractError("contract_input_missing")
try:
raw = _root_file_bytes(path)
except SignerContractError:
raise
try:
value = json.loads(raw.decode("utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError) as exc:
raise SignerContractError("contract_input_invalid") from exc
if not isinstance(value, dict):
raise SignerContractError("contract_object_required")
return value
def _root_is_readable(path: Path) -> bool:
try:
path.read_bytes()
return True
except (OSError, PermissionError):
return False
def _read_root_config(config_path: Path) -> dict[str, Any]:
"""Read the GrantRoot config. When the API service runs with governed sudo
(signer/validator/provisioner invoke via `sudo -n --`), this config is
root-only (0600); read it through sudo the same way instead of relaxing
permissions or breaking the private-key isolation boundary."""
if (
os.getenv("AIOS_GITOPS_GOVERNED_USE_SUDO", "").strip().lower()
in {"1", "true", "yes", "on"}
and not _root_is_readable(config_path)
):
sudo = os.getenv("AIOS_GITOPS_SUDO_PATH", "/usr/bin/sudo")
result = subprocess.run(
[sudo, "-n", "--", "cat", str(config_path)],
text=True,
capture_output=True,
check=False,
timeout=15,
)
if result.returncode != 0:
raise SignerContractError(f"root_config_unreadable:{result.returncode}")
try:
value = json.loads(result.stdout)
except (json.JSONDecodeError, UnicodeDecodeError) as exc:
raise SignerContractError("root_config_invalid") from exc
if not isinstance(value, dict):
raise SignerContractError("contract_object_required")
return value
return _read(config_path)
def _root_file_bytes(path: Path) -> bytes:
"""Read a raw file that may be root-only (0600 root:root). When the API
service runs with governed sudo, signer/validator/provisioner write
signature artifacts as root; read them through sudo the same way instead of
relaxing permissions or breaking the private-key isolation boundary."""
if (
os.getenv("AIOS_GITOPS_GOVERNED_USE_SUDO", "").strip().lower()
in {"1", "true", "yes", "on"}
and not _root_is_readable(path)
):
sudo = os.getenv("AIOS_GITOPS_SUDO_PATH", "/usr/bin/sudo")
result = subprocess.run(
[sudo, "-n", "--", "cat", str(path)],
text=False,
capture_output=True,
check=False,
timeout=15,
)
if result.returncode != 0:
raise SignerContractError(f"root_file_unreadable:{result.returncode}")
return result.stdout
return path.read_bytes()
def _write(path: Path, value: Any) -> None:
if path.is_symlink():
raise SignerContractError("contract_output_symlink")
@@ -200,10 +119,8 @@ def _run(
env=dict(env) if env is not None else None,
)
if result.returncode != 0:
detail = (result.stderr or result.stdout or "").strip()[-400:]
raise SignerContractError(
f"deployed_tool_failed:{Path(argv[0]).name}:{result.returncode}"
+ (f":{detail}" if detail else "")
)
return result
@@ -245,7 +162,7 @@ def _manifest(grant_dir: Path, names: Sequence[str]) -> None:
if not path.is_file() or path.is_symlink():
raise SignerContractError("manifest_input_missing")
rows.append(
f"{hashlib.sha256(_root_file_bytes(path)).hexdigest()} {name}"
f"{hashlib.sha256(path.read_bytes()).hexdigest()} {name}"
)
(grant_dir / "MANIFEST.sha256").write_text(
"\n".join(rows) + "\n",
@@ -404,7 +321,7 @@ def run_compatible_signer(
config_path = Path(
os.environ.get("AIOS_GRANT_ROOT_CONFIG", "/etc/aios/grant-root/config.json")
)
config = _read_root_config(config_path)
config = _read(config_path)
grant_namespace = str(config["grant_namespace"])
approval_namespace = str(config["approval_namespace"])
_sign(
@@ -422,10 +339,8 @@ def run_compatible_signer(
grant_dir / "signed-grant.key-id",
)
key_identity = (
_root_file_bytes(grant_dir / "signed-grant.key-id")
.decode("utf-8")
.strip()
)
grant_dir / "signed-grant.key-id"
).read_text(encoding="utf-8").strip()
if not key_identity:
raise SignerContractError("signer_identity_missing")
_manifest(
@@ -479,9 +394,10 @@ def _validator_config(
root_config = Path(
os.environ.get("AIOS_GRANT_ROOT_CONFIG", "/etc/aios/grant-root/config.json")
)
config = _read_root_config(root_config)
config = _read(root_config)
repo_path = (
os.environ.get("AIOS_GITOPS_VALIDATOR_REPO_PATH")
or os.environ.get("AIOS_GITOPS_WORKSPACE")
or config.get("repo_path")
)
config.update(
@@ -505,42 +421,6 @@ def _validator_config(
return path
def _validator_run_argv(
validator_command: Sequence[str],
config_path: Path,
grant_dir: Path,
result_path: Path,
) -> list[str]:
"""Build the validator argv. Production sudoers `Defaults env_reset` strips
AIOS_GRANT_ROOT_CONFIG from sudo, so when the command already carries a
sudo prefix we inject it explicitly via `sudo -n env VAR=... tool ...`.
A direct (non-sudo) invocation keeps its original form."""
validator_argv = list(validator_command)
if Path(validator_argv[0]).name in {"sudo", "sudo.exe"}:
rest = validator_argv[1:]
tool_at = 0
while tool_at < len(rest) and rest[tool_at].startswith("-"):
tool_at += 1
return [
validator_argv[0],
"-n",
"env",
f"AIOS_GRANT_ROOT_CONFIG={config_path}",
*rest[tool_at:],
"--grant-dir",
str(grant_dir),
"--result-json",
str(result_path),
]
return [
*validator_command,
"--grant-dir",
str(grant_dir),
"--result-json",
str(result_path),
]
def run_compatible_validator(
validator_command: Sequence[str],
grant_dir: Path,
@@ -558,7 +438,7 @@ def run_compatible_validator(
],
timeout=60,
)
legacy = _read_json(result_path)
legacy = _read(result_path)
if legacy.get("status") not in {"PASS", "VALID", "COMPLETED"}:
raise SignerContractError("legacy_validator_not_ready")
return Path(
@@ -580,19 +460,19 @@ def run_compatible_validator(
config_path = _validator_config(grant_dir, request)
env = os.environ.copy()
env["AIOS_GRANT_ROOT_CONFIG"] = str(config_path)
# sudo `Defaults env_reset` (production sudoers) strips AIOS_GRANT_ROOT_CONFIG
# from sudo, so the validator would read the default root config and hash/link
# the wrong approval-plan. Explicitly inject via `sudo -n env VAR=... tool ...`.
run_argv = _validator_run_argv(
validator_command, config_path, grant_dir, result_path
)
try:
_run(
run_argv,
[
*validator_command,
"--grant-dir",
str(grant_dir),
"--result-json",
str(result_path),
],
env=env,
timeout=60,
)
actual = _read_json(result_path)
actual = _read(result_path)
if actual.get("result") != "READY":
raise SignerContractError("validator_not_ready")
checks = {
+167 -71
View File
@@ -105,6 +105,28 @@ pre { white-space: pre-wrap; word-break: break-word; }
</div>
</div>
<script>
// Security: all dynamic values (API/git/agent/owner-gate) are rendered with
// textContent / DOM APIs. innerHTML is never used to inject untrusted data.
function clearNode(node) {
if (!node) return;
while (node.firstChild) node.removeChild(node.firstChild);
}
function setText(id, text) {
const n = document.getElementById(id);
if (n) n.textContent = text;
}
function addClass(node, cls) { if (node) node.classList.add(cls); }
function div(cls) { const d = document.createElement('div'); if (cls) addClass(d, cls); return d; }
function span(cls, text) { const s = document.createElement('span'); if (cls) addClass(s, cls); if (text != null) s.textContent = text; return s; }
// Centralized CSRF token source: server-issued meta tag, or double-submit cookie.
function csrfToken() {
const meta = document.querySelector('meta[name="csrf-token"]');
if (meta && meta.content) return meta.content;
const m = document.cookie.match(/(?:^|;\s*)aios_csrf=([^;]+)/);
return m ? decodeURIComponent(m[1]) : '';
}
async function fetchData() {
const r = await fetch('/admin/ai-evolution/api/data');
const data = await r.json();
@@ -124,130 +146,204 @@ async function fetchData() {
async function loadVersions() {
const el = document.getElementById('versions-list');
el.innerHTML = '<div class="loading">Loading versions...</div>';
clearNode(el);
const loading = div('loading'); loading.textContent = 'Loading versions...'; el.appendChild(loading);
try {
const r = await fetch('/admin/ai-evolution/api/versions');
const data = await r.json();
if (data.error) { el.innerHTML = '<div class="error">' + data.error + '</div>'; return; }
let html = '<div class="grid">';
clearNode(el);
if (data.error) { const e = div('error'); e.textContent = data.error; el.appendChild(e); return; }
const grid = div('grid');
for (const v of data.versions || []) {
html += '<div class="card"><div class="card-title">' + v.tag + '</div><div class="card-value">' + (v.version || v.tag) + '</div></div>';
const card = div('card');
const t = div('card-title'); t.textContent = v.tag || '';
const val = div('card-value'); val.textContent = v.version || v.tag || '';
card.appendChild(t); card.appendChild(val); grid.appendChild(card);
}
html += '</div>';
el.innerHTML = html;
} catch(e) {
el.innerHTML = '<div class="error">Failed to load versions: ' + e.message + '</div>';
el.appendChild(grid);
} catch (e) {
const err = div('error'); err.textContent = 'Failed to load versions: ' + e.message; el.appendChild(err);
}
}
async function loadCommands() {
const el = document.getElementById('commands-list');
el.innerHTML = '<div class="loading">Loading commands...</div>';
clearNode(el);
const loading = div('loading'); loading.textContent = 'Loading commands...'; el.appendChild(loading);
try {
const r = await fetch('/admin/ai-evolution/api/data');
const data = await r.json();
let cmds = data.commands || [];
let html = '<div style="display:flex;gap:24px;margin-bottom:16px;">';
html += '<div><div class="card"><div class="card-title">Total</div><div class="card-value">' + cmds.length + '</div></div></div>';
html += '<div><div class="card"><div class="card-title">By Category</div>';
const cmds = data.commands || [];
clearNode(el);
const wrapper = div('');
wrapper.style.display = 'flex'; wrapper.style.gap = '24px'; wrapper.style.marginBottom = '16px';
const totalCard = div('card');
const tcTitle = div('card-title'); tcTitle.textContent = 'Total';
const tcVal = div('card-value'); tcVal.textContent = String(cmds.length);
totalCard.appendChild(tcTitle); totalCard.appendChild(tcVal); wrapper.appendChild(totalCard);
const catCard = div('card');
const ccTitle = div('card-title'); ccTitle.textContent = 'By Category';
catCard.appendChild(ccTitle);
const cats = {};
for (const c of cmds) { const k = c.category || 'uncategorized'; cats[k] = (cats[k] || 0) + 1; }
for (const [k, v] of Object.entries(cats)) html += '<div style="font-size:14px;color:#8b949e;margin:4px 0;">' + k + ': ' + v + '</div>';
html += '</div></div></div>';
for (const cmd of cmds.slice().reverse()) {
html += '<div class="command-item"><div class="meta">' + cmd.category + ' | ' + cmd.filename + '</div><div class="name">' + cmd.title + '</div><div style="font-size:13px;color:#c9d1d9;margin-top:4px;">' + cmd.summary + '</div></div>';
for (const [k, v] of Object.entries(cats)) {
const line = div(''); line.style.fontSize = '14px'; line.style.color = '#8b949e'; line.style.margin = '4px 0';
line.textContent = k + ': ' + v; catCard.appendChild(line);
}
el.innerHTML = html;
} catch(e) {
el.innerHTML = '<div class="error">Failed to load commands: ' + e.message + '</div>';
wrapper.appendChild(catCard); el.appendChild(wrapper);
for (const cmd of cmds.slice().reverse()) {
const item = div('command-item');
const meta = div('meta'); meta.textContent = (cmd.category || '') + ' | ' + (cmd.filename || '');
const name = div('name'); name.textContent = cmd.title || '';
const summary = div(''); summary.style.fontSize = '13px'; summary.style.color = '#c9d1d9'; summary.style.marginTop = '4px';
summary.textContent = cmd.summary || '';
item.appendChild(meta); item.appendChild(name); item.appendChild(summary);
el.appendChild(item);
}
} catch (e) {
const err = div('error'); err.textContent = 'Failed to load commands: ' + e.message; el.appendChild(err);
}
}
async function loadAgents() {
const el = document.getElementById('agents-list');
el.innerHTML = '<div class="loading">Loading agent runs...</div>';
clearNode(el);
const loading = div('loading'); loading.textContent = 'Loading agent runs...'; el.appendChild(loading);
try {
const r = await fetch('/admin/ai-evolution/api/agents');
const data = await r.json();
if (data.error) { el.innerHTML = '<div class="error">' + data.error + '</div>'; return; }
let html = '';
for (const agent of data.agents || []) {
html += '<div class="agent-item"><div class="agent-title">' + agent.name + ' <span class="state state-' + (agent.status === 'running' ? 'running' : (agent.status === 'ok' ? 'completed' : 'failed')) + '">' + agent.status + '</span></div><div class="agent-meta">' + (agent.runs || 0) + ' runs | last: ' + (agent.last_run || '-') + '</div></div>';
clearNode(el);
if (data.error) { const e = div('error'); e.textContent = data.error; el.appendChild(e); return; }
const agents = data.agents || [];
if (!agents.length) {
const none = div(''); none.style.padding = '24px'; none.style.color = '#8b949e';
none.textContent = 'No agent runs recorded yet.'; el.appendChild(none); return;
}
el.innerHTML = html || '<div style="padding:24px;color:#8b949e;">No agent runs recorded yet.</div>';
} catch(e) {
el.innerHTML = '<div class="error">Failed to load agents: ' + e.message + '</div>';
for (const agent of agents) {
const item = div('agent-item');
const title = div('agent-title');
title.appendChild(document.createTextNode(agent.name || ''));
const st = (agent.status === 'running') ? 'running' : ((agent.status === 'ok') ? 'completed' : 'failed');
const badge = span('state state-' + st, agent.status || '');
title.appendChild(badge);
const meta = div('agent-meta');
meta.textContent = (agent.runs || 0) + ' runs | last: ' + (agent.last_run || '-');
item.appendChild(title); item.appendChild(meta); el.appendChild(item);
}
} catch (e) {
const err = div('error'); err.textContent = 'Failed to load agents: ' + e.message; el.appendChild(err);
}
}
function showPanel(name) {
async function loadOwnerGate() {
const el = document.getElementById('arch-owner-content');
el.innerHTML = '<div class="loading">Loading owner gate data...</div>';
clearNode(el);
const loading = div('loading'); loading.textContent = 'Loading owner gate data...'; el.appendChild(loading);
try {
const r = await fetch('/admin/ai-evolution/api/owner-gate');
const data = await r.json();
let html = '<div class="row">';
html += '<div class="col"><div class="card"><div class="card-title">Pending Requests</div><div class="card-value">' + data.pending_count + '</div></div></div>';
html += '<div class="col"><div class="card"><div class="card-title">Approved</div><div class="card-value">' + data.approved_count + '</div></div></div>';
html += '<div class="col"><div class="card"><div class="card-title">Rejected</div><div class="card-value">' + data.rejected_count + '</div></div></div>';
html += '</div>';
html += '<div class="card"><div class="card-title">Pending Approval Requests</div>';
if (data.pending_requests && data.pending_requests.length) {
for (const req of data.pending_requests) {
html += '<div class="command-item">';
html += '<div class="name">' + (req.title || req.change_id) + '</div>';
html += '<div class="meta">ID: ' + req.change_id + ' | Risk: ' + req.risk + '</div>';
html += '<div style="font-size:13px;color:#c9d1d9;margin-top:4px;">' + (req.reason || '') + '</div>';
html += '</div>';
clearNode(el);
const row = div('row');
const cols = [['Pending Requests', data.pending_count], ['Approved', data.approved_count], ['Rejected', data.rejected_count]];
for (const [label, value] of cols) {
const col = div('col'); const card = div('card');
const t = div('card-title'); t.textContent = label;
const val = div('card-value'); val.textContent = value;
card.appendChild(t); card.appendChild(val); col.appendChild(card); row.appendChild(col);
}
el.appendChild(row);
const listCard = div('card');
const lt = div('card-title'); lt.textContent = 'Pending Approval Requests'; listCard.appendChild(lt);
const reqs = data.pending_requests || [];
if (reqs.length) {
for (const req of reqs) {
const item = div('command-item');
const name = div('name'); name.textContent = req.title || req.change_id || '';
const meta = div('meta'); meta.textContent = 'ID: ' + (req.change_id || '') + ' | Risk: ' + (req.risk || '');
const reason = div(''); reason.style.fontSize = '13px'; reason.style.color = '#c9d1d9'; reason.style.marginTop = '4px';
reason.textContent = req.reason || '';
item.appendChild(name); item.appendChild(meta); item.appendChild(reason);
listCard.appendChild(item);
}
} else {
html += '<div style="color:#8b949e;padding:12px;">No pending requests.</div>';
const none = div(''); none.style.color = '#8b949e'; none.style.padding = '12px';
none.textContent = 'No pending requests.'; listCard.appendChild(none);
}
html += '</div>';
el.innerHTML = html;
} catch(e) {
el.innerHTML = '<div class="error">Failed to load: ' + e.message + '</div>';
el.appendChild(listCard);
} catch (e) {
const err = div('error'); err.textContent = 'Failed to load: ' + e.message; el.appendChild(err);
}
}
// Real governed approve: posts the decision to the server-side governed endpoint.
async function approveDecision(changeId, decision, btn) {
if (btn) btn.disabled = true;
try {
const res = await fetch('/admin/ai-evolution/api/owner-gate/decision', {
method: 'POST',
headers: { 'Content-Type': 'application/json', 'X-CSRF-Token': csrfToken() },
body: JSON.stringify({ change_id: changeId, decision: decision })
});
const payload = await res.json();
if (!res.ok) throw new Error(payload.error || ('HTTP ' + res.status));
await loadApprove();
} catch (e) {
if (btn) btn.disabled = false;
const out = document.getElementById('arch-approve-content');
const err = div('error'); err.textContent = 'Decision failed: ' + e.message;
if (out) out.appendChild(err);
}
}
async function loadApprove() {
const el = document.getElementById('arch-approve-content');
el.innerHTML = '<div class="loading">Loading pending approvals...</div>';
clearNode(el);
const loading = div('loading'); loading.textContent = 'Loading pending approvals...'; el.appendChild(loading);
try {
const r = await fetch('/admin/ai-evolution/api/owner-gate');
const data = await r.json();
let html = '';
if (data.pending_requests && data.pending_requests.length) {
html += '<div class="card"><div class="card-title">Pending changes requiring your approval</div>';
for (const req of data.pending_requests) {
html += '<div class="command-item">';
html += '<div class="name">' + (req.title || req.change_id) + '</div>';
html += '<div class="meta">Risk: <strong>' + req.risk + '</strong></div>';
html += '<div style="font-size:13px;color:#c9d1d9;margin:8px 0;">' + (req.reason || '') + '</div>';
html += '<div style="display:flex;gap:8px;">';
html += '<button onclick="ownerAction.classList.add('\'approved\');this.disabled=true" style="padding:6px 16px;background:#238636;color:#fff;border:none;border-radius:6px;cursor:pointer;font-size:13px;">✓ Approve</button>';
html += '</div></div>';
clearNode(el);
const reqs = data.pending_requests || [];
if (reqs.length) {
const card = div('card');
const t = div('card-title'); t.textContent = 'Pending changes requiring your approval'; card.appendChild(t);
for (const req of reqs) {
const item = div('command-item');
const name = div('name'); name.textContent = req.title || req.change_id || '';
const meta = div('meta');
const strongRisk = document.createElement('strong'); strongRisk.textContent = req.risk || '';
meta.appendChild(document.createTextNode('Risk: ')); meta.appendChild(strongRisk);
const reason = div(''); reason.style.fontSize = '13px'; reason.style.color = '#c9d1d9'; reason.style.margin = '8px 0';
reason.textContent = req.reason || '';
const actions = div(''); actions.style.display = 'flex'; actions.style.gap = '8px';
const btn = document.createElement('button');
btn.textContent = '✓ Approve';
btn.style.padding = '6px 16px'; btn.style.background = '#238636'; btn.style.color = '#fff';
btn.style.border = 'none'; btn.style.borderRadius = '6px'; btn.style.cursor = 'pointer'; btn.style.fontSize = '13px';
const changeId = req.change_id;
btn.addEventListener('click', function () { approveDecision(changeId, 'approved', btn); });
actions.appendChild(btn);
item.appendChild(name); item.appendChild(meta); item.appendChild(reason); item.appendChild(actions);
card.appendChild(item);
}
html += '</div>';
el.appendChild(card);
} else {
html += '<div class="card"><div style="padding:24px;text-align:center;color:#8b949e;">✅ All changes processed. No pending approval requests.</div></div>';
const done = div('card');
const inner = div(''); inner.style.padding = '24px'; inner.style.textAlign = 'center'; inner.style.color = '#8b949e';
inner.textContent = '✅ All changes processed. No pending approval requests.';
done.appendChild(inner); el.appendChild(done);
}
el.innerHTML = html;
} catch(e) {
el.innerHTML = '<div class="error">Failed to load: ' + e.message + '</div>';
} catch (e) {
const err = div('error'); err.textContent = 'Failed to load: ' + e.message; el.appendChild(err);
}
}
function showPanel(name) {
document.querySelectorAll('.panel').forEach(p => p.style.display = 'none');
document.getElementById('panel-' + name).style.display = '';
document.querySelectorAll('.sidebar-nav li').forEach(li => li.classList.remove('active'));
document.querySelector('.sidebar-nav li:nth-child(' + ({overview:1,versions:2,commands:3,agents:4}[name] || 1) + ')').classList.add('active');
if (name === 'overview') {
document.getElementById('panel-overview').innerHTML = '<div class="main-header"><h2>System Overview</h2><div id="header-status" class="status-badge">Loading...</div></div><div id="loading" class="loading">Loading...</div><div id="error" class="error" style="display:none;"></div><div id="content"><div class="row"><div class="col"><div class="card"><div class="card-title">Current Baseline</div><div id="base" class="card-value">-</div></div></div><div class="col"><div class="card"><div class="card-title">Git Branch</div><div id="branch" class="card-value">-</div></div></div><div class="col"><div class="card"><div class="card-title">Uptime</div><div id="uptime" class="card-value">-</div></div></div></div><div class="card" style="margin-top:16px"><div class="card-title">Event Bus Status</div><pre id="bus-status">-</pre></div></div>';
fetchData();
}
document.querySelector('.sidebar-nav li:nth-child(' + ({overview:1,versions:2,commands:3,agents:4,['arch-owner']:5,['arch-approve']:6}[name] || 1) + ')').classList.add('active');
if (name === 'overview') fetchData();
const loaders = {
'arch-owner': loadOwnerGate,
'arch-approve': loadApprove,
+25 -1
View File
@@ -12,6 +12,29 @@ from .task_lifecycle import TaskLifecycle
lifecycle = TaskLifecycle()
# Role must come from an authenticated principal, never from the client body.
_ROLE_WHITELIST = {"user", "agent", "executor", "dispatcher", "notifier", "group_admin", "owner"}
_authenticator = None # callable(payload) -> {"role": ...} | None, injected by host
def set_authenticator(fn):
"""Install a server-side authenticator used to derive the actor role.
`fn` receives the raw request payload and must return either None (or a
dict without a whitelisted "role") for anonymous, or a dict containing a
server-verified "role". The client-supplied `payload["role"]` is ignored.
"""
global _authenticator
_authenticator = fn
def effective_role(payload: dict[str, Any]) -> str:
if _authenticator is not None:
principal = _authenticator(payload)
if isinstance(principal, dict) and principal.get("role") in _ROLE_WHITELIST:
return principal["role"]
return "user"
def get_task_actions(task: dict[str, Any], role: str) -> dict[str, Any]:
return {
@@ -31,7 +54,8 @@ def build_router():
@router.post("/{task_id}/actions/{action}", operation_id="executeTaskAction")
def action(task_id: str, action: str, payload: dict[str, Any]):
return execute_task_action(payload.get("task", {}), action, payload.get("role", "user"), payload.get("reason"))
# Role is derived from the authenticated principal; body role is ignored.
return execute_task_action(payload.get("task", {}), action, effective_role(payload), payload.get("reason"))
return router
+386 -121
View File
@@ -8,6 +8,7 @@ import time
import uuid
import urllib.parse
from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
from datetime import datetime, timezone
from typing import Any, Callable
from .delegation import DelegationEngine
@@ -23,6 +24,8 @@ from .errors import (
from .evidence import EvidenceService
from .journal import TaskJournal
from .lease import LeaseManager
from .lineage_store import TaskLineagePolicy, TaskPlaneLineageCoordinator
from .mission import MissionAuthority
from .models import (
EvidenceEnvelope,
ResultEnvelope,
@@ -33,7 +36,11 @@ from .models import (
validate_identifier,
)
from .policy import PolicyEngine
from .schedule_service import SchedulingService
from .root_action_consumer import (
RootActionConsumer,
RootActionExecutor,
RootActionRunResult,
)
from .store import TaskPlaneStore
from .worker_registry import WorkerRegistry
@@ -54,7 +61,15 @@ def _bounded_int(value: Any, name: str, minimum: int, maximum: int) -> int:
class TaskPlaneService:
def __init__(self, store: TaskPlaneStore, policy: PolicyEngine | None = None):
def __init__(
self,
store: TaskPlaneStore,
policy: PolicyEngine | None = None,
*,
lineage_policy: TaskLineagePolicy | None = None,
mission_authority: MissionAuthority | None = None,
root_action_executor: RootActionExecutor | None = None,
):
self.store = store
self.policy = policy or PolicyEngine(
execution_enabled=False,
@@ -67,31 +82,257 @@ class TaskPlaneService:
self.workers = WorkerRegistry(store)
self.evidence = EvidenceService(store)
self.delegation = DelegationEngine(store)
self.scheduling = SchedulingService(store)
if mission_authority is not None and mission_authority.store is not store:
raise ValueError("mission authority must use the TaskPlane store")
self.missions = mission_authority
effective_lineage_policy = lineage_policy
if mission_authority is not None:
effective_lineage_policy = (
lineage_policy
or TaskLineagePolicy(
principal_bindings=(), acceptance_evaluators={}
)
).with_acceptance_evaluator_resolver(
mission_authority.resolve_acceptance_evaluator
)
self.lineage = (
TaskPlaneLineageCoordinator(store, effective_lineage_policy)
if effective_lineage_policy is not None
else None
)
self.root_actions = (
RootActionConsumer(
store,
plan_builder=lambda review: mission_authority.build_root_action_plan(
review=review, lineage=self.lineage
),
rereviewer=lambda plan, receipt: mission_authority.rereview_root_action(
plan=plan,
execution_receipt=receipt,
lineage=self.lineage,
),
executor=root_action_executor,
)
if mission_authority is not None and self.lineage is not None
else None
)
self.store.probe_attestation_verifier = self.policy.probe_attestation_verifier
def create_task(self, body: dict[str, Any], idempotency_key: str | None = None) -> dict[str, Any]:
def consume_one_root_action(
self,
*,
consumer_id: str,
now: datetime | None = None,
lease_seconds: int = 60,
) -> RootActionRunResult:
if self.root_actions is None:
raise ForbiddenError("root action consumer is not configured")
return self.root_actions.run_once(
consumer_id=consumer_id,
now=now or datetime.now(tz=timezone.utc),
lease_seconds=lease_seconds,
)
def create_mission(
self,
body: dict[str, Any],
idempotency_key: str | None,
*,
actor_identity: str | None,
) -> dict[str, Any]:
if self.missions is None or self.lineage is None:
raise ForbiddenError("mission authority is not configured")
if actor_identity is None:
raise ForbiddenError("mission creation requires authenticated identity")
if set(body) != {"goal_slug", "original_acceptance"}:
raise ValidationError(
"mission body requires only goal_slug and original_acceptance"
)
self.policy.assert_allowed("create", actor_identity)
mission = self.missions.create_mission(
goal_slug=body["goal_slug"],
original_acceptance=body["original_acceptance"],
idempotency_key=idempotency_key,
authenticated_identity=actor_identity,
lineage=self.lineage,
)
return {
**mission.to_dict(),
"root_task": self.store.get_task(mission.root_task_id).to_dict(),
}
def get_mission(
self, mission_id: str, *, actor_identity: str | None
) -> dict[str, Any]:
if self.missions is None:
raise ForbiddenError("mission authority is not configured")
if actor_identity is None:
raise ForbiddenError("mission read requires authenticated identity")
self.policy.assert_allowed("read", actor_identity)
mission = self.missions.get_mission(
mission_id, authenticated_identity=actor_identity
)
return {
**mission.to_dict(),
"root_task": self.store.get_task(mission.root_task_id).to_dict(),
}
def delegate_mission(
self,
parent_task_id: str,
body: dict[str, Any],
idempotency_key: str | None,
*,
actor_identity: str | None,
) -> dict[str, Any]:
"""Admit one server-identified child under an owned canonical Mission."""
if self.missions is None or self.lineage is None:
raise ForbiddenError("mission authority is not configured")
if actor_identity is None:
raise ForbiddenError("mission delegation requires authenticated identity")
# Mission delegation is an admission operation. It deliberately does
# not enable the worker execution actions guarded by execution_enabled.
self.policy.assert_allowed("create", actor_identity)
granted = set(
self.policy.identity_capabilities.get(
actor_identity,
self.policy.identity_capabilities.get("*", set()),
)
)
if "task.delegate" not in granted:
raise ForbiddenError("action delegate not permitted")
child = self.missions.delegate_child(
parent_task_id=parent_task_id,
body=body,
idempotency_key=idempotency_key,
authenticated_identity=actor_identity,
lineage=self.lineage,
delegation=self.delegation,
)
return child.to_dict()
def attach_legacy_attachment(
self,
mission_id: str,
body: dict[str, Any],
idempotency_key: str | None,
*,
actor_identity: str | None,
) -> dict[str, Any]:
if self.missions is None:
raise ForbiddenError("mission authority is not configured")
if actor_identity is None:
raise ForbiddenError("legacy attachment requires authenticated identity")
self.policy.assert_allowed("create", actor_identity)
granted = set(
self.policy.identity_capabilities.get(
actor_identity, self.policy.identity_capabilities.get("*", set())
)
)
if "task.delegate" not in granted:
raise ForbiddenError("action delegate not permitted")
return self.missions.attach_legacy(
mission_id=mission_id,
body=body,
idempotency_key=idempotency_key,
authenticated_identity=actor_identity,
)
def get_legacy_attachment(
self,
mission_id: str,
attachment_id: str,
*,
actor_identity: str | None,
) -> dict[str, Any]:
if self.missions is None:
raise ForbiddenError("mission authority is not configured")
if actor_identity is None:
raise ForbiddenError("legacy attachment read requires authenticated identity")
self.policy.assert_allowed("read", actor_identity)
return self.missions.get_legacy_attachment(
mission_id=mission_id,
attachment_id=attachment_id,
authenticated_identity=actor_identity,
)
def create_task(
self,
body: dict[str, Any],
idempotency_key: str | None = None,
*,
actor_identity: str | None = None,
) -> dict[str, Any]:
if body.get("parent_task_id"):
raise ForbiddenError("child tasks must use the delegate endpoint")
issuer = body.get("issuer", {})
actor = issuer.get("id", "anon") if isinstance(issuer, dict) else "anon"
if actor_identity is not None:
if issuer and (
not isinstance(issuer, dict) or issuer.get("id") != actor_identity
):
raise ForbiddenError("issuer identity mismatch")
body = dict(body)
body["issuer"] = {"type": "authenticated", "id": actor_identity}
actor = actor_identity or (
issuer.get("id", "anon") if isinstance(issuer, dict) else "anon"
)
self.policy.assert_allowed("create", actor)
try:
plan = TaskPlan.from_dict(body)
except (TypeError, ValueError) as exc:
raise ValidationError("invalid task plan") from exc
task = self.store.create_task_with_event(plan, idempotency_key=idempotency_key)
if self.lineage is not None:
if actor_identity is None:
raise ForbiddenError("lineage admission requires authenticated identity")
task = self.lineage.create_root(
plan=plan,
authenticated_identity=actor_identity,
request_digest=self.lineage._request_digest(body),
idempotency_key=idempotency_key,
)
else:
task = self.store.create_task_with_event(
plan, idempotency_key=idempotency_key
)
return task.to_dict()
def _can_read_mission_task(self, task, actor: str | None) -> bool:
# Every service sharing the store must honor persisted Mission ownership,
# including a legacy launcher that has no Mission authority configured.
if self.missions is not None:
return self.missions.can_read_task(
task, authenticated_identity=actor or ""
)
with self.store._lock:
return self.store.conn.execute(
"SELECT 1 FROM task_missions WHERE root_task_id=?",
(task.root_task_id,),
).fetchone() is None
def _assert_mission_read(self, task, actor: str | None) -> None:
if not self._can_read_mission_task(task, actor):
raise ForbiddenError("mission task requires its configured owner authority")
def list_tasks(self, actor: str | None = None) -> list[dict[str, Any]]:
if actor is not None:
self.policy.assert_allowed("read", actor)
return [task.to_dict(include_task=False) for task in self.store.list_tasks()]
tasks = self.store.list_tasks()
tasks = [task for task in tasks if self._can_read_mission_task(task, actor)]
return [task.to_dict(include_task=False) for task in tasks]
def get_task(self, task_id: str, actor: str | None = None) -> dict[str, Any]:
task = self.store.get_task(task_id)
self._assert_mission_read(task, actor)
if actor is not None:
self.policy.assert_allowed("read", actor, task=self.store.get_task(task_id))
return self.store.get_task(task_id).to_dict()
self.policy.assert_allowed("read", actor, task=task)
if self.missions is not None and self.missions.manages_root(
task.root_task_id
):
self.missions.assert_task_read(
task, authenticated_identity=actor
)
return task.to_dict()
def get_worker(self, worker_id: str, actor: str | None = None) -> dict[str, Any]:
if actor is not None:
@@ -309,17 +550,51 @@ class TaskPlaneService:
)
evidence_entries.append(envelope)
previous = envelope.evidence_digest
self.store.complete_claim_with_evidence_and_event(
task_id,
claim_id,
final_status = (
status
if status
in {"COMPLETED", "FAILED_FINAL", "FAILED_RECOVERABLE", "CANCELLED"}
else "COMPLETED",
result.to_dict(),
evidence_entries,
else "COMPLETED"
)
return result.to_dict()
lineage_metadata = task.task_plan.metadata.get("aios_task_lineage")
is_lineage_root = bool(
self.lineage is not None
and isinstance(lineage_metadata, dict)
and lineage_metadata.get("relation_type") == "ROOT"
)
if is_lineage_root and final_status == "COMPLETED":
raise ForbiddenError("lineage root completion requires root acceptance review")
review_outcome = None
with self.store._transaction():
self.store.complete_claim_with_evidence_and_event(
task_id,
claim_id,
final_status,
result.to_dict(),
evidence_entries,
)
is_lineage_child = bool(
self.lineage is not None
and isinstance(lineage_metadata, dict)
and lineage_metadata.get("relation_type") != "ROOT"
)
if is_lineage_child and final_status in {"COMPLETED", "FAILED_FINAL", "CANCELLED"}:
review_outcome = self.lineage.review_proven_terminal_child(
child_task_id=task_id,
claim_id=str(claim_id),
result_digest=str(result.result_digest),
evidence_refs=[entry.evidence_id for entry in evidence_entries],
)
response = result.to_dict()
if review_outcome is not None:
response["root_review"] = {
"review_id": review_outcome.review.review_id,
"root_task_id": review_outcome.review.root_task_id,
"root_complete": review_outcome.review.root_complete,
"next_legal_action": review_outcome.review.next_legal_action.value,
"dispatch_status": review_outcome.dispatch.status,
}
return response
def fail(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
claim_id = body.get("claim_id")
@@ -348,14 +623,38 @@ class TaskPlaneService:
)
return updated.to_dict()
def delegate(self, task_id: str, body: dict[str, Any]) -> dict[str, Any]:
def delegate(
self,
task_id: str,
body: dict[str, Any],
idempotency_key: str | None = None,
) -> dict[str, Any]:
actor = body.get("actor_identity") or "owner"
self.policy.assert_allowed("delegate", actor)
parent = self.store.get_task(task_id)
mission_managed = bool(
self.lineage is not None
and self.missions is not None
and self.missions.manages_root(parent.root_task_id)
)
if mission_managed:
plan_body = body.get("task") if "task" in body else {
key: value
for key, value in body.items()
if key != "actor_identity"
}
return self.delegate_mission(
task_id,
plan_body,
idempotency_key,
actor_identity=str(actor),
)
self.policy.assert_allowed("delegate", actor)
self.policy.assert_allowed("delegate", actor, task=parent)
if "task" in body:
plan_body = body["task"]
try:
child_plan = TaskPlan.from_dict(body["task"])
child_plan = TaskPlan.from_dict(plan_body)
except (TypeError, ValueError) as exc:
raise ValidationError("invalid child task plan") from exc
else:
@@ -366,93 +665,68 @@ class TaskPlaneService:
except (TypeError, ValueError) as exc:
raise ValidationError("invalid child task plan") from exc
child_plan.parent_task_id = task_id
child = self.delegation.delegate(task_id, child_plan, caller=actor)
if self.lineage is not None:
if not isinstance(plan_body, dict):
raise ValidationError("invalid child task plan")
if "task_id" not in plan_body:
raise ValidationError("lineage delegation requires a stable task_id")
plan_body = dict(plan_body)
issuer = plan_body.get("issuer")
if issuer and (
not isinstance(issuer, dict) or issuer.get("id") != actor
):
raise ForbiddenError("child issuer identity mismatch")
plan_body["issuer"] = {"type": "authenticated", "id": actor}
child_plan.issuer = dict(plan_body["issuer"])
child = self.lineage.delegate_child(
parent_task_id=task_id,
plan=child_plan,
authenticated_identity=str(actor),
request_digest=self.lineage._request_digest(plan_body),
delegation=self.delegation,
)
else:
child = self.delegation.delegate(task_id, child_plan, caller=actor)
return child.to_dict()
def get_events(self, task_id: str, actor: str | None = None) -> list[dict[str, Any]]:
task = self.store.get_task(task_id)
self._assert_mission_read(task, actor)
if actor is not None:
self.policy.assert_allowed("audit", actor, task=self.store.get_task(task_id))
self.store.get_task(task_id)
self.policy.assert_allowed("audit", actor, task=task)
if self.missions is not None and self.missions.manages_root(
task.root_task_id
):
self.missions.assert_task_read(
task, authenticated_identity=actor
)
return self.store.list_events(task_id)
def get_evidence(self, task_id: str, actor: str | None = None) -> list[dict[str, Any]]:
task = self.store.get_task(task_id)
self._assert_mission_read(task, actor)
if actor is not None:
self.policy.assert_allowed("audit", actor, task=self.store.get_task(task_id))
self.store.get_task(task_id)
self.policy.assert_allowed("audit", actor, task=task)
if self.missions is not None and self.missions.manages_root(
task.root_task_id
):
self.missions.assert_task_read(
task, authenticated_identity=actor
)
return self.evidence.list(task_id)
def get_children(self, task_id: str, actor: str | None = None) -> list[str]:
if actor is not None:
self.policy.assert_allowed("read", actor, task=self.store.get_task(task_id))
self.store.get_task(task_id)
return self.store.list_children(task_id)
# -- ANNEX-2.3 R2 governed claim / pull endpoints -------------------------
def governed_claim(self, task_id: str, body: dict[str, Any], actor: str) -> dict[str, Any]:
self.policy.assert_allowed("claim", actor)
task = self.store.get_task(task_id)
self.policy.assert_allowed("claim", actor, task=task)
worker_id = body.get("worker_id")
if not worker_id:
raise ValidationError("worker_id required")
worker = self.store.get_worker(worker_id)
if worker.identity != actor:
raise ForbiddenError("worker identity mismatch")
self.policy.assert_readonly_worker_binding(actor, worker_id)
lease_seconds = _bounded_int(
body.get("lease_seconds", 30), "lease_seconds", 1, 3600
)
return self.scheduling.claim_governed(
worker_id, task_id, lease_seconds=lease_seconds
)
def long_poll(self, worker_id: str, actor: str, timeout_seconds: int | None = None) -> dict[str, Any]:
self._assert_worker_pull(worker_id, actor)
return self.scheduling.long_poll(worker_id, timeout_seconds=timeout_seconds)
def poll(self, worker_id: str, actor: str) -> dict[str, Any]:
self._assert_worker_pull(worker_id, actor)
return self.scheduling.poll(worker_id)
def schedule(self, worker_id: str, actor: str) -> dict[str, Any]:
worker = self.store.get_worker(worker_id)
if worker.identity != actor:
raise ForbiddenError("worker identity mismatch")
return self.scheduling.schedule(worker_id)
def _assert_worker_pull(self, worker_id: str, actor: str) -> None:
self.policy.assert_allowed("claim", actor)
worker = self.store.get_worker(worker_id)
if worker.identity != actor:
raise ForbiddenError("worker identity mismatch")
self.policy.assert_readonly_worker_binding(actor, worker_id)
def get_proof(self, proof_id: str, actor: str | None = None) -> dict[str, Any]:
self._assert_mission_read(task, actor)
if actor is not None:
self.policy.assert_allowed("audit", actor)
return self.store.get_schedulability_proof(proof_id)
def get_receipt(self, receipt_id: str, actor: str | None = None) -> dict[str, Any]:
if actor is not None:
self.policy.assert_allowed("audit", actor)
return self.store.get_receipt(receipt_id)
def list_task_receipts(self, task_id: str, actor: str | None = None) -> list[dict[str, Any]]:
if actor is not None:
self.policy.assert_allowed("audit", actor, task=self.store.get_task(task_id))
self.store.get_task(task_id)
return self.store.list_receipts(task_id)
def get_active_claim(self, task_id: str, actor: str) -> dict[str, Any]:
"""Return the caller's active claim for a task (used by renew/readback)."""
self.policy.assert_allowed("read", actor, task=self.store.get_task(task_id))
claim = self.store.active_claim(task_id)
if claim is None:
return {"task_id": task_id, "active": False}
if claim.worker_identity != actor and actor not in self.policy.admin_identities:
raise ForbiddenError("claim identity mismatch")
return {"task_id": task_id, "active": True, "claim": claim.to_dict()}
self.policy.assert_allowed("read", actor, task=task)
if self.missions is not None and self.missions.manages_root(
task.root_task_id
):
self.missions.assert_task_read(
task, authenticated_identity=actor
)
return self.store.list_children(task_id)
class TaskPlaneRequestHandler(BaseHTTPRequestHandler):
@@ -543,6 +817,10 @@ class TaskPlaneRequestHandler(BaseHTTPRequestHandler):
if len(parts) == 1 and parts[0] == "tasks":
return 200, self.service.list_tasks(identity)
if len(parts) == 2 and parts[0] == "missions":
return 200, self.service.get_mission(
parts[1], actor_identity=identity
)
if len(parts) == 2 and parts[0] == "tasks":
task_id = parts[1]
return 200, self.service.get_task(task_id, identity)
@@ -556,30 +834,6 @@ class TaskPlaneRequestHandler(BaseHTTPRequestHandler):
return 200, self.service.list_workers(identity)
if len(parts) == 2 and parts[0] == "workers":
return 200, self.service.get_worker(parts[1], identity)
if len(parts) == 3 and parts[0] == "workers" and parts[2] == "long_poll":
query = urllib.parse.parse_qs(
urllib.parse.urlparse(self.path).query
)
raw = query.get("timeout_seconds", ["15"])[0]
try:
timeout = int(raw)
except (TypeError, ValueError):
timeout = 15
return 200, self.service.long_poll(
parts[1], identity, timeout_seconds=timeout
)
if len(parts) == 3 and parts[0] == "workers" and parts[2] == "poll":
return 200, self.service.poll(parts[1], identity)
if len(parts) == 3 and parts[0] == "workers" and parts[2] == "schedule":
return 200, self.service.schedule(parts[1], identity)
if len(parts) == 2 and parts[0] == "proofs":
return 200, self.service.get_proof(parts[1], identity)
if len(parts) == 2 and parts[0] == "receipts":
return 200, self.service.get_receipt(parts[1], identity)
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "receipts":
return 200, self.service.list_task_receipts(parts[1], identity)
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "active-claim":
return 200, self.service.get_active_claim(parts[1], identity)
raise NotFoundError("unknown endpoint")
self._handle(route)
@@ -592,6 +846,13 @@ class TaskPlaneRequestHandler(BaseHTTPRequestHandler):
parts = [p for p in path.split("/") if p]
headers = {k.lower(): v for k, v in self.headers.items()}
if len(parts) == 1 and parts[0] == "missions":
return 201, self.service.create_mission(
body or {},
headers.get("idempotency-key"),
actor_identity=identity,
)
if len(parts) == 1 and parts[0] == "tasks":
issuer = body.get("issuer")
if issuer and (
@@ -599,14 +860,16 @@ class TaskPlaneRequestHandler(BaseHTTPRequestHandler):
):
raise ForbiddenError("issuer identity mismatch")
body["issuer"] = {"type": "authenticated", "id": identity}
return 201, self.service.create_task(body or {}, headers.get("idempotency-key"))
return 201, self.service.create_task(
body or {},
headers.get("idempotency-key"),
actor_identity=identity,
)
if body.get("actor_identity") not in {None, identity}:
raise ForbiddenError("actor identity mismatch")
body["actor_identity"] = identity
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "claim":
return 200, self.service.claim(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "claim-governed":
return 200, self.service.governed_claim(parts[1], body or {}, identity)
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "heartbeat":
return 200, self.service.heartbeat(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "progress":
@@ -618,7 +881,9 @@ class TaskPlaneRequestHandler(BaseHTTPRequestHandler):
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "cancel":
return 200, self.service.cancel(parts[1], body or {})
if len(parts) == 3 and parts[0] == "tasks" and parts[2] == "delegate":
return 200, self.service.delegate(parts[1], body or {})
return 200, self.service.delegate(
parts[1], body or {}, headers.get("idempotency-key")
)
if len(parts) == 2 and parts[0] == "workers" and parts[1] == "register":
if (
@@ -1,286 +0,0 @@
"""Read-only Runtime / Authority Catalog (Catalog V1).
Pure evidence-producer / resolve layer. It deliberately does NOT create an
authority store, does NOT write to any database, and does NOT mutate
worker / task / business state (red line: no parallel authority store
without an ADR).
Result shape follows the Catalog V1 contract (V4 section 4.1)::
{
"schema_version": "aios.catalog.v1",
"resource_type": ...,
"resource_id": ...,
"authority": {"kind", "endpoint_or_path", "revision"},
"runtime": {"service", "pid", "loaded_source_path", "source_commit",
"artifact_sha256", "state_root"},
"contracts": [...],
"conflicts": [],
"evidence": [],
}
Read-only principle: parse as much as is observable from the supplied
``TaskPlaneStore`` and the resolve arguments. Anything not observable is
left ``None`` / unknown -- we never fabricate live evidence. If no store
is supplied, every resolve returns a structurally complete record with all
observation fields set to ``None`` and an empty ``evidence`` list.
"""
from __future__ import annotations
import dataclasses
from dataclasses import dataclass, field
from typing import Any
SCHEMA_VERSION = "aios.catalog.v1"
# Runtime fields that describe the loaded source of a runtime instance.
# Their absence is what the ``runtime_source_unknown`` metric counts.
_RUNTIME_SOURCE_FIELDS = ("loaded_source_path", "source_commit", "artifact_sha256")
def _null_runtime() -> dict[str, Any]:
return {
"service": None,
"pid": None,
"loaded_source_path": None,
"source_commit": None,
"artifact_sha256": None,
"state_root": None,
}
def _null_authority(kind: str | None) -> dict[str, Any]:
return {
"kind": kind,
"endpoint_or_path": None,
"revision": None,
}
@dataclass(frozen=True)
class CatalogResource:
"""A single resolved catalog entry (immutable value object).
Every field is JSON-serialisable. Fields that could not be observed are
intentionally ``None`` rather than fabricated.
"""
schema_version: str = SCHEMA_VERSION
resource_type: str | None = None
resource_id: str | None = None
authority: dict = field(default_factory=lambda: _null_authority(None))
runtime: dict = field(default_factory=_null_runtime)
contracts: list = field(default_factory=list)
conflicts: list = field(default_factory=list)
evidence: list = field(default_factory=list)
def to_dict(self) -> dict[str, Any]:
return {
"schema_version": self.schema_version,
"resource_type": self.resource_type,
"resource_id": self.resource_id,
"authority": dict(self.authority),
"runtime": dict(self.runtime),
"contracts": list(self.contracts),
"conflicts": list(self.conflicts),
"evidence": list(self.evidence),
}
class AuthorityCatalog:
"""Read-only authority / runtime directory.
``store`` is optional. When ``None`` every resolve returns a complete
but field-empty record (no store -> nothing observable -> no fabrication).
Any read through the store is wrapped so a missing record degrades to the
same empty-but-complete shape instead of raising.
"""
def __init__(self, store: Any = None) -> None:
self.store = store
self._attempts = 0
self._resolved: dict[str, int] = {}
self._unknown: dict[str, int] = {}
self._missing = {
"authority_endpoint_or_path": 0,
"authority_revision": 0,
"loaded_source_path": 0,
"source_commit": 0,
"artifact_sha256": 0,
}
self._runtime_source_unknown = 0
# -- public resolve API ----------------------------------------------
def resolve_worker(self, worker_id: str) -> CatalogResource:
"""service/runtime/authority for a registered worker (or UNKNOWN)."""
resource = self._worker_or_unknown(str(worker_id))
self._record(resource, resolved=resource.authority.get("revision") is not None)
return resource
def resolve_task(self, task_id: str) -> CatalogResource:
"""task plan / revision authority (or UNKNOWN)."""
resource = self._task_or_unknown(str(task_id))
self._record(resource, resolved=resource.authority.get("revision") is not None)
return resource
def resolve_repository(
self, repo_id: str, authority_ref: str | None = None
) -> CatalogResource:
"""repo -> authority ref -> remote OID.
Remote OID is only reflected from the supplied ``authority_ref``; with
no live repository probe it reads as null/unknown rather than guessed.
"""
authority = {
"kind": "repository",
"endpoint_or_path": str(repo_id),
"revision": authority_ref,
}
resource = dataclasses.replace(
CatalogResource(resource_type="repository", resource_id=str(repo_id)),
authority=authority,
)
self._record(resource, resolved=authority.get("revision") is not None)
return resource
def resolve_generic(self, resource_type: str, resource_id: str) -> CatalogResource:
"""Dispatch known types; anything else -> complete-but-empty UNKNOWN."""
rtype = str(resource_type)
rid = str(resource_id)
if rtype == "worker":
return self.resolve_worker(rid)
if rtype == "task":
return self.resolve_task(rid)
if rtype == "repository":
return self.resolve_repository(rid)
# job / auth / release / schema / state_root / overlay / unknown:
# no live source -> never fabricated live state.
resource = CatalogResource(
resource_type=rtype,
resource_id=rid,
authority=_null_authority(rtype),
)
self._record(resource, resolved=False)
return resource
def summary(self) -> dict[str, Any]:
"""Resolve statistics, including ``runtime_source_unknown`` counters.
Consumed by the runtime-source-unknown metric: counts how many
resolved entries had no observable loaded source path / commit / SHA.
"""
return {
"schema_version": SCHEMA_VERSION,
"resolve_attempts": self._attempts,
"resolved": dict(self._resolved),
"unknown": dict(self._unknown),
"runtime_source_unknown": self._runtime_source_unknown,
"missing_fields": dict(self._missing),
}
# -- internal helpers ------------------------------------------------
def _worker_or_unknown(self, worker_id: str) -> CatalogResource:
base = CatalogResource(
resource_type="worker",
resource_id=worker_id,
authority=_null_authority("worker.registration"),
)
if self.store is None:
return base
try:
worker = self.store.get_worker(worker_id)
except Exception:
return base
return CatalogResource(
resource_type="worker",
resource_id=worker.worker_id,
authority={
"kind": "worker.registration",
"endpoint_or_path": worker.identity,
"revision": worker.registration_revision,
},
runtime={
"service": worker.worker_type,
"pid": None,
"loaded_source_path": None,
"source_commit": None,
"artifact_sha256": None,
"state_root": None,
},
contracts=[
{"field": "capabilities", "value": list(worker.capabilities)},
{"field": "supported_protocols", "value": list(worker.supported_protocols)},
{"field": "trust_zone", "value": worker.trust_zone},
{"field": "status", "value": worker.status},
],
conflicts=[],
evidence=[],
)
def _task_or_unknown(self, task_id: str) -> CatalogResource:
base = CatalogResource(
resource_type="task",
resource_id=task_id,
authority=_null_authority("task.plan"),
)
if self.store is None:
return base
try:
task = self.store.get_task(task_id)
except Exception:
return base
issuer = task.task_plan.issuer or {}
issuer_id = issuer.get("id") if isinstance(issuer, dict) else None
authorities = task.task_plan.authorities if hasattr(task.task_plan, "authorities") else None
return CatalogResource(
resource_type="task",
resource_id=task.task_id,
authority={
"kind": "task.plan",
"endpoint_or_path": issuer_id or None,
"revision": task.revision,
},
runtime={
"service": issuer.get("type") if isinstance(issuer, dict) else None,
"pid": None,
"loaded_source_path": authorities.get("loaded_source_path")
if isinstance(authorities, dict)
else None,
"source_commit": authorities.get("source_commit")
if isinstance(authorities, dict)
else None,
"artifact_sha256": authorities.get("artifact_sha256")
if isinstance(authorities, dict)
else None,
"state_root": task.task_plan.execution_zone,
},
contracts=[
{"field": "task_type", "value": task.task_plan.task_type},
{"field": "status", "value": task.status},
{"field": "risk", "value": task.task_plan.risk},
{"field": "data_classification", "value": task.task_plan.data_classification},
{"field": "execution_zone", "value": task.task_plan.execution_zone},
{"field": "network_policy", "value": task.task_plan.network_policy},
{"field": "required_capabilities", "value": list(task.task_plan.required_capabilities)},
],
conflicts=[],
evidence=[],
)
def _record(self, resource: CatalogResource, *, resolved: bool) -> None:
self._attempts += 1
rtype = resource.resource_type or "unknown"
bucket = self._resolved if resolved else self._unknown
bucket[rtype] = bucket.get(rtype, 0) + 1
if not resource.authority.get("endpoint_or_path"):
self._missing["authority_endpoint_or_path"] += 1
if resource.authority.get("revision") is None:
self._missing["authority_revision"] += 1
for name in _RUNTIME_SOURCE_FIELDS:
if resource.runtime.get(name) is None:
self._missing[name] += 1
if all(resource.runtime.get(name) is None for name in _RUNTIME_SOURCE_FIELDS):
self._runtime_source_unknown += 1
@@ -0,0 +1,94 @@
"""Agent discovery/claim coordinator over diagnostic schedulability evidence."""
from __future__ import annotations
from dataclasses import dataclass
from typing import Any, Mapping
from .schedulability import (
DiscoveryDiagnostics,
SchedulabilityProof,
SchedulabilityService,
)
@dataclass(frozen=True)
class DiscoveryResult:
status: str
proof: SchedulabilityProof | None
blocker_codes: tuple[str, ...] = ()
candidate_count: int = 0
def to_dict(self) -> dict[str, Any]:
value: dict[str, Any] = {
"status": self.status,
"blocker_codes": list(self.blocker_codes),
"candidate_count": self.candidate_count,
"proof": None,
}
if self.proof is not None:
value["proof"] = {
"schema": self.proof.schema,
"proof_id": self.proof.proof_id,
"task_id": self.proof.task_id,
"worker_id": self.proof.worker_id,
"created_at": self.proof.created_at,
"expires_at": self.proof.expires_at,
"blockers": [b.to_dict() for b in self.proof.blockers],
"proof_digest": self.proof.proof_digest,
}
return value
def _waiting_status(diag: DiscoveryDiagnostics) -> str:
codes = set(diag.blocker_codes)
if diag.candidate_count == 0:
return "NO_CANDIDATES"
if not codes:
return "NO_CLAIMABLE_TASK"
if codes <= {"CAPACITY_EXHAUSTED"}:
return "WAITING_CAPACITY"
if "AUTHORIZATION_BLOCKED" in codes or "EXACT_SCOPE_MISMATCH" in codes:
return "WAITING_AUTHORIZATION"
if "DEPENDENCY_BLOCKED" in codes:
return "WAITING_DEPENDENCY"
if "FREEZE_BLOCKED" in codes:
return "WAITING_FREEZE"
if "CONFLICTING_LEASE" in codes:
return "WAITING_LEASE"
if {"STALE_OR_INACTIVE_WORKER", "MISSING_CAPABILITY", "INVALID_AUTHORITY_VALUE"} & codes:
return "WORKER_NOT_ELIGIBLE"
if "TASK_STATE_NOT_CLAIMABLE" in codes:
return "NO_CLAIMABLE_TASK"
return "BLOCKED"
class AutoClaimCoordinator:
"""No implicit credential derivation; authenticated identity is supplied by host."""
def __init__(self, schedulability: SchedulabilityService) -> None:
self.schedulability = schedulability
def discover(self, *, worker_id: str, actor_identity: str) -> DiscoveryResult:
proof, diag = self.schedulability.discover_with_diagnostics(
worker_id=worker_id, actor_identity=actor_identity
)
if proof is not None:
return DiscoveryResult(
"CLAIMABLE", proof, diag.blocker_codes, diag.candidate_count
)
return DiscoveryResult(
_waiting_status(diag), None, diag.blocker_codes, diag.candidate_count
)
def claim(
self,
*,
proof_id: str,
worker_id: str,
actor_identity: str,
lease_seconds: int = 30,
) -> Mapping[str, Any]:
return self.schedulability.claim_with_recheck(
proof_id=proof_id,
worker_id=worker_id,
actor_identity=actor_identity,
lease_seconds=lease_seconds,
)
File diff suppressed because it is too large Load Diff
@@ -1,320 +0,0 @@
"""caller_scan: 只读的模块调用方扫描与孤儿判定骨架。
实现《AIOS_caller_scan_设计规格_V1》V1 骨架的只读部分:仅通过静态 AST 证据
识别哪些文件引用了目标模块(subject),并对"是否可能成为孤儿"给出评估。本模块
**绝不执行被扫描模块、不 import 任何业务模块、不读 secret、不做任何写操作**。
红线(本骨架强制):
- import 的唯一证据是 AST(标准库 ast),regex 仅用于 recall 并在 evidence 标注
evidence kind。
- 任一 probe 超时/无法解析 → coverage 非完整 → 禁止判 PROVEN_ORPHAN。
- UNKNOWN 禁止被误降为 orphan(除非显式 allow_unknown_lower)。
- 测试引用 (filename 含 tests_marker) 绝不能证明 live。
"""
from __future__ import annotations
import ast
import json
from dataclasses import dataclass
from enum import StrEnum
from pathlib import Path
from typing import Optional
class CallerKind(StrEnum):
STATIC_PRODUCTION = "STATIC_PRODUCTION"
STATIC_TEST_ONLY = "STATIC_TEST_ONLY"
SYSTEMD = "SYSTEMD"
API_ROUTE = "API_ROUTE"
SCHEDULER = "SCHEDULER"
RUNTIME_LOADED = "RUNTIME_LOADED"
WRAPPER = "WRAPPER"
UNKNOWN_DYNAMIC = "UNKNOWN_DYNAMIC"
class OrphanAssessment(StrEnum):
LIVE = "LIVE"
UNKNOWN = "UNKNOWN"
LIKELY_ORPHAN = "LIKELY_ORPHAN"
PROVEN_ORPHAN = "PROVEN_ORPHAN"
@dataclass(frozen=True)
class CallerReference:
"""一个静态调用方文件引用。"""
path: str
kind: CallerKind
line: int
evidence: str # 触发判定的语句片段
evidence_kind: str # "ast" 或 "regex-recall"
import_text: str = "" # 上一条 import 原文(recall / 说明用)
def to_dict(self) -> dict:
return {
"path": self.path,
"kind": str(self.kind),
"line": self.line,
"evidence": self.evidence,
"evidence_kind": self.evidence_kind,
"import_text": self.import_text,
}
@dataclass(frozen=True)
class CallerScanResult:
subject: str
callers: tuple[CallerReference, ...] = ()
runtime_loaded: tuple[str, ...] = ()
unknown_dynamic_refs: tuple[dict, ...] = ()
test_only_callers: tuple[CallerReference, ...] = ()
caller_count: int = 0
assessment: OrphanAssessment = OrphanAssessment.UNKNOWN
coverage: bool = False
evidence_refs: tuple[str, ...] = ()
def to_dict(self) -> dict:
return {
"subject": self.subject,
"callers": [c.to_dict() for c in self.callers],
"runtime_loaded": list(self.runtime_loaded),
"unknown_dynamic_refs": list(self.unknown_dynamic_refs),
"test_only_callers": [c.to_dict() for c in self.test_only_callers],
"caller_count": self.caller_count,
"assessment": str(self.assessment),
"coverage": self.coverage,
"evidence_refs": list(self.evidence_refs),
}
def as_json(self) -> str:
return json.dumps(self.to_dict(), ensure_ascii=False, sort_keys=True)
# ---------------------------------------------------------------- helpers
_IMPORTLIB_CALLS = {"import_module", "find_module", "__import__"}
_API_ROUTE_MARKERS = ("@app.", "add_url_rule", "APIRouter", "Blueprint", "ROUTES =")
_SCHEDULER_MARKERS = ("schedule.", "cron", "while True", "APScheduler", "every(")
_SYSTEMD_MARKERS = ("systemd", "Unit]", "ExecStart", ".service")
def _import_statement_text(source: str, node: ast.AST) -> str:
"""尽力还原一条 import 语句的原文(用于 evidence 说明,不判读逻辑依据)。"""
try:
lines = source.splitlines()
start = getattr(node, "lineno", 1)
end = getattr(node, "end_lineno", start)
return "\n".join(lines[start - 1 : end]).strip()
except Exception: # pragma: no cover - 尽力而为的展示
return ""
def _root(name: str) -> str:
return name.split(".")[0]
def _classify_file_kind(filename: str, text: str) -> CallerKind:
"""对非测试文件做启发式归类(骨架级,浅表判断)。"""
low = filename.lower()
if "wrapper" in low:
return CallerKind.WRAPPER
if any(m in text for m in _SCHEDULER_MARKERS):
return CallerKind.SCHEDULER
if any(m in text for m in _API_ROUTE_MARKERS):
return CallerKind.API_ROUTE
if any(m in text for m in _SYSTEMD_MARKERS):
return CallerKind.SYSTEMD
return CallerKind.STATIC_PRODUCTION
def _is_test_file(filename: str, tests_marker: str) -> bool:
base = Path(filename).name
low = base.lower()
norm = filename.replace("\\", "/")
return (
low.startswith(tests_marker)
or ("_" + tests_marker) in low
or ("/" + tests_marker + "/") in norm
or (tests_marker in low and "tests" in low)
)
# ---------------------------------------------------------------- main scan
def scan_subject(
subject: str,
repo_path: Optional[str] = None,
tests_marker: str = "test",
allow_unknown_lower: bool = False,
) -> CallerScanResult:
"""只读扫描 repo_path 目录下的 .py 文件,收集对 subject 的静态引用。
只引入标准库 ast,不执行/不 import 被扫描文件。
默认即保守(allow_unknown_lower=False):只要存在不可解析的 opaque 动态
import,assessment 一律返回 UNKNOWN,不把 UNKNOWN 误降为 orphan;仅当调用方
显式传入 allow_unknown_lower=True 时才允许降为 LIKELY_ORPHAN。
"""
if not subject:
raise ValueError("subject must not be empty")
root = Path(repo_path or Path.cwd())
callers: list[CallerReference] = []
test_only: list[CallerReference] = []
runtime_literals: set[str] = set()
unknown_dynamic_refs: list[dict] = []
evidence_refs: list[str] = []
scan_errors = 0
py_files = sorted(str(p) for p in root.rglob("*.py") if p.is_file())
for fpath in py_files:
try:
source = Path(fpath).read_text(encoding="utf-8", errors="replace")
tree = ast.parse(source, filename=fpath)
except (SyntaxError, OSError, UnicodeDecodeError):
scan_errors += 1
continue
is_test = _is_test_file(fpath, tests_marker)
lines = source.splitlines()
for node in ast.walk(tree):
# --- 静态 import / from ... import ---
if isinstance(node, ast.Import):
for alias in node.names:
if _root(alias.name) == _root(subject):
kind = (
CallerKind.STATIC_TEST_ONLY
if is_test
else _classify_file_kind(fpath, source)
)
ref = CallerReference(
path=fpath,
kind=kind,
line=node.lineno,
evidence=f"import {alias.name}",
evidence_kind="ast",
import_text=_import_statement_text(source, node),
)
(test_only if kind is CallerKind.STATIC_TEST_ONLY else callers).append(
ref
)
evidence_refs.append(
f"{fpath}:{node.lineno} (ast) import {alias.name}"
)
elif isinstance(node, ast.ImportFrom):
if node.module and _root(node.module) == _root(subject):
kind = (
CallerKind.STATIC_TEST_ONLY
if is_test
else _classify_file_kind(fpath, source)
)
ref = CallerReference(
path=fpath,
kind=kind,
line=node.lineno,
evidence=f"from {node.module} import",
evidence_kind="ast",
import_text=_import_statement_text(source, node),
)
(test_only if kind is CallerKind.STATIC_TEST_ONLY else callers).append(
ref
)
evidence_refs.append(
f"{fpath}:{node.lineno} (ast) from {node.module} import"
)
# --- 动态 import(importlib 调用 / __import__)---
elif _is_dynamic_import(node):
target = None
args = node.args
if args and isinstance(args[0], ast.Constant):
target = str(args[0].value)
known = target is not None
hits_subject = known and (
_root(target) == _root(subject)
or target.startswith(_root(subject) + ".")
)
if known and hits_subject:
runtime_literals.add(target)
# 无论是否点名 subject,动态 / 不可解析都计入 unknown_dynamic_refs
unknown_dynamic_refs.append(
{
"path": fpath,
"line": node.lineno,
"target": target,
"resolvable": known,
"evidence_kind": "ast" if known else "regex-recall",
}
)
has_production = any(
c.kind is not CallerKind.STATIC_TEST_ONLY for c in callers
)
has_runtime_literal = bool(runtime_literals)
has_unknown_dynamic = bool(unknown_dynamic_refs)
coverage = scan_errors == 0
assessment = _assess(
has_production=has_production,
has_runtime_literal=has_runtime_literal,
has_unknown_dynamic=has_unknown_dynamic,
scan_complete=coverage,
allow_unknown_lower=allow_unknown_lower,
)
caller_count = len(
[c for c in callers if c.kind is not CallerKind.STATIC_TEST_ONLY]
)
return CallerScanResult(
subject=subject,
callers=tuple(callers),
runtime_loaded=tuple(sorted(runtime_literals)),
unknown_dynamic_refs=tuple(unknown_dynamic_refs),
test_only_callers=tuple(test_only),
caller_count=caller_count,
assessment=assessment,
coverage=coverage,
evidence_refs=tuple(evidence_refs),
)
def _is_dynamic_import(node: ast.AST) -> bool:
"""判断是否为 importlib 动态加载调用(import_module / find_module / __import__)。"""
if not isinstance(node, ast.Call):
return False
if isinstance(node.func, ast.Name):
return node.func.id == "__import__"
if isinstance(node.func, ast.Attribute):
return (
getattr(node.func, "attr", "") in _IMPORTLIB_CALLS
and _root(getattr(node.func.value, "id", "")) == "importlib"
)
return False
def _assess(
has_production: bool,
has_runtime_literal: bool,
has_unknown_dynamic: bool,
scan_complete: bool,
allow_unknown_lower: bool,
) -> OrphanAssessment:
"""判定评估等级(只读、保守)。"""
if has_production:
return OrphanAssessment.LIVE
# 存在运行时字面量引用(importlib 显式点名 subject)→ 潜在 live,不降 orphan
if has_runtime_literal:
return OrphanAssessment.UNKNOWN
if has_unknown_dynamic:
# 无法解析的动态引用:coverage 不完整,禁止 PROVEN_ORPHAN
if allow_unknown_lower:
return OrphanAssessment.LIKELY_ORPHAN
return OrphanAssessment.UNKNOWN
if not scan_complete:
# coverage 不完整 → 禁止 PROVEN_ORPHAN
return OrphanAssessment.LIKELY_ORPHAN
# 全量扫描且无任何引用 → 可 PROVEN
return OrphanAssessment.PROVEN_ORPHAN
-210
View File
@@ -1,210 +0,0 @@
"""Single Canonical Contract (P0-2 / WP3).
Deterministic, replayable, cross-process encoding and hashing for every
digest / evidence / readback value in the AIOS platform.
ARCHITECTURAL RED LINE
----------------------
This module is the ONLY canonical serialization entry point. Do NOT hand-write
``json.dumps(sort_keys=True, ...)`` anywhere else for digest/evidence/readback
calculation. Everything must flow through :func:`encode` (and :func:`sha256`).
Design / cross-process replayability
------------------------------------
* ``encode`` reuses ``task_plane.store._canonical`` (which is
``task_plane.models.canonical_json`` = ``json.dumps(..., sort_keys=True,
separators=(",", ":"))``). We do NOT re-implement serialization here; we
normalize the object into JSON-safe plain data first, then delegate to that
single base implementation.
* Determinism is guaranteed by construction:
- object keys are sorted (``sort_keys=True``);
- ``separators=(",", ":")`` removes insignificant whitespace;
- floats are normalized by Python's ``repr(float)`` round-trip (so ``1.0``
and ``1.00`` encode identically) — this is exactly what ``json.dumps``
emits;
- booleans are distinguished from ints; ``True`` never collides with ``1``.
* Non-finite floats (``inf`` / ``nan``) are rejected loudly instead of being
silently serialized as ``Infinity`` / ``NaN``, so a digest can never be
produced from an undefined value.
* dataclass instances are flattened via :func:`dataclasses.asdict` before
serialization, so typed envelopes (``EvidenceEnvelope`` and friends) are
first-class citizens.
* Because the canonical bytes are a pure function of the logical object, two
processes on two hosts reproduce identical bytes without any shared state ->
digests are replayable and comparable across processes.
Only the standard library is used (no heavy framework).
"""
from __future__ import annotations
import dataclasses
import hashlib
import json
import math
from enum import Enum
from pathlib import Path
from typing import Any
from .store import _canonical as _canonical_base
# ---------------------------------------------------------------------------
# Schema names (the six share one canonical system-truth contract; see PRD P0-2)
# ---------------------------------------------------------------------------
class SchemaName(str, Enum):
EVIDENCE_ENVELOPE = "EVIDENCE_ENVELOPE"
RELEASE_PLAN = "RELEASE_PLAN"
TASK_EXECUTION_ENVELOPE = "TASK_EXECUTION_ENVELOPE"
AUTHORIZATION_ENVELOPE = "AUTHORIZATION_ENVELOPE"
RUNTIME_IDENTITY = "RUNTIME_IDENTITY"
RECOVERY_RECEIPT = "RECOVERY_RECEIPT"
SCHEMA_NAMES = frozenset(name.value for name in SchemaName)
# ---------------------------------------------------------------------------
# Contract schema source of truth
# ---------------------------------------------------------------------------
_CONFIG_RELATIVE = ("config", "gitops", "system_truth_contract.json")
# Overridable for exotic layouts; never required at import time.
CONTRACT_SCHEMA_PATH_ENV = "AIOS_SYSTEM_TRUTH_CONTRACT_PATH"
def _config_path() -> Path | None:
"""Locate ``system_truth_contract.json`` without importing config packages."""
import os
override = os.environ.get(CONTRACT_SCHEMA_PATH_ENV)
if override:
cand = Path(override)
if cand.is_file():
return cand
# repo_root == contract.py parents[3] (task_plane -> app -> active-runtime -> aios)
cand = Path(__file__).resolve().parents[3].joinpath(*_CONFIG_RELATIVE)
return cand if cand.is_file() else None
def _load_contract_version() -> str | None:
"""Return the ``schema`` version string from the system-truth contract.
Returns ``None`` if the file is missing or does not expose a ``schema``
key (caller decides whether that is acceptable).
"""
path = _config_path()
if path is None:
return None
try:
data = json.loads(path.read_text(encoding="utf-8"))
except (json.JSONDecodeError, OSError, UnicodeError):
return None
if not isinstance(data, dict):
return None
value = data.get("schema")
return str(value) if isinstance(value, str) else None
# ---------------------------------------------------------------------------
# Encoding primitives
# ---------------------------------------------------------------------------
class ContractError(ValueError):
"""Raised when an object cannot be canonically encoded."""
def _normalize(obj: Any) -> Any:
"""Reduce ``obj`` to JSON-safe plain data, rejecting what must not be
silently serialized (non-finite floats, sets, arbitrary objects)."""
if obj is None or isinstance(obj, (str, bool, int)):
return obj
if isinstance(obj, float):
if math.isinf(obj) or math.isnan(obj):
raise ContractError(
f"non-finite float ({obj!r}) cannot be canonically encoded"
)
return obj
if isinstance(obj, dict):
return {str(key): _normalize(value) for key, value in obj.items()}
if isinstance(obj, (list, tuple)):
return [_normalize(value) for value in obj]
if dataclasses.is_dataclass(obj) and not isinstance(obj, type):
return _normalize(dataclasses.asdict(obj))
raise ContractError(
f"unsupported type for canonical encoding: {type(obj).__name__}"
)
def encode(obj: Any) -> str:
"""Return the deterministic canonical byte-string for ``obj``.
This is the single canonical serialization implementation. It normalizes
the object into JSON-safe plain data and delegates to the shared base
encoder (``store._canonical`` / ``models.canonical_json``).
"""
return _canonical_base(_normalize(obj))
def sha256(obj: Any) -> str:
"""Return the lowercase hex sha256 of :func:`encode` of ``obj``."""
return hashlib.sha256(encode(obj).encode("utf-8")).hexdigest()
# ---------------------------------------------------------------------------
# Validation
# ---------------------------------------------------------------------------
@dataclasses.dataclass(frozen=True)
class ValidationResult:
valid: bool
reason: str | None = None
schema_name: str | None = None
version: str | None = None
def __bool__(self) -> bool:
return self.valid
def validate(obj: Any, schema_name: str | SchemaName | None = None) -> ValidationResult:
"""Validate that ``obj`` is canonically encodable and, when a schema name is
given, that the schema is a known member of the system-truth contract."""
try:
encode(obj)
except ContractError as exc:
return ValidationResult(valid=False, reason=str(exc))
version = None
if schema_name is not None:
name = _coerce_schema_name(schema_name)
if name not in SCHEMA_NAMES:
return ValidationResult(
valid=False,
reason=f"unknown schema name: {name}",
schema_name=str(name),
)
version = _load_contract_version()
if version is None:
return ValidationResult(
valid=False,
reason="canonical contract schema file unavailable",
schema_name=str(name),
)
return ValidationResult(
valid=True, schema_name=str(name), version=version
)
return ValidationResult(valid=True)
def known_schema(name: str | SchemaName) -> str | None:
"""Return the canonical contract version for a *known* schema name, or
``None`` for an unknown name / unavailable schema file."""
key = _coerce_schema_name(name)
if key not in SCHEMA_NAMES:
return None
return _load_contract_version()
def _coerce_schema_name(name: str | SchemaName) -> str:
if isinstance(name, SchemaName):
return name.value
if isinstance(name, str):
return name
return str(name)
+16 -2
View File
@@ -31,7 +31,16 @@ class DelegationEngine:
current = self.store.parent_of(current)
return True
def delegate(self, parent_id: str, child_plan: TaskPlan, *, caller: str) -> TaskRecord:
def delegate(
self,
parent_id: str,
child_plan: TaskPlan,
*,
caller: str,
owner_principal_id: str = "",
mission_id: str = "",
created_at: float | None = None,
) -> TaskRecord:
self.can_delegate(parent_id, child_plan)
parent = self.store.get_task(parent_id)
if child_plan.required_capabilities and not set(child_plan.required_capabilities).issubset(
@@ -80,7 +89,12 @@ class DelegationEngine:
raise ConflictError("delegation task_id conflict")
return self.store.delegate_task_with_budget_and_events(
parent_id, child_plan, caller=caller
parent_id,
child_plan,
caller=caller,
owner_principal_id=owner_principal_id,
mission_id=mission_id,
created_at=created_at,
)
def _depth(self, task_id: str) -> int:
@@ -1,173 +0,0 @@
"""env_profile: 只读的环境能力画像骨架。
实现 V4 §4.2 / V3 TrackA A1 的只读环境画像:探测操作系统、shell、posix、
常用工具、能力与凭证机制,**绝不自动扩权、绝不改 sudoers、绝不安装生产依赖、
绝不收集 secret**。全部探测只读并用 try/except 包裹,失败项置 None/'Unknown'
并记录到 missing_requirements。
"""
from __future__ import annotations
import json
import os
import platform
import shutil
import sys
from dataclasses import dataclass, field
from typing import Optional
SCHEMA_VERSION = "aios.environment-profile.v1"
def _probe(call):
"""只读探测;失败返回 None。"""
try:
return call()
except Exception:
return None
def _which(tool: str) -> Optional[str]:
try:
w = shutil.which(tool)
return w if w else None
except Exception:
return None
@dataclass(frozen=True)
class EnvironmentCapabilityReport:
schema_version: str = SCHEMA_VERSION
host_id: Optional[str] = None
os: Optional[str] = None
shell: Optional[str] = None
posix: Optional[bool] = None
tools: dict = field(default_factory=dict)
capabilities: dict = field(default_factory=dict)
credential_mechanisms: list = field(default_factory=list)
filesystem: dict = field(default_factory=dict)
missing_requirements: list = field(default_factory=list)
def to_dict(self) -> dict:
return {
"schema_version": self.schema_version,
"host_id": self.host_id,
"os": self.os,
"shell": self.shell,
"posix": self.posix,
"tools": dict(self.tools),
"capabilities": dict(self.capabilities),
"credential_mechanisms": list(self.credential_mechanisms),
"filesystem": dict(self.filesystem),
"missing_requirements": list(self.missing_requirements),
}
def as_json(self) -> str:
return json.dumps(self.to_dict(), ensure_ascii=False, sort_keys=True)
def build_report() -> EnvironmentCapabilityReport:
"""采集当前环境的只读画像。永不抛异常,字段可能为 None。"""
os_name = _probe(lambda: os.name)
sys_platform = _probe(lambda: sys.platform)
system = _probe(lambda: platform.system())
posix = True if os_name == "posix" else (False if os_name is not None else None)
hostname = None
if os_name == "posix":
hostname = _probe(lambda: os.uname().nodename)
if not hostname:
hostname = _probe(lambda: os.environ.get("COMPUTERNAME")) or _probe(
lambda: os.environ.get("HOSTNAME")
)
python_ver = _probe(lambda: ".".join(platform.python_version_tuple()[:2]))
tools = {}
for tool, label in (
("git", "git"),
("python", "python"),
("python3", "python3"),
("node", "node"),
("npm", "npm"),
("ssh", "ssh"),
("gpg", "gpg"),
("docker", "docker"),
):
tools[label] = _which(tool)
missing = []
if tools["git"] is None:
missing.append("git")
if tools["python"] is None and tools["python3"] is None:
missing.append("python")
if types := [t for t in ("git", "python") if tools[t] is None]:
missing.extend(types)
capabilities = {
"os_name": os_name,
"system": system,
"sys_platform": sys_platform,
"python": python_ver,
"posix": posix,
}
# 凭证机制:只做"存在性"探测(env var / 配置目录是否存在),绝不读取 secret 值本身
credential_mechanisms = []
for var, label in (
("GIT_SSH_COMMAND", "git_ssh"),
("AGENT_SSH_KEY_FILE", "agent_ssh_key_file"),
("SSH_AUTH_SOCK", "ssh_agent"),
):
present = _probe(lambda v=var: bool(os.environ.get(v)))
if present:
credential_mechanisms.append(label)
# 配置目录存在性(不读内容)
for path, label in (
("~/.ssh", "ssh_config_dir"),
("~/.gnupg", "gnupg_dir"),
):
try:
expanded = os.path.expanduser(path)
if os.path.isdir(expanded):
credential_mechanisms.append(label)
except Exception:
pass
filesystem = {"cwd": _probe(lambda: os.getcwd())}
return EnvironmentCapabilityReport(
schema_version=SCHEMA_VERSION,
host_id=hostname,
os=os_name,
shell=_probe(lambda: os.environ.get("SHELL")) or _probe(lambda: os.environ.get("COMSPEC")),
posix=posix,
tools=tools,
capabilities=capabilities,
credential_mechanisms=sorted(set(credential_mechanisms)),
filesystem=filesystem,
missing_requirements=sorted(set(missing)),
)
def check_requirements(required: list[str]) -> EnvironmentCapabilityReport:
"""在 build_report 基础上,按 required 校验缺失项并写入 missing_requirements。
只读,不安装任何东西。
"""
report = build_report()
missing = set(report.missing_requirements)
tool_requirements = ("git", "python", "node", "ssh", "gpg", "docker", "npm", "python3")
for req in required or []:
if req in tool_requirements:
if report.tools.get(req) is None:
missing.add(req)
else:
# 非工具类要求的可验证性有限;不做过度校验,仅当明确缺失时记录
val = report.capabilities.get(req)
if val is not None:
missing.discard(req)
return EnvironmentCapabilityReport(
**{**report.to_dict(), "missing_requirements": sorted(missing)}
)
-114
View File
@@ -1,114 +0,0 @@
"""Gate / Evidence Engine — gate decisions over Verdicts (WP5 / P0-5).
A GateDecision wraps a Verdict with the gate kind that produced it. The
harden rule is that ok == (verdict.status == PASS); a gate never "passes"
on partial evidence, service liveness, or historical proof.
"""
from __future__ import annotations
import time
from dataclasses import dataclass, field
from enum import StrEnum
from typing import Any
from .verdict import Verdict, VerdictStatus, evaluate_verdict
class GateKind(StrEnum):
CLAIM = "CLAIM"
ADMIT = "ADMIT"
EXECUTE = "EXECUTE"
EVIDENCE = "EVIDENCE"
AUTHORIZE = "AUTHORIZE"
READBACK = "READBACK"
@dataclass(frozen=True)
class GateDecision:
"""A gate's decision: the Verdict plus the gate kind it came from."""
gate: GateKind
ok: bool
verdict: Verdict
checks: tuple[dict[str, Any], ...] = ()
reason: str = ""
decided_at: float = field(default_factory=time.time)
def to_dict(self) -> dict[str, Any]:
return {
"gate": self.gate.value,
"ok": self.ok,
"verdict": self.verdict.to_dict(),
"checks": list(self.checks),
"reason": self.reason,
"decided_at": self.decided_at,
}
def decide(gate_kind: GateKind | str, verdict: Verdict) -> GateDecision:
"""Build a GateDecision. ok is True IFF the verdict is PASS."""
ok = verdict.status == VerdictStatus.PASS
return GateDecision(
gate=GateKind(gate_kind),
ok=ok,
verdict=verdict,
checks=verdict.checks,
reason=verdict.reason,
decided_at=verdict.decided_at,
)
def claim_gate(recheck_decision: Any) -> GateDecision:
"""Wrap an authoritative_claim_recheck result as a CLAIM gate.
The CLAIM gate only passes when the full recheck is accepted.
RecheckDecision exposes `allowed` (not a `verdict` field) and its checks;
there is no REJECT status on the recheck side, so a denial is mapped to
the fail-closed BLOCKED verdict here (documented mapping).
"""
decided_at = getattr(recheck_decision, "observed_at", time.time())
checks = [c.to_dict() for c in getattr(recheck_decision, "checks", [])]
allowed = bool(getattr(recheck_decision, "allowed", False))
subject = getattr(recheck_decision, "task_id", "claim")
raw_reason = getattr(recheck_decision, "reason", "") or ""
if allowed:
status = VerdictStatus.PASS
reason = raw_reason or "all claim recheck checks accepted"
else:
# NOTE: recheck has no REJECT status -> map a denied recheck to BLOCKED
# (fail-closed). RecheckDecision.reason carries the failing check detail.
status = VerdictStatus.BLOCKED
reason = raw_reason or "claim recheck rejected"
verdict = Verdict(
scope="contract",
status=status,
subject=subject,
checks=tuple(checks),
evidence_refs=(),
missing_evidence=(),
reason=reason,
decided_at=decided_at,
)
return decide(GateKind.CLAIM, verdict)
def evidence_gate(
required_evidence: Any,
available_evidence: Any,
*,
scope: str = "runtime",
subject: str = "evidence",
reason: str = "",
) -> GateDecision:
"""EVIDENCE gate: PASS only when the evidence closure is fully satisfied."""
verdict = evaluate_verdict(
scope=scope,
subject=subject,
required_evidence=required_evidence,
available_evidence=available_evidence,
reason=reason,
)
return decide(GateKind.EVIDENCE, verdict)
@@ -1,141 +0,0 @@
"""Real HTTP RunnerTransport for the Agent Pull Runner protocol.
Implements the RunnerTransport Protocol from agent_pull_runner.py over
stdlib-only urllib.request, talking to the governed TaskPlane HTTP API.
See ARCH-AUTO-CLAIM-R2-20260914 §2 / §6 T9.
"""
from __future__ import annotations
import json
import urllib.error
import urllib.parse
import urllib.request
from typing import Any, Mapping, Sequence
from agent_pull_runner import RunnerError
class HttpRunnerTransport:
def __init__(
self,
base_url: str,
*,
token: str = "",
actor_identity: str | None = None,
timeout: float = 10,
) -> None:
self.base_url = base_url.rstrip("/")
self.token = token
self.actor_identity = actor_identity
self.timeout = timeout
def _request(
self,
method: str,
path: str,
body: Mapping[str, Any] | None = None,
query: Mapping[str, Any] | None = None,
) -> Any:
url = self.base_url + path
if query:
url += "?" + urllib.parse.urlencode(query)
request = urllib.request.Request(url, method=method)
if self.token:
request.add_header("Authorization", "Bearer " + self.token)
data = None
if body is not None:
data = json.dumps(body).encode("utf-8")
request.add_header("Content-Type", "application/json")
request.add_header("Content-Length", str(len(data)))
try:
with urllib.request.urlopen(request, data=data, timeout=self.timeout) as response:
return json.loads(response.read().decode("utf-8"))
except urllib.error.HTTPError as exc:
raw = exc.read().decode("utf-8")
try:
payload = json.loads(raw)
except (ValueError, json.JSONDecodeError):
payload = {"error": "http_error", "status": exc.code}
raise RunnerError(payload) from None
# -- RunnerTransport protocol ------------------------------------------
def register_worker(self, worker_id: str, capabilities: Sequence[str]) -> Mapping[str, Any]:
return self._request(
"POST",
"/workers/register",
{
"worker_id": worker_id,
"worker_type": "code.change",
"identity": self.actor_identity or worker_id,
"capabilities": list(capabilities),
"trust_zone": "default",
"max_concurrency": 1,
"heartbeat_interval_seconds": 15,
"supported_protocols": ["aios-task-v1"],
},
)
def heartbeat(self, worker_id: str) -> Mapping[str, Any]:
return self._request("POST", f"/workers/{worker_id}/heartbeat", {})
def long_poll(self, worker_id: str, timeout_seconds: int) -> Mapping[str, Any] | None:
result = self._request(
"GET",
f"/workers/{worker_id}/long_poll",
query={"timeout_seconds": int(timeout_seconds)},
)
if isinstance(result, dict) and result.get("status") == "idle":
return None
return result
def poll(self, worker_id: str) -> Mapping[str, Any] | None:
result = self._request("GET", f"/workers/{worker_id}/poll")
if isinstance(result, dict) and result.get("status") == "idle":
return None
return result
def claim(self, worker_id: str, task: Mapping[str, Any]) -> Mapping[str, Any]:
if isinstance(task, dict) and task.get("claim_id"):
# In the governed flow the long_poll already claimed the task;
# remain idempotent and return the already-claimed task.
return task
task_id = task.get("task_id")
if not task_id:
raise RunnerError("task_id is required")
return self._request(
"POST",
f"/tasks/{task_id}/claim-governed",
{"worker_id": worker_id},
)
def renew(self, worker_id: str, task_id: str, claim_id: str | None = None) -> Mapping[str, Any]:
# Lease extension is delegated to the claim heartbeat endpoint, which
# requires the claim_id and matches the bearer identity to the claim.
# When the caller does not supply claim_id we resolve the caller's own
# active claim via the read endpoint first.
if not claim_id:
resolved = self._request("GET", f"/tasks/{task_id}/active-claim")
if isinstance(resolved, dict) and resolved.get("active"):
claim = resolved["claim"]
claim_id = claim.get("claim_id")
if not claim_id:
raise RunnerError("no active claim to renew")
body = {"lease_seconds": 30, "claim_id": claim_id}
return self._request("POST", f"/tasks/{task_id}/heartbeat", body)
def report_progress(
self, worker_id: str, task_id: str, payload: Mapping[str, Any]
) -> Mapping[str, Any]:
return self._request("POST", f"/tasks/{task_id}/progress", payload)
def report_evidence(
self, worker_id: str, task_id: str, payload: Mapping[str, Any]
) -> Mapping[str, Any]:
return self._request("POST", f"/tasks/{task_id}/result", payload)
def report_result(
self, worker_id: str, task_id: str, payload: Mapping[str, Any]
) -> Mapping[str, Any]:
return self._request("POST", f"/tasks/{task_id}/result", payload)
@@ -0,0 +1,536 @@
"""Mission-owned, append-only bindings for verified external legacy executions."""
from __future__ import annotations
import json
import re
from dataclasses import dataclass
from datetime import datetime, timezone
from typing import Mapping, Protocol, Sequence
from .errors import ConflictError, NotFoundError, ValidationError
from .models import compute_sha256_hex
ATTACH_REQUEST_SCHEMA = "aios.mission-legacy-attach-request.v1"
ATTACH_EVIDENCE_SCHEMA = "aios.verified-legacy-attachment-evidence.v1"
ATTACH_RECEIPT_SCHEMA = "aios.mission-legacy-attachment-receipt.v1"
ATTACHMENT_KIND = "LEGACY_EXECUTION_SUBJECT"
RELATION_TYPE = "RECOVERY"
SOURCE_KIND = "agent-task-bridge-job"
REQUIRED_READBACK_ROLES = frozenset(
{"agent_task", "admission", "owner_registry", "bridge_job"}
)
REQUIRED_C04_EVIDENCE = frozenset(
{
"verified legacy ownership and acceptance relevance",
"unchanged task/job/idempotency/reconciliation identity",
"authoritative attach readback",
}
)
_SHA256_RE = re.compile(r"^[0-9a-f]{64}$")
def _canonical(value: object) -> str:
try:
return json.dumps(
value,
ensure_ascii=False,
sort_keys=True,
separators=(",", ":"),
allow_nan=False,
)
except (TypeError, ValueError) as exc:
raise ValidationError("legacy attachment value must be JSON data") from exc
def _object(value: object, fields: frozenset[str], name: str) -> dict[str, object]:
if not isinstance(value, Mapping) or set(value) != fields:
raise ValidationError(f"invalid {name} fields")
return json.loads(_canonical(dict(value)))
def _identifier(value: object, name: str, *, maximum: int = 256) -> str:
if (
not isinstance(value, str)
or not value
or len(value) > maximum
or any(ord(character) < 33 for character in value)
):
raise ValidationError(f"invalid {name}")
return value
def _integer(value: object, name: str) -> int:
if type(value) is not int or value < 0:
raise ValidationError(f"invalid {name}")
return value
def _utc(value: object, name: str) -> str:
if not isinstance(value, str) or not value:
raise ValidationError(f"invalid {name}")
try:
parsed = datetime.fromisoformat(value.replace("Z", "+00:00"))
except ValueError as exc:
raise ValidationError(f"invalid {name}") from exc
if parsed.tzinfo is None or parsed.utcoffset() is None:
raise ValidationError(f"invalid {name}")
return parsed.astimezone(timezone.utc).isoformat().replace("+00:00", "Z")
@dataclass(frozen=True)
class AuthorityReadbackRef:
"""A non-secret pointer and digest issued by one independent authority."""
role: str
authority: str
readback_ref: str
canonical_sha256: str
observed_at: str
revision: int | None = None
def to_dict(self) -> dict[str, object]:
if self.role not in REQUIRED_READBACK_ROLES:
raise ValidationError("invalid legacy authority readback role")
authority = _identifier(self.authority, "readback authority")
reference = _identifier(self.readback_ref, "readback ref", maximum=1024)
if not _SHA256_RE.fullmatch(self.canonical_sha256):
raise ValidationError("invalid readback canonical SHA-256")
observed = _utc(self.observed_at, "readback observed_at")
revision = self.revision
if revision is not None:
revision = _integer(revision, "readback revision")
return {
"role": self.role,
"authority": authority,
"readback_ref": reference,
"canonical_sha256": self.canonical_sha256,
"observed_at": observed,
"revision": revision,
}
@dataclass(frozen=True)
class VerifiedLegacyAttachmentEvidence:
"""Typed output of a configured authority reader, never a caller payload."""
schema: str
source_kind: str
source_task_id: str
source_task_created_at: str
source_task_status: str
source_task_revision: int
source_task_attempt: int
source_task_max_attempts: int
external_job_id: str
internal_job_id: str
internal_idempotency_key: str
job_branch: str
job_state: str
job_revision: int
side_effect_state: str
grant_id: str
grant_digest: str
execution_digest: str
epoch_id: str
epoch_digest: str
historical_signer_id: str
historical_signer_fingerprint: str
registry_generation: int
signature_verified: bool
epoch_currently_valid: bool
readbacks: Sequence[AuthorityReadbackRef]
def to_dict(self) -> dict[str, object]:
if self.schema != ATTACH_EVIDENCE_SCHEMA or self.source_kind != SOURCE_KIND:
raise ValidationError("invalid legacy attachment evidence schema")
if self.signature_verified is not True or self.epoch_currently_valid is not False:
raise ConflictError("legacy historical Owner evidence is not verified-expired")
values = {
"source_task_id": self.source_task_id,
"source_task_status": self.source_task_status,
"external_job_id": self.external_job_id,
"internal_job_id": self.internal_job_id,
"internal_idempotency_key": self.internal_idempotency_key,
"job_branch": self.job_branch,
"job_state": self.job_state,
"side_effect_state": self.side_effect_state,
"grant_id": self.grant_id,
"historical_signer_id": self.historical_signer_id,
"historical_signer_fingerprint": self.historical_signer_fingerprint,
"epoch_id": self.epoch_id,
}
normalized = {
key: _identifier(value, key, maximum=1024)
for key, value in values.items()
}
for name, value in {
"grant_digest": self.grant_digest,
"execution_digest": self.execution_digest,
"epoch_digest": self.epoch_digest,
}.items():
if not _SHA256_RE.fullmatch(value):
raise ValidationError(f"invalid {name}")
normalized[name] = value
task_revision = _integer(self.source_task_revision, "source_task_revision")
task_attempt = _integer(self.source_task_attempt, "source_task_attempt")
task_max_attempts = _integer(
self.source_task_max_attempts, "source_task_max_attempts"
)
if task_attempt > task_max_attempts:
raise ConflictError("legacy task attempt exceeds max attempts")
job_revision = _integer(self.job_revision, "job_revision")
registry_generation = _integer(
self.registry_generation, "registry_generation"
)
readbacks = [value.to_dict() for value in self.readbacks]
roles = [str(value["role"]) for value in readbacks]
if len(roles) != len(set(roles)) or set(roles) != REQUIRED_READBACK_ROLES:
raise ValidationError("legacy evidence requires four authority readbacks")
readbacks.sort(key=lambda value: str(value["role"]))
return {
"schema": ATTACH_EVIDENCE_SCHEMA,
"source_kind": SOURCE_KIND,
**normalized,
"source_task_created_at": _utc(
self.source_task_created_at, "source_task_created_at"
),
"source_task_revision": task_revision,
"source_task_attempt": task_attempt,
"source_task_max_attempts": task_max_attempts,
"job_revision": job_revision,
"registry_generation": registry_generation,
"signature_verified": True,
"epoch_currently_valid": False,
"readbacks": readbacks,
}
class LegacyAttachmentEvidenceProvider(Protocol):
"""Configured server-side reader for the four independent authority sources."""
def read_verified(
self,
*,
mission_id: str,
root_task_id: str,
anchor_task_id: str,
source_kind: str,
source_task_id: str,
external_job_id: str,
original_acceptance_digest: str,
) -> VerifiedLegacyAttachmentEvidence: ...
def normalize_attach_request(value: object) -> dict[str, object]:
body = _object(
value,
frozenset({"schema", "anchor_task_id", "source", "expected"}),
"legacy attachment request",
)
if body["schema"] != ATTACH_REQUEST_SCHEMA:
raise ValidationError("invalid legacy attachment request schema")
source = _object(
body["source"],
frozenset({"kind", "task_id", "external_job_id"}),
"legacy attachment source",
)
if source["kind"] != SOURCE_KIND:
raise ValidationError("invalid legacy attachment source kind")
expected = _object(
body["expected"],
frozenset(
{
"root_revision",
"anchor_revision",
"agent_task_revision",
"agent_task_attempt",
"internal_job_id",
"job_revision",
"internal_idempotency_key",
}
),
"legacy attachment expected",
)
return {
"schema": ATTACH_REQUEST_SCHEMA,
"anchor_task_id": _identifier(body["anchor_task_id"], "anchor_task_id"),
"source": {
"kind": SOURCE_KIND,
"task_id": _identifier(source["task_id"], "source task_id"),
"external_job_id": _identifier(
source["external_job_id"], "external_job_id"
),
},
"expected": {
"root_revision": _integer(expected["root_revision"], "root_revision"),
"anchor_revision": _integer(
expected["anchor_revision"], "anchor_revision"
),
"agent_task_revision": _integer(
expected["agent_task_revision"], "agent_task_revision"
),
"agent_task_attempt": _integer(
expected["agent_task_attempt"], "agent_task_attempt"
),
"internal_job_id": _identifier(
expected["internal_job_id"], "internal_job_id"
),
"job_revision": _integer(expected["job_revision"], "job_revision"),
"internal_idempotency_key": _identifier(
expected["internal_idempotency_key"], "internal_idempotency_key"
),
},
}
def validate_c04_acceptance(
acceptance: Mapping[str, object], *, source_task_id: str, external_job_id: str
) -> None:
candidate = acceptance.get("legacy_candidate")
if (
not isinstance(candidate, Mapping)
or candidate.get("task_id") != source_task_id
or candidate.get("job_id") != external_job_id
):
raise ConflictError("mission legacy candidate authority mismatch")
conditions = acceptance.get("conditions")
if not isinstance(conditions, list):
raise ConflictError("mission C04 acceptance condition is unavailable")
matches = [
condition
for condition in conditions
if isinstance(condition, Mapping) and condition.get("key") == "C04"
]
if len(matches) != 1:
raise ConflictError("mission C04 acceptance condition is unavailable")
condition = matches[0]
evidence = condition.get("required_evidence")
if (
condition.get("purpose") != "LEGACY-LINEAGE-MIGRATION"
or not isinstance(evidence, list)
or not REQUIRED_C04_EVIDENCE.issubset(
value for value in evidence if isinstance(value, str)
)
):
raise ConflictError("mission C04 acceptance contract mismatch")
def validate_evidence_matches_request(
evidence: Mapping[str, object], request: Mapping[str, object]
) -> None:
source = request["source"]
expected = request["expected"]
assert isinstance(source, Mapping) and isinstance(expected, Mapping)
pairs = {
"source_kind": source["kind"],
"source_task_id": source["task_id"],
"external_job_id": source["external_job_id"],
"source_task_revision": expected["agent_task_revision"],
"source_task_attempt": expected["agent_task_attempt"],
"internal_job_id": expected["internal_job_id"],
"job_revision": expected["job_revision"],
"internal_idempotency_key": expected["internal_idempotency_key"],
}
mismatches = [name for name, value in pairs.items() if evidence.get(name) != value]
if mismatches:
raise ConflictError(
"legacy provider evidence mismatch: " + ",".join(sorted(mismatches))
)
class LegacyAttachmentLedger:
"""Append-only persistence owned by one Mission authority store."""
def __init__(self, store) -> None:
self.store = store
with store._transaction():
store.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_mission_legacy_attachments (
attachment_id TEXT PRIMARY KEY,
mission_id TEXT NOT NULL,
root_task_id TEXT NOT NULL,
anchor_task_id TEXT NOT NULL,
authenticated_identity TEXT NOT NULL,
owner_principal_id TEXT NOT NULL,
attachment_kind TEXT NOT NULL,
relation_type TEXT NOT NULL,
attachment_sequence INTEGER NOT NULL,
display_name TEXT NOT NULL,
source_kind TEXT NOT NULL,
source_task_id TEXT NOT NULL,
external_job_id TEXT NOT NULL,
internal_job_id TEXT NOT NULL,
internal_idempotency_key TEXT NOT NULL,
request_digest TEXT NOT NULL,
idempotency_key TEXT NOT NULL,
evidence_digest TEXT NOT NULL,
receipt_digest TEXT NOT NULL,
receipt_json TEXT NOT NULL,
created_at REAL NOT NULL,
UNIQUE (authenticated_identity, idempotency_key),
UNIQUE (source_kind, source_task_id),
UNIQUE (source_kind, external_job_id),
UNIQUE (mission_id, anchor_task_id, attachment_sequence),
UNIQUE (mission_id, anchor_task_id, source_kind, source_task_id, external_job_id)
)
"""
)
store.conn.execute(
"CREATE INDEX IF NOT EXISTS ix_mission_legacy_attachments "
"ON task_mission_legacy_attachments(mission_id, anchor_task_id)"
)
store.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_mission_acceptance_child_bindings (
mission_id TEXT NOT NULL,
condition_key TEXT NOT NULL,
child_task_id TEXT NOT NULL UNIQUE,
original_acceptance_digest TEXT NOT NULL,
binding_digest TEXT NOT NULL,
created_at REAL NOT NULL,
PRIMARY KEY (mission_id, condition_key)
)
"""
)
def acceptance_child_binding(self, mission_id: str, condition_key: str):
return self.store.conn.execute(
"SELECT * FROM task_mission_acceptance_child_bindings "
"WHERE mission_id=? AND condition_key=?",
(mission_id, condition_key),
).fetchone()
def bind_acceptance_child(
self,
*,
mission_id: str,
condition_key: str,
child_task_id: str,
original_acceptance_digest: str,
binding_digest: str,
created_at: float,
) -> None:
existing = self.acceptance_child_binding(mission_id, condition_key)
if existing is not None:
if (
existing["child_task_id"] != child_task_id
or existing["original_acceptance_digest"]
!= original_acceptance_digest
or existing["binding_digest"] != binding_digest
):
raise ConflictError(
"mission acceptance condition already has a canonical child"
)
return
child_binding = self.store.conn.execute(
"SELECT mission_id,condition_key FROM "
"task_mission_acceptance_child_bindings WHERE child_task_id=?",
(child_task_id,),
).fetchone()
if child_binding is not None:
raise ConflictError("mission child already belongs to an acceptance condition")
self.store.conn.execute(
"INSERT INTO task_mission_acceptance_child_bindings "
"(mission_id,condition_key,child_task_id,original_acceptance_digest,"
"binding_digest,created_at) VALUES(?,?,?,?,?,?)",
(
mission_id,
condition_key,
child_task_id,
original_acceptance_digest,
binding_digest,
created_at,
),
)
def by_idempotency(self, identity: str, key: str):
return self.store.conn.execute(
"SELECT * FROM task_mission_legacy_attachments "
"WHERE authenticated_identity=? AND idempotency_key=?",
(identity, key),
).fetchone()
def by_source(self, source_kind: str, task_id: str, external_job_id: str):
return self.store.conn.execute(
"SELECT * FROM task_mission_legacy_attachments "
"WHERE source_kind=? AND (source_task_id=? OR external_job_id=?)",
(source_kind, task_id, external_job_id),
).fetchone()
def by_id(self, mission_id: str, attachment_id: str):
row = self.store.conn.execute(
"SELECT * FROM task_mission_legacy_attachments "
"WHERE mission_id=? AND attachment_id=?",
(mission_id, attachment_id),
).fetchone()
if row is None:
raise NotFoundError("legacy attachment not found")
return row
def next_sequence(self, mission_id: str, anchor_task_id: str) -> int:
row = self.store.conn.execute(
"SELECT COALESCE(MAX(attachment_sequence),0)+1 "
"FROM task_mission_legacy_attachments "
"WHERE mission_id=? AND anchor_task_id=?",
(mission_id, anchor_task_id),
).fetchone()
sequence = int(row[0])
if not 1 <= sequence <= 99:
raise ConflictError("legacy attachment sequence exhausted")
return sequence
def insert(
self,
*,
receipt: Mapping[str, object],
authenticated_identity: str,
idempotency_key: str,
request_digest: str,
evidence_digest: str,
created_at: float,
) -> None:
self.store.conn.execute(
"INSERT INTO task_mission_legacy_attachments "
"(attachment_id,mission_id,root_task_id,anchor_task_id,"
"authenticated_identity,owner_principal_id,attachment_kind,"
"relation_type,attachment_sequence,display_name,source_kind,"
"source_task_id,external_job_id,internal_job_id,"
"internal_idempotency_key,request_digest,idempotency_key,"
"evidence_digest,receipt_digest,receipt_json,created_at) "
"VALUES(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)",
(
receipt["attachment_id"], receipt["mission_id"],
receipt["root_task_id"], receipt["anchor_task_id"],
authenticated_identity, receipt["owner_principal_id"],
receipt["attachment_kind"], receipt["relation_type"],
receipt["attachment_sequence"], receipt["display_name"],
receipt["source"]["source_kind"],
receipt["source"]["source_task_id"],
receipt["source"]["external_job_id"],
receipt["source"]["internal_job_id"],
receipt["source"]["internal_idempotency_key"],
request_digest, idempotency_key, evidence_digest,
receipt["receipt_digest"], _canonical(receipt), created_at,
),
)
@staticmethod
def receipt(row) -> dict[str, object]:
try:
value = json.loads(row["receipt_json"])
except (TypeError, ValueError, json.JSONDecodeError) as exc:
raise ConflictError("persisted legacy attachment receipt is invalid") from exc
if (
not isinstance(value, dict)
or value.get("schema") != ATTACH_RECEIPT_SCHEMA
or value.get("receipt_digest") != row["receipt_digest"]
):
raise ConflictError("persisted legacy attachment receipt is invalid")
digest_payload = dict(value)
digest_payload.pop("receipt_digest", None)
digest_payload.pop("idempotent", None)
if compute_sha256_hex(digest_payload) != row["receipt_digest"]:
raise ConflictError("persisted legacy attachment receipt digest mismatch")
return value
@@ -0,0 +1,846 @@
"""Actual TaskPlane binding for task identity, lineage and root review.
All writes use the existing :class:`TaskPlaneStore` connection and transaction
fence. Request metadata is never used as an authorization source.
"""
from __future__ import annotations
import json
from dataclasses import dataclass
from datetime import datetime, timezone
from typing import Callable, Mapping, Sequence
if __package__ and "." in __package__: # app.task_plane package import
from ..aios_task_lineage import (
AcceptanceCondition,
AcceptanceEvaluator,
NextLegalAction,
RelationType,
RootAcceptanceSnapshot,
RootReviewRecord,
TaskRecord as LineageTaskRecord,
lineage_from_metadata,
normalize_goal_slug,
root_review_from_record,
)
from ..aios_task_lineage_integration import (
DISPLAY_NAME_METADATA_KEY,
LINEAGE_METADATA_KEY,
LINEAGE_REQUEST_DIGEST_KEY,
NewTaskIdentityRequest,
PERSISTED_DISPATCH_STATUSES,
RootActionDispatchReceipt,
TaskAuthoritySnapshot,
TaskIdentityLineageService,
TerminalReviewOutcome,
_as_task_record,
)
from ..aios_task_lineage_taskplane_adapter import (
AuthenticatedTaskContext,
InProcessTaskPlaneLineageAdapter,
TaskPlaneServiceBindings,
)
from ..aios_task_lineage_hooks import TaskLineageLifecycleHooks
else: # top-level task_plane import used by direct-module test harnesses
from aios_task_lineage import (
AcceptanceCondition,
AcceptanceEvaluator,
NextLegalAction,
RelationType,
RootAcceptanceSnapshot,
RootReviewRecord,
TaskRecord as LineageTaskRecord,
lineage_from_metadata,
normalize_goal_slug,
root_review_from_record,
)
from aios_task_lineage_integration import (
DISPLAY_NAME_METADATA_KEY,
LINEAGE_METADATA_KEY,
LINEAGE_REQUEST_DIGEST_KEY,
NewTaskIdentityRequest,
PERSISTED_DISPATCH_STATUSES,
RootActionDispatchReceipt,
TaskAuthoritySnapshot,
TaskIdentityLineageService,
TerminalReviewOutcome,
_as_task_record,
)
from aios_task_lineage_taskplane_adapter import (
AuthenticatedTaskContext,
InProcessTaskPlaneLineageAdapter,
TaskPlaneServiceBindings,
)
from aios_task_lineage_hooks import TaskLineageLifecycleHooks
from .delegation import DelegationEngine
from .errors import ConflictError, ForbiddenError, NotFoundError, ValidationError
from .models import TaskPlan, TaskRecord, canonical_json, compute_sha256_hex
from .store import TaskPlaneStore
RESERVED_LINEAGE_METADATA_KEYS = frozenset(
{LINEAGE_METADATA_KEY, DISPLAY_NAME_METADATA_KEY, LINEAGE_REQUEST_DIGEST_KEY}
)
PROVEN_TERMINAL_STATES = frozenset({"COMPLETED", "FAILED_FINAL", "CANCELLED"})
@dataclass(frozen=True)
class PrincipalLineageBinding:
"""Server-owned projection of one authenticated service identity."""
authenticated_identity: str
owner_principal_id: str
mission_id: str
preferred_owner_alias: str | None
root_acceptance_ref: str
def __post_init__(self) -> None:
for name in (
"authenticated_identity",
"owner_principal_id",
"mission_id",
"root_acceptance_ref",
):
if not str(getattr(self, name)).strip():
raise ValueError(f"{name} is required")
class TaskLineagePolicy:
"""Immutable service policy for authenticated identity and acceptance."""
def __init__(
self,
*,
principal_bindings: Sequence[PrincipalLineageBinding],
acceptance_evaluators: Mapping[str, AcceptanceEvaluator],
acceptance_evaluator_resolver: Callable[
[str], AcceptanceEvaluator | None
]
| None = None,
) -> None:
self._bindings: dict[str, PrincipalLineageBinding] = {}
for binding in principal_bindings:
if binding.authenticated_identity in self._bindings:
raise ValueError("duplicate authenticated lineage identity")
self._bindings[binding.authenticated_identity] = binding
self._evaluators = dict(acceptance_evaluators)
self._acceptance_evaluator_resolver = acceptance_evaluator_resolver
def with_acceptance_evaluator_resolver(
self, resolver: Callable[[str], AcceptanceEvaluator | None]
) -> "TaskLineagePolicy":
previous = self._acceptance_evaluator_resolver
def combined(reference: str) -> AcceptanceEvaluator | None:
resolved = resolver(reference)
if resolved is not None:
return resolved
return previous(reference) if previous is not None else None
return TaskLineagePolicy(
principal_bindings=tuple(self._bindings.values()),
acceptance_evaluators=self._evaluators,
acceptance_evaluator_resolver=combined,
)
def resolve_principal(self, authenticated_identity: str) -> PrincipalLineageBinding:
try:
return self._bindings[authenticated_identity]
except KeyError as exc:
raise ForbiddenError("authenticated identity has no lineage policy binding") from exc
def _acceptance_evaluator(self, reference: str) -> AcceptanceEvaluator | None:
evaluator = self._evaluators.get(reference)
if evaluator is None and self._acceptance_evaluator_resolver is not None:
evaluator = self._acceptance_evaluator_resolver(reference)
return evaluator
def assert_acceptance_ref_registered(self, reference: str) -> None:
if self._acceptance_evaluator(reference) is None:
raise ForbiddenError("root acceptance policy is not registered")
def evaluate_acceptance(
self,
root: LineageTaskRecord,
descendants: Sequence[LineageTaskRecord],
) -> RootAcceptanceSnapshot:
reference = root.lineage.root_acceptance_ref
evaluator = self._acceptance_evaluator(reference)
if evaluator is None:
raise ForbiddenError("root acceptance policy is not registered")
snapshot = evaluator(root, descendants)
if not isinstance(snapshot, RootAcceptanceSnapshot) or not snapshot.conditions:
return RootAcceptanceSnapshot(
(
AcceptanceCondition(
key="root-acceptance-proof-unavailable",
satisfied=False,
owner_boundary=True,
),
)
)
normalized: list[AcceptanceCondition] = []
for condition in snapshot.conditions:
if condition.satisfied and not condition.evidence_refs:
normalized.append(
AcceptanceCondition(
key=condition.key,
satisfied=False,
owner_boundary=True,
)
)
else:
normalized.append(condition)
return RootAcceptanceSnapshot(tuple(normalized))
class TaskPlaneLineageStoreBinding:
"""Official in-process read/write surface over the existing TaskPlane DB."""
def __init__(self, store: TaskPlaneStore) -> None:
self.store = store
def service_bindings(self) -> TaskPlaneServiceBindings:
return TaskPlaneServiceBindings(
get_task_snapshot=self.get_task_snapshot,
list_descendant_snapshots=self.list_descendant_snapshots,
reserve_sibling_sequence=self.reserve_sibling_sequence,
compare_and_swap_task_metadata=self.compare_and_swap_task_metadata,
get_root_review_record=self.get_root_review_record,
compare_and_swap_root_review_record=self.compare_and_swap_root_review_record,
get_root_action_dispatch=self.get_root_action_dispatch,
dispatch_root_action=self.dispatch_root_action,
)
def _terminal_proof(self, task: TaskRecord) -> Mapping[str, object] | None:
row = self.store.conn.execute(
"SELECT * FROM task_terminal_proofs WHERE task_id=?", (task.task_id,)
).fetchone()
if row is None or task.status not in PROVEN_TERMINAL_STATES:
return None
claim = self.store.conn.execute(
"SELECT status,result_payload FROM claims WHERE claim_id=? AND task_id=?",
(row["claim_id"], task.task_id),
).fetchone()
if claim is None or claim["status"] != "COMPLETED" or not claim["result_payload"]:
return None
try:
result = json.loads(claim["result_payload"])
refs = tuple(json.loads(row["evidence_refs_json"]))
except (TypeError, ValueError, json.JSONDecodeError):
return None
if result.get("result_digest") != row["result_digest"]:
return None
if task.revision < int(row["task_revision"]):
return None
evidence_ids = {entry["evidence_id"] for entry in self.store.list_evidence(task.task_id)}
if not set(refs).issubset(evidence_ids):
return None
if not self.store.verify_event_chain(task.task_id):
return None
terminal_event = any(
event["type"] in {"TASK_COMPLETED", "TASK_FAILED"}
and event["data"].get("claim_id") == row["claim_id"]
for event in self.store.list_events(task.task_id)
)
if not terminal_event or not self.store.verify_evidence_integrity(task.task_id):
return None
return dict(row)
def get_task_snapshot(self, task_id: str) -> TaskAuthoritySnapshot:
task = self.store.get_task(task_id)
if not task.owner_principal_id or not task.mission_id:
raise ForbiddenError("task lacks server-owned lineage authority binding")
proof = self._terminal_proof(task)
return TaskAuthoritySnapshot(
task_id=task.task_id,
revision=task.revision,
status=task.status,
owner_principal_id=task.owner_principal_id,
mission_id=task.mission_id,
created_at=datetime.fromtimestamp(task.created_at, tz=timezone.utc),
metadata=dict(task.task_plan.metadata),
evidence_refs=(
tuple(json.loads(proof["evidence_refs_json"])) if proof else ()
),
terminal_evidence_proven=proof is not None,
)
def list_descendant_snapshots(
self, root_task_id: str
) -> tuple[TaskAuthoritySnapshot, ...]:
with self.store._lock:
rows = self.store.conn.execute(
"SELECT task_id FROM tasks WHERE root_task_id=? AND task_id<>? "
"ORDER BY created_at,task_id",
(root_task_id, root_task_id),
).fetchall()
return tuple(self.get_task_snapshot(row["task_id"]) for row in rows)
def reserve_sibling_sequence(
self, parent_task_id: str, relation_type: RelationType, child_task_id: str
) -> int:
with self.store._transaction():
existing = self.store.conn.execute(
"SELECT parent_task_id,relation_type,sibling_sequence "
"FROM task_lineage_sequences WHERE child_task_id=?",
(child_task_id,),
).fetchone()
if existing is not None:
if (
existing["parent_task_id"] != parent_task_id
or existing["relation_type"] != relation_type.value
):
raise ConflictError("child sibling reservation conflicts with lineage")
return int(existing["sibling_sequence"])
row = self.store.conn.execute(
"SELECT COALESCE(MAX(sibling_sequence),0)+1 AS next_sequence "
"FROM task_lineage_sequences WHERE parent_task_id=? AND relation_type=?",
(parent_task_id, relation_type.value),
).fetchone()
sequence = int(row["next_sequence"])
if sequence > 99:
raise ConflictError("lineage sibling sequence exhausted")
self.store.conn.execute(
"INSERT INTO task_lineage_sequences "
"(parent_task_id,relation_type,child_task_id,sibling_sequence,created_at) "
"VALUES(?,?,?,?,?)",
(parent_task_id, relation_type.value, child_task_id, sequence, datetime.now(tz=timezone.utc).timestamp()),
)
return sequence
def compare_and_swap_task_metadata(
self, task_id: str, expected_revision: int, metadata_patch: Mapping[str, object]
) -> bool:
if not set(metadata_patch).issubset(RESERVED_LINEAGE_METADATA_KEYS):
raise ForbiddenError("lineage CAS contains non-lineage metadata")
with self.store._transaction():
task = self.store.get_task(task_id)
if task.revision != expected_revision:
return False
lineage_value = metadata_patch.get(LINEAGE_METADATA_KEY)
if isinstance(lineage_value, Mapping):
lineage = lineage_from_metadata(lineage_value)
if lineage.relation_type != RelationType.ROOT:
# The root check and child metadata CAS share this BEGIN
# IMMEDIATE transaction, closing the legacy attach race.
self.store.assert_root_open_for_child(lineage.root_task_id)
if task.status in PROVEN_TERMINAL_STATES:
raise ConflictError("terminal task lineage metadata is immutable")
payload = task.task_plan.to_dict(redact=False)
metadata = dict(payload.get("metadata") or {})
for key, value in metadata_patch.items():
if key in metadata and metadata[key] != value:
raise ConflictError("authoritative lineage metadata conflict")
metadata[key] = value
payload["metadata"] = metadata
now = datetime.now(tz=timezone.utc).timestamp()
updated = self.store.conn.execute(
"UPDATE tasks SET task_json=?,revision=revision+1,updated_at=? "
"WHERE task_id=? AND revision=?",
(json.dumps(payload, ensure_ascii=False), now, task_id, expected_revision),
)
if updated.rowcount != 1:
return False
self.store.append_event(
task_id,
"TASK_LINEAGE_BOUND",
{"metadata_keys": sorted(metadata_patch)},
subject=f"task/{task_id}",
)
return True
def record_terminal_proof(
self,
*,
task_id: str,
claim_id: str,
result_digest: str,
evidence_refs: Sequence[str],
) -> None:
with self.store._transaction():
task = self.store.get_task(task_id)
if task.status not in PROVEN_TERMINAL_STATES:
raise ConflictError("terminal proof requires a final task status")
claim = self.store.conn.execute(
"SELECT status,result_payload FROM claims WHERE claim_id=? AND task_id=?",
(claim_id, task_id),
).fetchone()
if claim is None or claim["status"] != "COMPLETED" or not claim["result_payload"]:
raise ConflictError("terminal proof requires a completed claim")
stored_result = json.loads(claim["result_payload"])
if stored_result.get("result_digest") != result_digest:
raise ConflictError("terminal proof result digest mismatch")
refs = tuple(str(value) for value in evidence_refs)
evidence_ids = {entry["evidence_id"] for entry in self.store.list_evidence(task_id)}
if not set(refs).issubset(evidence_ids):
raise ConflictError("terminal proof evidence reference mismatch")
existing = self.store.conn.execute(
"SELECT * FROM task_terminal_proofs WHERE task_id=?", (task_id,)
).fetchone()
expected = (claim_id, result_digest, canonical_json(list(refs)))
if existing is not None:
actual = (
existing["claim_id"],
existing["result_digest"],
canonical_json(json.loads(existing["evidence_refs_json"])),
)
if actual != expected:
raise ConflictError("terminal proof replay conflict")
return
if not self.store.verify_event_chain(task_id) or not self.store.verify_evidence_integrity(task_id):
raise ConflictError("terminal proof journal integrity failed")
self.store.conn.execute(
"INSERT INTO task_terminal_proofs "
"(task_id,claim_id,task_revision,result_digest,evidence_refs_json,created_at) "
"VALUES(?,?,?,?,?,?)",
(
task_id,
claim_id,
task.revision,
result_digest,
json.dumps(list(refs), ensure_ascii=False),
datetime.now(tz=timezone.utc).timestamp(),
),
)
def get_root_review_record(self, review_id: str) -> Mapping[str, object] | None:
row = self.store.conn.execute(
"SELECT record_json FROM task_root_reviews WHERE review_id=?", (review_id,)
).fetchone()
return None if row is None else json.loads(row["record_json"])
def get_root_review_for_trigger(
self, trigger_child_task_id: str
) -> Mapping[str, object] | None:
row = self.store.conn.execute(
"SELECT record_json FROM task_root_reviews "
"WHERE trigger_child_task_id=? ORDER BY created_at LIMIT 1",
(trigger_child_task_id,),
).fetchone()
return None if row is None else json.loads(row["record_json"])
def compare_and_swap_root_review_record(
self,
root_task_id: str,
expected_root_revision: int,
review_record: Mapping[str, object],
) -> bool:
with self.store._transaction():
review_id = str(review_record.get("review_id") or "")
trigger_id = str(review_record.get("trigger_child_task_id") or "")
if (
review_record.get("root_task_id") != root_task_id
or int(review_record.get("root_revision_before", -1)) != expected_root_revision
or int(review_record.get("root_revision_after", -1)) != expected_root_revision + 1
):
raise ConflictError("root review authority binding mismatch")
existing = self.get_root_review_record(review_id)
if existing is not None:
if canonical_json(existing) != canonical_json(dict(review_record)):
raise ConflictError("root review replay conflict")
return True
trigger = self.get_task_snapshot(trigger_id)
if not trigger.terminal_evidence_proven:
raise ConflictError("root review trigger is not proven terminal")
trigger_revision = trigger.revision
frozen = self.store.conn.execute(
"SELECT review_id FROM task_root_reviews "
"WHERE trigger_child_task_id=? AND trigger_child_revision=?",
(trigger_id, trigger_revision),
).fetchone()
if frozen is not None:
raise ConflictError("terminal trigger already has a different root review")
root = self.store.get_task(root_task_id)
if root.revision != expected_root_revision:
return False
now = datetime.now(tz=timezone.utc).timestamp()
self.store.conn.execute(
"INSERT INTO task_root_reviews "
"(review_id,root_task_id,trigger_child_task_id,trigger_child_revision,"
"root_revision_before,root_revision_after,record_json,created_at) "
"VALUES(?,?,?,?,?,?,?,?)",
(
review_id,
root_task_id,
trigger_id,
trigger_revision,
expected_root_revision,
expected_root_revision + 1,
json.dumps(dict(review_record), ensure_ascii=False, sort_keys=True),
now,
),
)
updated = self.store.conn.execute(
"UPDATE tasks SET revision=revision+1,updated_at=? "
"WHERE task_id=? AND revision=?",
(now, root_task_id, expected_root_revision),
)
if updated.rowcount != 1:
raise ConflictError("root review CAS lost after reservation")
self.store.append_event(
root_task_id,
"ROOT_REVIEW",
{
"review_id": review_id,
"trigger_child_task_id": trigger_id,
"next_legal_action": review_record.get("next_legal_action"),
"root_complete": bool(review_record.get("root_complete")),
},
subject=f"task/{root_task_id}",
)
return True
def get_root_action_dispatch(self, review_id: str) -> RootActionDispatchReceipt | None:
row = self.store.conn.execute(
"SELECT * FROM task_root_action_dispatches WHERE review_id=?", (review_id,)
).fetchone()
if row is None:
return None
return RootActionDispatchReceipt(
review_id=row["review_id"],
root_task_id=row["root_task_id"],
action=NextLegalAction(row["action"]),
idempotency_key=row["idempotency_key"],
status=row["status"],
mission_id=row["mission_id"],
root_revision=int(row["root_revision"]),
plan_digest=row["plan_digest"],
execution_key=row["execution_key"],
)
def dispatch_root_action(
self, review: RootReviewRecord, idempotency_key: str
) -> RootActionDispatchReceipt:
with self.store._transaction():
persisted = self.get_root_review_record(review.review_id)
if persisted is None:
raise ConflictError("root action requires a persisted root review")
existing = self.get_root_action_dispatch(review.review_id)
if existing is not None:
if (
existing.root_task_id != review.root_task_id
or existing.action != review.next_legal_action
or existing.idempotency_key != idempotency_key
):
raise ConflictError("root action dispatch replay conflict")
return existing
now = datetime.now(tz=timezone.utc).timestamp()
self.store.conn.execute(
"INSERT INTO task_root_action_dispatches "
"(review_id,root_task_id,action,idempotency_key,status,created_at,"
"mission_id,root_revision,updated_at) VALUES(?,?,?,?,?,?,?,?,?)",
(
review.review_id,
review.root_task_id,
review.next_legal_action.value,
idempotency_key,
"PENDING",
now,
self.store.get_task(review.root_task_id).mission_id,
review.root_revision_after,
now,
),
)
self.store.append_event(
review.root_task_id,
"ROOT_NEXT_ACTION_PENDING",
{
"review_id": review.review_id,
"action": review.next_legal_action.value,
"idempotency_key": idempotency_key,
},
subject=f"task/{review.root_task_id}",
)
return self.get_root_action_dispatch(review.review_id) # type: ignore[return-value]
class TaskPlaneLineageCoordinator:
"""Coordinates lineage with real TaskPlane admission and completion paths."""
def __init__(self, store: TaskPlaneStore, policy: TaskLineagePolicy) -> None:
self.store = store
self.policy = policy
self.binding = TaskPlaneLineageStoreBinding(store)
self.adapter = InProcessTaskPlaneLineageAdapter(
self.binding.service_bindings(), readback=self.binding
)
self.service = TaskIdentityLineageService(self.adapter)
self.hooks = TaskLineageLifecycleHooks(self.adapter)
@staticmethod
def _request_digest(payload: Mapping[str, object]) -> str:
return compute_sha256_hex(dict(payload))
@staticmethod
def _reject_reserved_metadata(plan: TaskPlan) -> None:
if set(plan.metadata).intersection(RESERVED_LINEAGE_METADATA_KEYS):
raise ValidationError("lineage metadata is server-owned")
@staticmethod
def _goal_slug(plan: TaskPlan) -> str:
for candidate in (plan.intent, plan.task_type, f"TASK-{plan.task_id}"):
if not candidate:
continue
try:
return normalize_goal_slug(str(candidate))
except ValueError:
continue
raise ValidationError("task goal cannot be normalized for lineage")
def _existing_by_idempotency(self, key: str | None) -> TaskRecord | None:
if not key:
return None
row = self.store.conn.execute(
"SELECT task_id FROM tasks WHERE idempotency_key=?", (key,)
).fetchone()
return None if row is None else self.store.get_task(row["task_id"])
@staticmethod
def _verify_replay(
task: TaskRecord,
*,
request_digest: str,
principal: PrincipalLineageBinding,
parent_task_id: str | None,
) -> TaskRecord:
metadata = task.task_plan.metadata
if (
metadata.get(LINEAGE_REQUEST_DIGEST_KEY) != request_digest
or task.owner_principal_id != principal.owner_principal_id
or task.mission_id != principal.mission_id
or task.task_plan.parent_task_id != parent_task_id
):
raise ConflictError("lineage admission replay conflict")
return task
def create_root(
self,
*,
plan: TaskPlan,
authenticated_identity: str,
request_digest: str,
idempotency_key: str | None,
) -> TaskRecord:
principal = self.policy.resolve_principal(authenticated_identity)
return self.create_root_for_principal(
plan=plan,
principal=principal,
request_digest=request_digest,
idempotency_key=idempotency_key,
)
def create_root_for_principal(
self,
*,
plan: TaskPlan,
principal: PrincipalLineageBinding,
request_digest: str,
idempotency_key: str | None,
created_at: datetime | None = None,
) -> TaskRecord:
self.policy.assert_acceptance_ref_registered(principal.root_acceptance_ref)
self._reject_reserved_metadata(plan)
with self.store._transaction():
existing = self._existing_by_idempotency(idempotency_key)
if existing is not None:
return self._verify_replay(
existing,
request_digest=request_digest,
principal=principal,
parent_task_id=None,
)
created_at = created_at or datetime.now(tz=timezone.utc)
if created_at.tzinfo is None or created_at.utcoffset() is None:
raise ValidationError("lineage created_at must be timezone-aware")
created_at = created_at.astimezone(timezone.utc)
prepared = self.service.prepare_new_task_identity(
NewTaskIdentityRequest(
task_id=str(plan.task_id),
owner_principal_id=principal.owner_principal_id,
preferred_owner_alias=principal.preferred_owner_alias,
created_at=created_at,
mission_id=principal.mission_id,
goal_slug=self._goal_slug(plan),
root_acceptance_ref=principal.root_acceptance_ref,
relation_type=RelationType.ROOT,
)
)
plan.metadata = {
**plan.metadata,
**prepared.metadata_patch,
LINEAGE_REQUEST_DIGEST_KEY: request_digest,
}
return self.store.create_task_with_event(
plan,
idempotency_key=idempotency_key,
owner_principal_id=principal.owner_principal_id,
mission_id=principal.mission_id,
created_at=created_at.timestamp(),
)
def delegate_child(
self,
*,
parent_task_id: str,
plan: TaskPlan,
authenticated_identity: str,
request_digest: str,
delegation: DelegationEngine,
principal: PrincipalLineageBinding | None = None,
) -> TaskRecord:
if not plan.task_id:
raise ValidationError("lineage delegation requires a stable task_id")
principal = principal or self.policy.resolve_principal(authenticated_identity)
if principal.authenticated_identity != authenticated_identity:
raise ForbiddenError("delegation principal identity mismatch")
self._reject_reserved_metadata(plan)
with self.store._transaction():
parent = self.store.get_task(parent_task_id)
root = self.store.get_task(parent.root_task_id)
if (
root.owner_principal_id != principal.owner_principal_id
or root.mission_id != principal.mission_id
):
raise ForbiddenError("delegation actor does not match root authority")
try:
existing = self.store.get_task(str(plan.task_id))
except NotFoundError:
existing = None
if existing is not None:
return self._verify_replay(
existing,
request_digest=request_digest,
principal=principal,
parent_task_id=parent_task_id,
)
self.store.assert_root_open_for_child(root.task_id)
created_at = datetime.now(tz=timezone.utc)
prepared = self.service.prepare_new_task_identity(
NewTaskIdentityRequest(
task_id=str(plan.task_id),
owner_principal_id=root.owner_principal_id,
preferred_owner_alias=principal.preferred_owner_alias,
created_at=created_at,
mission_id=root.mission_id,
goal_slug=self._goal_slug(plan),
root_acceptance_ref=str(
root.task_plan.metadata[LINEAGE_METADATA_KEY]["root_acceptance_ref"]
),
relation_type=RelationType.CHILD,
parent_task_id=parent_task_id,
)
)
plan.metadata = {
**plan.metadata,
**prepared.metadata_patch,
LINEAGE_REQUEST_DIGEST_KEY: request_digest,
}
return delegation.delegate(
parent_task_id,
plan,
caller=authenticated_identity,
owner_principal_id=root.owner_principal_id,
mission_id=root.mission_id,
created_at=created_at.timestamp(),
)
def current_acceptance_snapshot(self, *, root_task_id: str) -> RootAcceptanceSnapshot:
"""Evaluate the current authoritative root/descendant projection."""
root = _as_task_record(self.binding.get_task_snapshot(root_task_id))
descendants = tuple(
_as_task_record(snapshot)
for snapshot in self.binding.list_descendant_snapshots(root_task_id)
)
return self.policy.evaluate_acceptance(root, descendants)
def review_proven_terminal_child(
self,
*,
child_task_id: str,
claim_id: str,
result_digest: str,
evidence_refs: Sequence[str],
):
with self.store._transaction():
self.binding.record_terminal_proof(
task_id=child_task_id,
claim_id=claim_id,
result_digest=result_digest,
evidence_refs=evidence_refs,
)
return self.resume_terminal_review(child_task_id=child_task_id)
def resume_terminal_review(self, *, child_task_id: str) -> TerminalReviewOutcome:
"""Resume after restart without re-completing a claim or re-evaluating a review.
Once a trigger has a persisted review, that exact decision is frozen.
Policy/evaluator drift therefore cannot create a second decision for the
same terminal transition.
"""
with self.store._transaction():
child = self.binding.get_task_snapshot(child_task_id)
if not child.terminal_evidence_proven:
raise ConflictError("terminal review resume requires durable proof")
raw_lineage = child.metadata.get(LINEAGE_METADATA_KEY)
if not isinstance(raw_lineage, Mapping):
raise ConflictError("lineage child lacks authoritative lineage metadata")
try:
child_lineage = lineage_from_metadata(dict(raw_lineage))
except ValueError as exc:
raise ConflictError("lineage child metadata is invalid") from exc
persisted = self.binding.get_root_review_for_trigger(child_task_id)
if persisted is not None:
review_id = persisted.get("review_id")
if not isinstance(review_id, str) or not review_id:
raise ConflictError("persisted root review identity is invalid")
try:
review = root_review_from_record(
persisted, expected_review_id=review_id
)
except ValueError as exc:
raise ConflictError("persisted root review is invalid") from exc
root = self.binding.get_task_snapshot(child_lineage.root_task_id)
if (
review.trigger_child_task_id != child_task_id
or review.root_task_id != child_lineage.root_task_id
or child_lineage.task_id != child_task_id
or child_lineage.owner_principal_id != child.owner_principal_id
or child_lineage.mission_id != child.mission_id
or root.owner_principal_id != child.owner_principal_id
or root.mission_id != child.mission_id
or root.revision < review.root_revision_after
):
raise ConflictError("persisted root review authority mismatch")
dispatch = self.binding.get_root_action_dispatch(review.review_id)
if dispatch is None:
dispatch = self.binding.dispatch_root_action(
review,
f"root-review:{review.review_id}:{review.next_legal_action.value}",
)
expected_key = (
f"root-review:{review.review_id}:{review.next_legal_action.value}"
)
if (
dispatch.review_id != review.review_id
or dispatch.root_task_id != review.root_task_id
or dispatch.action != review.next_legal_action
or dispatch.idempotency_key != expected_key
or dispatch.status not in PERSISTED_DISPATCH_STATUSES
):
raise ConflictError("persisted root action dispatch is invalid")
return TerminalReviewOutcome(review=review, dispatch=dispatch)
hook_outcome = self.hooks.on_proven_terminal_child(
child_task_id=child_task_id,
context=AuthenticatedTaskContext(
principal_id=child.owner_principal_id,
preferred_alias=child_lineage.owner_alias,
mission_id=child.mission_id,
),
acceptance_evaluator=self.policy.evaluate_acceptance,
reviewed_at=datetime.now(tz=timezone.utc),
require_independent_readback=False,
)
return hook_outcome.outcome
File diff suppressed because it is too large Load Diff
+7
View File
@@ -388,6 +388,8 @@ class TaskRecord:
root_task_id: str
idempotency_key: str | None = None
failure_reason: str | None = None
owner_principal_id: str = ""
mission_id: str = ""
def with_status(self, status: str) -> "TaskRecord":
allowed = TASK_NEXT_STATES.get(self.status, set())
@@ -407,6 +409,8 @@ class TaskRecord:
root_task_id=self.root_task_id,
idempotency_key=self.idempotency_key,
failure_reason=self.failure_reason,
owner_principal_id=self.owner_principal_id,
mission_id=self.mission_id,
)
def to_dict(self, *, include_task=True, redact_payload: bool = True) -> dict[str, Any]:
@@ -422,6 +426,9 @@ class TaskRecord:
}
if include_task:
payload["plan"] = self.task_plan.to_dict(redact=redact_payload)
display_name = self.task_plan.metadata.get("aios_task_display_name")
if self.owner_principal_id and self.mission_id and isinstance(display_name, str):
payload["display_name"] = display_name
if self.idempotency_key:
payload["idempotency_key"] = self.idempotency_key
return payload
-78
View File
@@ -1,78 +0,0 @@
"""Claim/proof receipts for ANNEX-2.3 R2 governed claim audit read-back.
A receipt is a minimal, tamper-evident record generated alongside a governed
claim. Its digest is sha256(canonical_json(payload)) so the payload can be
re-verified on read-back, consistent with the evidence and event chain.
"""
from __future__ import annotations
import time
import uuid
from dataclasses import dataclass
from typing import Any
from .models import compute_sha256_hex
def receipt_digest(payload: dict[str, Any]) -> str:
return compute_sha256_hex(payload)
@dataclass
class ClaimReceipt:
receipt_id: str
kind: str
task_id: str
worker_id: str
claim_id: str | None
proof_id: str | None
digest: str
payload: dict[str, Any]
created_at: float
def to_dict(self) -> dict[str, Any]:
return {
"receipt_id": self.receipt_id,
"kind": self.kind,
"task_id": self.task_id,
"worker_id": self.worker_id,
"claim_id": self.claim_id,
"proof_id": self.proof_id,
"digest": self.digest,
"payload": self.payload,
"created_at": self.created_at,
}
def build_receipt(
*,
kind: str,
task_id: str,
worker_id: str,
claim_id: str | None = None,
proof_id: str | None = None,
extra: dict[str, Any] | None = None,
created_at: float | None = None,
) -> ClaimReceipt:
"""Build a ClaimReceipt whose digest covers the full digestable payload."""
extra = extra or {}
payload = {
"kind": kind,
"task_id": task_id,
"worker_id": worker_id,
"claim_id": claim_id,
"proof_id": proof_id,
**extra,
}
return ClaimReceipt(
receipt_id=f"rct-{uuid.uuid4()}",
kind=kind,
task_id=task_id,
worker_id=worker_id,
claim_id=claim_id,
proof_id=proof_id,
digest=receipt_digest(payload),
payload=payload,
created_at=created_at if created_at is not None else time.time(),
)
-240
View File
@@ -1,240 +0,0 @@
"""RecheckEngine producing auditable RecheckDecision for governed claims.
The authoritative_claim_recheck re-reads live state at claim time and evaluates
the full A23 check set (both positive evidence and the five A23 negative cases).
See ANNEX-2.3 R2 ARCH-AUTO-CLAIM-R2-20260914 §4.2.
"""
from __future__ import annotations
import time
import uuid
from dataclasses import dataclass, field
from typing import Any, Callable
from .capabilities import has_capabilities
from .errors import ConflictError, ForbiddenError, LockedError, TaskPlaneHTTPError
from .models import compute_sha256_hex
from .schedulability import default_tenant_of_task, default_tenant_of_worker
# check -> (error_code, exception_factory)
_A23_BY_CHECK = {
"tenant_match": ("cross_tenant_denied", ForbiddenError),
"principal_match": ("task_binding_mismatch", ConflictError),
"task_match": ("task_binding_mismatch", ConflictError),
"generation_ok": ("generation_stale", ConflictError),
"lease_valid": ("lease_expired_or_closed", LockedError),
"capacity_ok": ("capacity_exhausted", LockedError),
}
CLAIMABLE_STATUSES = {"CREATED", "QUEUED", "FAILED_RECOVERABLE"}
@dataclass
class Check:
check: str
ok: bool
detail: str = ""
def to_dict(self) -> dict[str, Any]:
return {"check": self.check, "ok": self.ok, "detail": self.detail}
@dataclass
class RecheckDecision:
recheck_id: str
proof_id: str
task_id: str
worker_id: str
claim_id: str | None
allowed: bool
reason: str | None
checks: list[Check]
observed_at: float
digest: str = ""
error_code: str | None = None
status_code: int = 400
def payload(self) -> dict[str, Any]:
return {
"recheck_id": self.recheck_id,
"proof_id": self.proof_id,
"task_id": self.task_id,
"worker_id": self.worker_id,
"claim_id": self.claim_id,
"allowed": self.allowed,
"reason": self.reason,
"checks": [c.to_dict() for c in self.checks],
"observed_at": self.observed_at,
}
def to_dict(self) -> dict[str, Any]:
return {
**self.payload(),
"digest": self.digest,
"error_code": self.error_code,
"status_code": self.status_code,
}
def failed_check(self) -> Check | None:
for check in self.checks:
if not check.ok:
return check
return None
class RecheckEngine:
def __init__(
self,
store: Any,
*,
tenant_of_task: Callable[[Any], str] | None = None,
tenant_of_worker: Callable[[Any], str] | None = None,
authorization_ok: Callable[[Any], bool] | None = None,
) -> None:
self.store = store
self.tenant_of_task = tenant_of_task or default_tenant_of_task
self.tenant_of_worker = tenant_of_worker or default_tenant_of_worker
self.authorization_ok = authorization_ok or (lambda worker: True)
def authoritative_claim_recheck(
self,
*,
proof: Any,
task: Any,
worker: Any,
now: float | None = None,
) -> RecheckDecision:
now = now if now is not None else time.time()
checks: list[Check] = []
tenant_task = self.tenant_of_task(task)
tenant_worker = self.tenant_of_worker(worker)
checks.append(
Check(
"tenant_match",
tenant_task == tenant_worker,
f"task_tenant={tenant_task} worker_tenant={tenant_worker}"
if tenant_task != tenant_worker
else "tenant matched",
)
)
principal_ok = proof.binding.get("worker_identity") == worker.identity
checks.append(
Check(
"principal_match",
principal_ok,
f"proof_identity={proof.binding.get('worker_identity')} actual={worker.identity}"
if not principal_ok
else "identity matched",
)
)
authz_ok = bool(self.authorization_ok(worker))
checks.append(
Check(
"authorization_ok",
authz_ok,
"worker not in governed authorization profile" if not authz_ok else "authorized",
)
)
task_match = task.task_id == proof.task_id and task.status in CLAIMABLE_STATUSES
checks.append(
Check(
"task_match",
task_match,
f"task_status={task.status}" if not task_match else "task matched",
)
)
gen_ok = task.revision == proof.runtime["generation"]["task_revision"]
checks.append(
Check(
"generation_ok",
gen_ok,
f"proof_revision={proof.runtime['generation']['task_revision']} actual={task.revision}"
if not gen_ok
else "generation current",
)
)
lease_ok = self.store.active_claim(task.task_id) is None
checks.append(
Check(
"lease_valid",
lease_ok,
"task already actively claimed" if not lease_ok else "lease free",
)
)
current_active = self.store.get_active_claim_count(worker.worker_id)
capacity_ok = current_active < worker.max_concurrency
checks.append(
Check(
"capacity_ok",
capacity_ok,
f"active={current_active} max={worker.max_concurrency}"
if not capacity_ok
else "capacity available",
)
)
online = (not worker.disabled) and (
now <= worker.last_heartbeat + (worker.heartbeat_interval_seconds * 3)
)
checks.append(
Check(
"worker_online",
online,
"worker offline or disabled" if not online else "online",
)
)
caps_ok = has_capabilities(worker.capabilities, task.task_plan.required_capabilities)
checks.append(
Check(
"capabilities_match",
caps_ok,
"worker lacks required capabilities" if not caps_ok else "capabilities match",
)
)
allowed = all(check.ok for check in checks)
failed = next((check for check in checks if not check.ok), None)
reason = failed.detail if failed else None
error_code, status_code = self._error_for(failed)
decision = RecheckDecision(
recheck_id=f"rc-{uuid.uuid4()}",
proof_id=proof.proof_id,
task_id=task.task_id,
worker_id=worker.worker_id,
claim_id=proof.claim_id,
allowed=allowed,
reason=reason,
checks=checks,
observed_at=now,
error_code=error_code,
status_code=status_code,
)
decision.digest = compute_sha256_hex(decision.payload())
return decision
def _error_for(self, failed: Check | None) -> tuple[str | None, int]:
if failed is None:
return None, 400
entry = _A23_BY_CHECK.get(failed.check)
if entry is None:
return "forbidden", 403
return entry[0], entry[1].status_code
def to_error(self, decision: RecheckDecision) -> TaskPlaneHTTPError:
if decision.allowed:
raise ValueError("cannot build an error from an allowed decision")
failed = decision.failed_check()
if failed is None:
return ForbiddenError(decision.reason or "claim rejected")
code, exc_factory = _A23_BY_CHECK.get(failed.check, ("forbidden", ForbiddenError))
return exc_factory(decision.reason or failed.detail, code=code)
File diff suppressed because it is too large Load Diff
+466 -144
View File
@@ -1,191 +1,513 @@
"""SchedulabilityProof model and ProofBuilder for governed claim.
"""Diagnostic schedulability proof and claim-time authoritative recheck.
A SchedulabilityProof is a durable, server-side diagnostic snapshot that records
the state evaluated before an Agent may claim a task. It is NOT a bearer
credential; it is persisted for auditability and idempotent read-back. The five
sections captured are binding / worker / runtime / observations / status,
following ANNEX-2.3 R2 ARCH-AUTO-CLAIM-R2-20260914 §4.1.
The proof is evidence, never a bearer credential.
A claim proceeds in three stages:
1. read-only preflight/discovery;
2. claim-time read-only recheck;
3. final *atomic* authority operation that revalidates all mutable claim
conditions at the linearization point and returns lease/fencing receipt.
This module does not create a second scheduler or a second TaskPlane authority.
The host adapter must bind ``claim_atomically`` to the existing authoritative
claim transaction / repository operation.
"""
from __future__ import annotations
from dataclasses import dataclass, asdict
from typing import Any, Mapping, Protocol, Sequence
import hashlib
import time
import uuid
from dataclasses import dataclass, field
from typing import Any, Callable
from .models import compute_sha256_hex
from .models import canonical_json
class ProofStatus:
PENDING = "PENDING"
ACCEPTED = "ACCEPTED"
REJECTED = "REJECTED"
CLAIMED = "CLAIMED"
FAILED = "FAILED"
CLAIMABLE_TASK_STATES = frozenset({"CREATED", "QUEUED", "FAILED_RECOVERABLE"})
@dataclass
class Generation:
class SchedulabilityError(RuntimeError):
pass
class SchedulabilityChanged(SchedulabilityError):
pass
def _canonical(value: object) -> bytes:
# Reuse the existing TaskPlane canonical JSON contract. Do not introduce
# another sort/separator encoder for schedulability evidence.
return canonical_json(value).encode("utf-8")
def _sha(value: object) -> str:
return hashlib.sha256(_canonical(value)).hexdigest()
def _text(value: object, name: str, *, max_len: int = 256) -> str:
if not isinstance(value, str):
raise SchedulabilityError(f"{name} must be a string")
value = value.strip()
if not value or len(value) > max_len or any(ord(ch) < 32 for ch in value):
raise SchedulabilityError(f"invalid {name}")
return value
def _int_authority(
value: object,
*,
name: str,
default: int | None = None,
minimum: int = 0,
) -> tuple[int, "Blocker | None"]:
if value is None:
if default is None:
return 0, Blocker("INVALID_AUTHORITY_VALUE", "worker", f"{name} missing")
return default, None
if isinstance(value, bool) or not isinstance(value, int):
return 0, Blocker("INVALID_AUTHORITY_VALUE", "worker", f"{name} must be integer")
if value < minimum:
return value, Blocker("INVALID_AUTHORITY_VALUE", "worker", f"{name} below {minimum}")
return value, None
@dataclass(frozen=True)
class Blocker:
code: str
plane: str
detail: str
def to_dict(self) -> dict[str, str]:
return asdict(self)
@dataclass(frozen=True)
class ClaimConditions:
"""Mutable authority conditions that MUST be checked atomically with claim."""
task_revision: int
attempt: int | None = None
lease_generation: int | None = None
task_generation: int
task_status: str
worker_generation: int
worker_identity: str
authorization_ref: str
authorization_revision: str
scope_digest: str
capacity_limit: int
def to_dict(self) -> dict[str, Any]:
def to_dict(self) -> dict[str, object]:
return asdict(self)
@dataclass(frozen=True)
class SchedulabilityObservation:
task_id: str
task_revision: int
task_generation: int
task_status: str
worker_id: str
worker_generation: int
worker_identity: str
heartbeat_observed_at: float
heartbeat_fresh: bool
required_capabilities: tuple[str, ...]
worker_capabilities: tuple[str, ...]
scope_digest: str
authorization_ref: str
authorization_revision: str
authorization_allowed: bool
capacity_limit: int
active_claims: int
dependencies_satisfied: bool
freeze_blocked: bool
conflicting_lease: bool
@property
def capacity_remaining(self) -> int:
return max(0, self.capacity_limit - self.active_claims)
def stable_binding(self) -> dict[str, object]:
"""Fields whose change invalidates an earlier proof."""
return {
"task_id": self.task_id,
"task_revision": self.task_revision,
"attempt": self.attempt,
"lease_generation": self.lease_generation,
"task_generation": self.task_generation,
"task_status": self.task_status,
"worker_id": self.worker_id,
"worker_generation": self.worker_generation,
"worker_identity": self.worker_identity,
"required_capabilities": list(self.required_capabilities),
"worker_capabilities": list(self.worker_capabilities),
"scope_digest": self.scope_digest,
"authorization_ref": self.authorization_ref,
"authorization_revision": self.authorization_revision,
"authorization_allowed": self.authorization_allowed,
"capacity_limit": self.capacity_limit,
"active_claims": self.active_claims,
"dependencies_satisfied": self.dependencies_satisfied,
"freeze_blocked": self.freeze_blocked,
"conflicting_lease": self.conflicting_lease,
"heartbeat_fresh": self.heartbeat_fresh,
}
@classmethod
def from_dict(cls, data: dict[str, Any] | None) -> "Generation":
data = data or {}
return cls(
task_revision=int(data.get("task_revision") or 0),
attempt=data.get("attempt"),
lease_generation=data.get("lease_generation"),
def claim_conditions(self) -> ClaimConditions:
return ClaimConditions(
task_revision=self.task_revision,
task_generation=self.task_generation,
task_status=self.task_status,
worker_generation=self.worker_generation,
worker_identity=self.worker_identity,
authorization_ref=self.authorization_ref,
authorization_revision=self.authorization_revision,
scope_digest=self.scope_digest,
capacity_limit=self.capacity_limit,
)
@dataclass
@dataclass(frozen=True)
class SchedulabilityProof:
schema: str
proof_id: str
task_id: str
worker_id: str
generated_at: float
binding: dict[str, Any]
worker: dict[str, Any]
runtime: dict[str, Any]
observations: list[dict[str, Any]]
status: str = ProofStatus.PENDING
schema_version: int = 1
claim_id: str | None = None
digest: str = ""
actor_identity: str
created_at: float
expires_at: float
observation: SchedulabilityObservation
blockers: tuple[Blocker, ...]
observation_digest: str
proof_digest: str
def payload(self) -> dict[str, Any]:
"""The digest-stable payload (excludes the digest field itself)."""
return {
"proof_id": self.proof_id,
"schema_version": self.schema_version,
"task_id": self.task_id,
"worker_id": self.worker_id,
"claim_id": self.claim_id,
"generated_at": self.generated_at,
"binding": self.binding,
"worker": self.worker,
"runtime": self.runtime,
"observations": self.observations,
"status": self.status,
@classmethod
def create(
cls,
*,
task_id: str,
worker_id: str,
actor_identity: str,
observation: SchedulabilityObservation,
blockers: Sequence[Blocker],
now: float,
ttl_seconds: float,
proof_id: str | None = None,
) -> "SchedulabilityProof":
if ttl_seconds <= 0 or ttl_seconds > 300:
raise ValueError("schedulability proof ttl must be in (0,300]")
payload = {
"schema": "aios.schedulability-proof.v2",
"proof_id": proof_id or f"sched-{uuid.uuid4()}",
"task_id": _text(task_id, "task_id"),
"worker_id": _text(worker_id, "worker_id"),
"actor_identity": _text(actor_identity, "actor_identity", max_len=512),
"created_at": float(now),
"expires_at": float(now + ttl_seconds),
"observation": observation.stable_binding(),
"blockers": [b.to_dict() for b in blockers],
}
obs_digest = _sha(payload["observation"])
proof_digest = _sha({**payload, "observation_digest": obs_digest})
return cls(
schema=payload["schema"],
proof_id=payload["proof_id"],
task_id=payload["task_id"],
worker_id=payload["worker_id"],
actor_identity=payload["actor_identity"],
created_at=payload["created_at"],
expires_at=payload["expires_at"],
observation=observation,
blockers=tuple(blockers),
observation_digest=obs_digest,
proof_digest=proof_digest,
)
def to_dict(self) -> dict[str, Any]:
return {**self.payload(), "digest": self.digest}
@property
def claimable(self) -> bool:
return not self.blockers
def compute_proof_digest(proof_or_dict: Any) -> str:
"""sha256 over canonical_json of the digest-stable proof payload."""
if isinstance(proof_or_dict, SchedulabilityProof):
return compute_sha256_hex(proof_or_dict.payload())
payload = dict(proof_or_dict)
payload.pop("digest", None)
return compute_sha256_hex(payload)
@dataclass(frozen=True)
class DiscoveryDiagnostics:
candidate_count: int
blocker_codes: tuple[str, ...]
def observation(source: str, field_: str, value: Any, ok: bool, at: float) -> dict[str, Any]:
return {"source": source, "field": field_, "value": value, "ok": ok, "at": at}
class SchedulabilityAuthority(Protocol):
"""Read authorities and perform the final existing claim transaction."""
def read_task(self, task_id: str) -> Mapping[str, Any]: ...
def list_candidate_tasks(self) -> Sequence[Mapping[str, Any]]: ...
def read_worker(self, worker_id: str) -> Mapping[str, Any]: ...
def read_authorization(
self, *, task: Mapping[str, Any], worker: Mapping[str, Any], actor_identity: str
) -> Mapping[str, Any]: ...
def active_claim_count(self, worker_id: str) -> int: ...
def dependencies_satisfied(self, task: Mapping[str, Any]) -> bool: ...
def freeze_blocked(self, task: Mapping[str, Any]) -> bool: ...
def conflicting_lease(self, task_id: str) -> bool: ...
def exact_scope_digest(
self, *, task: Mapping[str, Any], worker: Mapping[str, Any], actor_identity: str
) -> tuple[bool, str]: ...
def claim_atomically(
self,
*,
task_id: str,
worker_id: str,
actor_identity: str,
expected: ClaimConditions,
lease_seconds: int,
schedulability_proof_digest: str,
) -> Mapping[str, Any]:
"""Claim at the existing TaskPlane linearization point.
The implementation MUST, in the same authoritative transaction / fenced
operation that writes the claim:
- verify task revision/generation/status is still claimable;
- verify worker registration/identity/health/capacity;
- verify authorization ref/revision remains allowed;
- verify exact scope;
- verify dependencies/freeze/conflicting lease;
- create claim + lease/fencing receipt.
This protocol is an adapter contract; it is not evidence that a specific
production adapter already implements the transaction.
"""
def default_tenant_of_task(task: Any) -> str:
return task.task_plan.metadata.get("tenant_id", "default")
class ProofRepository(Protocol):
def put(self, proof: SchedulabilityProof) -> None: ...
def get(self, proof_id: str) -> SchedulabilityProof | None: ...
def default_tenant_of_worker(worker: Any) -> str:
return worker.trust_zone or "default"
class MemoryProofRepository:
def __init__(self) -> None:
self._values: dict[str, SchedulabilityProof] = {}
def put(self, proof: SchedulabilityProof) -> None:
existing = self._values.get(proof.proof_id)
if existing is not None and existing.proof_digest != proof.proof_digest:
raise SchedulabilityError("proof id replay conflict")
self._values[proof.proof_id] = proof
def get(self, proof_id: str) -> SchedulabilityProof | None:
return self._values.get(proof_id)
class ProofBuilder:
"""Builds a SchedulabilityProof from live store snapshots."""
class SchedulabilityService:
def __init__(
self,
store: Any,
authority: SchedulabilityAuthority,
proofs: ProofRepository,
*,
tenant_of_task: Callable[[Any], str] | None = None,
tenant_of_worker: Callable[[Any], str] | None = None,
clock=time.time,
proof_ttl_seconds: float = 30.0,
heartbeat_grace_factor: float = 3.0,
) -> None:
self.store = store
self.tenant_of_task = tenant_of_task or default_tenant_of_task
self.tenant_of_worker = tenant_of_worker or default_tenant_of_worker
self.authority = authority
self.proofs = proofs
self.clock = clock
self.proof_ttl_seconds = proof_ttl_seconds
self.heartbeat_grace_factor = heartbeat_grace_factor
def build(
@staticmethod
def _caps(value: object) -> tuple[str, ...]:
if not isinstance(value, (list, tuple)):
return ()
return tuple(sorted({str(v) for v in value if isinstance(v, str) and v}))
def observe(
self, task_id: str, worker_id: str, actor_identity: str
) -> tuple[SchedulabilityObservation, tuple[Blocker, ...]]:
task = self.authority.read_task(task_id)
worker = self.authority.read_worker(worker_id)
auth = self.authority.read_authorization(
task=task, worker=worker, actor_identity=actor_identity
)
required = self._caps(task.get("required_capabilities") or ())
offered = self._caps(worker.get("capabilities") or ())
now = float(self.clock())
last = float(worker.get("last_heartbeat") or worker.get("last_seen_at_epoch") or 0.0)
interval = float(worker.get("heartbeat_interval_seconds") or 15.0)
fresh = (
str(worker.get("status") or "").lower() == "active"
and not bool(worker.get("disabled"))
and last <= now
and now - last <= max(interval * self.heartbeat_grace_factor, 1.0)
)
scope_ok, scope_digest = self.authority.exact_scope_digest(
task=task, worker=worker, actor_identity=actor_identity
)
deps = bool(self.authority.dependencies_satisfied(task))
freeze = bool(self.authority.freeze_blocked(task))
lease = bool(self.authority.conflicting_lease(task_id))
active_raw = self.authority.active_claim_count(worker_id)
active, active_error = _int_authority(active_raw, name="active_claim_count", minimum=0)
capacity, capacity_error = _int_authority(
worker.get("max_concurrency"),
name="max_concurrency",
default=1,
minimum=0,
)
auth_allowed = bool(auth.get("allowed"))
task_status = str(task.get("status") or "")
blockers: list[Blocker] = []
if task_status not in CLAIMABLE_TASK_STATES:
blockers.append(
Blocker(
"TASK_STATE_NOT_CLAIMABLE",
"task",
f"task status {task_status or '<missing>'} is not claimable",
)
)
if active_error:
blockers.append(active_error)
if capacity_error:
blockers.append(capacity_error)
if not fresh:
blockers.append(Blocker("STALE_OR_INACTIVE_WORKER", "worker", "worker heartbeat is not fresh"))
missing = sorted(set(required) - set(offered))
if missing:
blockers.append(Blocker("MISSING_CAPABILITY", "worker", ",".join(missing)))
if not scope_ok:
blockers.append(Blocker("EXACT_SCOPE_MISMATCH", "authorization", "worker/task signed scope mismatch"))
if not auth_allowed:
blockers.append(Blocker("AUTHORIZATION_BLOCKED", "authorization", "current authorization does not allow claim"))
if capacity >= 0 and active >= capacity:
blockers.append(Blocker("CAPACITY_EXHAUSTED", "worker", f"{active}/{capacity} active"))
if not deps:
blockers.append(Blocker("DEPENDENCY_BLOCKED", "task", "task dependencies are not satisfied"))
if freeze:
blockers.append(Blocker("FREEZE_BLOCKED", "task", "task/environment freeze is active"))
if lease:
blockers.append(Blocker("CONFLICTING_LEASE", "task", "a live conflicting lease exists"))
obs = SchedulabilityObservation(
task_id=_text(task.get("task_id"), "task_id"),
task_revision=int(task.get("revision") or 0),
task_generation=int(task.get("generation") or task.get("metadata", {}).get("generation") or 0),
task_status=task_status,
worker_id=_text(worker.get("worker_id"), "worker_id"),
worker_generation=int(worker.get("registration_revision") or worker.get("generation") or 0),
worker_identity=_text(worker.get("identity"), "worker_identity", max_len=512),
heartbeat_observed_at=last,
heartbeat_fresh=fresh,
required_capabilities=required,
worker_capabilities=offered,
scope_digest=_text(scope_digest, "scope_digest", max_len=256),
authorization_ref=_text(auth.get("authorization_ref"), "authorization_ref", max_len=512),
authorization_revision=_text(auth.get("revision"), "authorization_revision", max_len=256),
authorization_allowed=auth_allowed,
capacity_limit=capacity,
active_claims=active,
dependencies_satisfied=deps,
freeze_blocked=freeze,
conflicting_lease=lease,
)
return obs, tuple(blockers)
def preflight(
self,
task: Any,
worker: Any,
*,
now: float | None = None,
claim_id: str | None = None,
task_id: str,
worker_id: str,
actor_identity: str,
persist: bool = False,
) -> SchedulabilityProof:
now = now if now is not None else time.time()
active = self.store.get_active_claim_count(worker.worker_id)
capacity_ok = active < worker.max_concurrency
online = (not worker.disabled) and (
now <= worker.last_heartbeat + (worker.heartbeat_interval_seconds * 3)
now = float(self.clock())
obs, blockers = self.observe(task_id, worker_id, actor_identity)
proof = SchedulabilityProof.create(
task_id=task_id,
worker_id=worker_id,
actor_identity=actor_identity,
observation=obs,
blockers=blockers,
now=now,
ttl_seconds=self.proof_ttl_seconds,
)
tenant_worker = self.tenant_of_worker(worker)
tenant_task = self.tenant_of_task(task)
binding = {
"worker_identity": worker.identity,
"capabilities": list(worker.capabilities),
"trust_zone": worker.trust_zone,
"tenant_id": tenant_worker,
}
binding["binding_digest"] = compute_sha256_hex(binding)
worker_snapshot = {
"status": worker.status,
"disabled": worker.disabled,
"last_heartbeat": worker.last_heartbeat,
"heartbeat_interval_seconds": worker.heartbeat_interval_seconds,
"online": online,
"worker_type": worker.worker_type,
"max_concurrency": worker.max_concurrency,
"current_active": active,
}
runtime = {
"max_concurrency": worker.max_concurrency,
"current_active": active,
"capacity_ok": capacity_ok,
"generation": {
"task_revision": task.revision,
"attempt": None,
"lease_generation": None,
},
}
observations = [
observation("tenant", "tenant_match", {"task": tenant_task, "worker": tenant_worker}, tenant_task == tenant_worker, now),
observation("principal", "worker_identity", worker.identity, True, now),
observation("authorization", "authorization_ok", None, True, now),
observation("task", "task_id", task.task_id, task.status in {"CREATED", "QUEUED", "FAILED_RECOVERABLE"}, now),
observation("worker", "worker_status", worker.status, not worker.disabled, now),
observation("generation", "task_revision", task.revision, True, now),
observation("capacity", "capacity_ok", capacity_ok, capacity_ok, now),
]
proof = SchedulabilityProof(
proof_id=f"proof-{uuid.uuid4()}",
task_id=task.task_id,
worker_id=worker.worker_id,
generated_at=now,
binding=binding,
worker=worker_snapshot,
runtime=runtime,
observations=observations,
status=ProofStatus.PENDING,
claim_id=claim_id,
)
proof.digest = compute_sha256_hex(proof.payload())
if persist:
self.proofs.put(proof)
return proof
def discover_with_diagnostics(
self, *, worker_id: str, actor_identity: str
) -> tuple[SchedulabilityProof | None, DiscoveryDiagnostics]:
candidates = sorted(
self.authority.list_candidate_tasks(),
key=lambda task: (
-int(task.get("priority") or 0),
str(task.get("created_at") or ""),
str(task.get("task_id") or ""),
),
)
codes: set[str] = set()
observed = 0
for task in candidates:
task_id = str(task.get("task_id") or "")
if not task_id:
continue
observed += 1
proof = self.preflight(
task_id=task_id,
worker_id=worker_id,
actor_identity=actor_identity,
persist=False,
)
if proof.claimable:
self.proofs.put(proof)
return proof, DiscoveryDiagnostics(observed, tuple(sorted(codes)))
codes.update(b.code for b in proof.blockers)
return None, DiscoveryDiagnostics(observed, tuple(sorted(codes)))
def discover(
self, *, worker_id: str, actor_identity: str
) -> SchedulabilityProof | None:
proof, _ = self.discover_with_diagnostics(
worker_id=worker_id, actor_identity=actor_identity
)
return proof
@staticmethod
def _validate_claim_receipt(result: Mapping[str, Any]) -> None:
required = ("claim_id", "lease_id", "fencing_generation", "receipt_revision")
missing = [name for name in required if result.get(name) in (None, "")]
if missing:
raise SchedulabilityChanged(
"claim authority did not return required lease/fence receipt: "
+ ",".join(missing)
)
def claim_with_recheck(
self,
*,
proof_id: str,
worker_id: str,
actor_identity: str,
lease_seconds: int = 30,
) -> Mapping[str, Any]:
proof = self.proofs.get(proof_id)
if proof is None:
raise SchedulabilityChanged("proof not found")
now = float(self.clock())
if proof.expires_at <= now:
raise SchedulabilityChanged("proof expired")
if proof.worker_id != worker_id or proof.actor_identity != actor_identity:
raise SchedulabilityChanged("proof identity mismatch")
current, blockers = self.observe(proof.task_id, worker_id, actor_identity)
if blockers:
raise SchedulabilityChanged(
"claim-time blockers: " + ",".join(sorted(b.code for b in blockers))
)
if current.stable_binding() != proof.observation.stable_binding():
raise SchedulabilityChanged("claim-time authority changed since proof")
result = self.authority.claim_atomically(
task_id=proof.task_id,
worker_id=worker_id,
actor_identity=actor_identity,
expected=current.claim_conditions(),
lease_seconds=lease_seconds,
schedulability_proof_digest=proof.proof_digest,
)
self._validate_claim_receipt(result)
return result
@@ -1,179 +0,0 @@
"""SchedulingService wiring governed claim dispatch + long_poll/poll/schedule.
The service owns the proof -> recheck -> govern pipeline and the worker pull
endpoints (long_poll / poll / schedule) with a process-local semaphore to bound
blocking long-poll concurrency. See ARCH-AUTO-CLAIM-R2-20260914 §1.5/§5/§6 T6.
"""
from __future__ import annotations
import threading
import time
from typing import Any, Callable
from .errors import ForbiddenError, LockedError, NotFoundError
from .recheck import RecheckEngine
from .schedulability import ProofBuilder, default_tenant_of_task, default_tenant_of_worker
from .scheduler import TaskScheduler
class SchedulingService:
def __init__(
self,
store: Any,
*,
recheck: RecheckEngine | None = None,
tenant_of_task: Callable[[Any], str] | None = None,
tenant_of_worker: Callable[[Any], str] | None = None,
long_poll_seconds: int = 15,
long_poll_max: int = 30,
poll_interval_seconds: int = 1,
long_poll_slots: int = 8,
) -> None:
self.store = store
self.recheck = recheck or RecheckEngine(
store,
tenant_of_task=tenant_of_task,
tenant_of_worker=tenant_of_worker,
)
self.builder = ProofBuilder(
store,
tenant_of_task=tenant_of_task or default_tenant_of_task,
tenant_of_worker=tenant_of_worker or default_tenant_of_worker,
)
self.scheduler = TaskScheduler(
store,
tenant_of_task=tenant_of_task or default_tenant_of_task,
tenant_of_worker=tenant_of_worker or default_tenant_of_worker,
)
self.long_poll_seconds = long_poll_seconds
self.long_poll_max = long_poll_max
self.poll_interval_seconds = poll_interval_seconds
self._semaphore = threading.Semaphore(long_poll_slots)
# -- internal pipeline -------------------------------------------------
def _claim_for(self, worker: Any, task: Any, lease_seconds: int, now: float) -> dict[str, Any]:
proof = self.builder.build(task, worker, now=now)
decision = self.recheck.authoritative_claim_recheck(
proof=proof, task=task, worker=worker, now=now
)
if not decision.allowed:
self.store.record_proof_rejection(
proof=proof, decision=decision, task=task, now=now
)
raise self.recheck.to_error(decision)
claim, receipt_id = self.store.governed_claim(
task=task,
worker=worker,
lease_seconds=lease_seconds,
proof=proof,
decision=decision,
now=now,
)
return {
"task": task.to_dict(),
"task_id": task.task_id,
"claim_id": claim.claim_id,
"worker_id": worker.worker_id,
"proof_id": proof.proof_id,
"receipt_id": receipt_id,
"generation": {
"task_revision": claim.task_revision,
"attempt": claim.attempt,
"lease_generation": claim.attempt,
},
}
# -- public endpoint semantics -----------------------------------------
def poll(self, worker_id: str, *, lease_seconds: int = 30) -> dict[str, Any]:
worker = self._validate_online(worker_id)
task = self.scheduler.next_queued_task(worker_id)
if task is None:
return {"status": "idle", "reason": "no_task"}
return self._claim_for(worker, task, lease_seconds, time.time())
def claim_governed(
self, worker_id: str, task_id: str, *, lease_seconds: int = 30
) -> dict[str, Any]:
worker = self.store.get_worker(worker_id)
task = self.store.get_task(task_id)
# G4: direct claim of a cross-tenant task stays FAIL-CLOSED (403 cross_tenant_denied),
# in contrast to enumeration which skips + audits.
w_tenant = self.builder.tenant_of_worker(worker)
t_tenant = self.builder.tenant_of_task(task)
if w_tenant != t_tenant:
raise ForbiddenError(
f"cross-tenant claim denied: task_tenant={t_tenant} worker_tenant={w_tenant}",
code="cross_tenant_denied",
)
return self._claim_for(worker, task, lease_seconds, time.time())
def long_poll(
self,
worker_id: str,
*,
timeout_seconds: int | None = None,
lease_seconds: int = 30,
) -> dict[str, Any]:
timeout = max(1, min(int(timeout_seconds or self.long_poll_seconds), self.long_poll_max))
worker = self._validate_online(worker_id)
acquired = self._semaphore.acquire(blocking=False)
try:
deadline = time.monotonic() + timeout
while True:
task = self.scheduler.next_queued_task(worker_id)
if task is not None:
return self._claim_for(worker, task, lease_seconds, time.time())
remaining = deadline - time.monotonic()
if remaining <= 0:
return {"status": "idle", "reason": "timeout"}
time.sleep(max(0.0, min(self.poll_interval_seconds, remaining)))
finally:
if acquired:
self._semaphore.release()
def schedule(self, worker_id: str) -> dict[str, Any]:
worker = self.store.get_worker(worker_id)
current = self.store.get_active_claim_count(worker_id)
online = (not worker.disabled) and (
time.time() <= worker.last_heartbeat + (worker.heartbeat_interval_seconds * 3)
)
worker_tenant = self.builder.tenant_of_worker(worker)
# G4: queued_count must exclude cross-tenant tasks (skip + audit; NOT fail-closed).
available: list[Any] = []
for t in self.store.list_tasks():
if t.status not in {"CREATED", "QUEUED", "FAILED_RECOVERABLE"}:
continue
if t.task_plan.task_type == "worker.registry.audit":
continue
t_tenant = self.builder.tenant_of_task(t)
if t_tenant != worker_tenant:
self.store.record_tenant_skip_audit(
t.task_id,
worker_id,
task_tenant=t_tenant,
worker_tenant=worker_tenant,
reason="cross_tenant_skip",
)
continue
available.append(t)
return {
"worker_id": worker_id,
"online": online,
"disabled": worker.disabled,
"max_concurrency": worker.max_concurrency,
"current_active": current,
"capacity_ok": current < worker.max_concurrency,
"queued_count": len(available),
"status": "ready" if online else "offline",
}
def _validate_online(self, worker_id: str) -> Any:
worker = self.store.get_worker(worker_id)
if worker.disabled:
raise LockedError("worker disabled")
if time.time() > worker.last_heartbeat + (worker.heartbeat_interval_seconds * 3):
raise LockedError("worker offline")
return worker
+12 -51
View File
@@ -3,37 +3,18 @@
from __future__ import annotations
import time
from typing import Any, Callable
from .errors import LockedError, NotFoundError
from .errors import NotFoundError
from .capabilities import has_capabilities
from .models import WorkerClaim
from .recheck import RecheckEngine
from .schedulability import (
ProofBuilder,
default_tenant_of_task,
default_tenant_of_worker,
)
from .store import TaskPlaneStore
class TaskScheduler:
def __init__(
self,
store: TaskPlaneStore,
*,
recheck: RecheckEngine | None = None,
tenant_of_task: Callable[[Any], str] | None = None,
tenant_of_worker: Callable[[Any], str] | None = None,
):
def __init__(self, store: TaskPlaneStore):
self.store = store
self.recheck = recheck or RecheckEngine(store)
self.builder = ProofBuilder(store)
self._tenant_of_task = tenant_of_task or default_tenant_of_task
self._tenant_of_worker = tenant_of_worker or default_tenant_of_worker
def next_queued_task(self, worker_id: str):
worker = self.store.get_worker(worker_id)
worker_tenant = self._tenant_of_worker(worker)
tasks = sorted(
self.store.list_tasks(),
key=lambda task: (
@@ -47,17 +28,6 @@ class TaskScheduler:
continue
if task.task_plan.task_type == "worker.registry.audit":
continue
# G4: enumeration = skip + audit for cross-tenant tasks (NOT fail-closed).
task_tenant = self._tenant_of_task(task)
if task_tenant != worker_tenant:
self.store.record_tenant_skip_audit(
task.task_id,
worker_id,
task_tenant=task_tenant,
worker_tenant=worker_tenant,
reason="cross_tenant_skip",
)
continue
if task.status == "FAILED_RECOVERABLE" and not self.store.task_has_retryable_probe(
task.task_id
):
@@ -83,33 +53,24 @@ class TaskScheduler:
def schedule_for_worker(self, worker_id: str, lease_seconds: int = 15) -> WorkerClaim:
worker = self.store.get_worker(worker_id)
if worker.disabled:
from .errors import LockedError
raise LockedError("worker disabled")
if time.time() > worker.last_heartbeat + (
worker.heartbeat_interval_seconds * 3
):
from .errors import LockedError
raise LockedError("worker offline")
task = self.next_queued_task(worker_id)
if task is None:
raise NotFoundError("no queued tasks available")
return self._governed_claim(worker, task, lease_seconds)
def _governed_claim(self, worker, task, lease_seconds: int) -> WorkerClaim:
now = time.time()
proof = self.builder.build(task, worker, now=now)
decision = self.recheck.authoritative_claim_recheck(
proof=proof, task=task, worker=worker, now=now
)
if not decision.allowed:
self.store.record_proof_rejection(
proof=proof, decision=decision, task=task, now=now
)
raise self.recheck.to_error(decision)
claim, _ = self.store.governed_claim(
task=task,
worker=worker,
claim = self.store.claim_task(
task.task_id,
worker_id,
worker.identity,
lease_seconds=lease_seconds,
proof=proof,
decision=decision,
now=now,
expected_revision=task.revision,
operation_digest=task.task_plan.operation_digest(),
)
return claim
+232 -440
View File
@@ -32,7 +32,6 @@ from .models import (
redact_value,
validate_identifier,
)
from .receipts import build_receipt
EventEnvelope = tuple[str, str, str, str, str, float]
@@ -273,46 +272,12 @@ class TaskPlaneStore:
)
"""
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS schedulability_proofs (
proof_id TEXT PRIMARY KEY,
task_id TEXT NOT NULL,
worker_id TEXT NOT NULL,
claim_id TEXT,
status TEXT NOT NULL,
envelope_json TEXT NOT NULL,
digest TEXT NOT NULL,
created_at REAL NOT NULL,
updated_at REAL NOT NULL
)
"""
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS receipts (
receipt_id TEXT PRIMARY KEY,
kind TEXT NOT NULL,
task_id TEXT NOT NULL,
claim_id TEXT,
proof_id TEXT,
worker_id TEXT NOT NULL,
digest TEXT NOT NULL,
payload_json TEXT NOT NULL,
created_at REAL NOT NULL
)
"""
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS tenant_skip_audit (
task_id TEXT NOT NULL,
worker_id TEXT NOT NULL,
finding_json TEXT NOT NULL,
created_at REAL NOT NULL,
PRIMARY KEY (task_id, worker_id)
)
"""
self._ensure_columns(
"tasks",
{
"owner_principal_id": "TEXT NOT NULL DEFAULT ''",
"mission_id": "TEXT NOT NULL DEFAULT ''",
},
)
self._ensure_columns(
"workers",
@@ -348,10 +313,165 @@ class TaskPlaneStore:
)
self.conn.execute("CREATE INDEX IF NOT EXISTS ix_outbox_status ON outbox(delivery_status)")
self.conn.execute(
"CREATE INDEX IF NOT EXISTS ix_proofs_task ON schedulability_proofs(task_id)"
"""
CREATE TABLE IF NOT EXISTS task_lineage_sequences (
parent_task_id TEXT NOT NULL,
relation_type TEXT NOT NULL,
child_task_id TEXT NOT NULL UNIQUE,
sibling_sequence INTEGER NOT NULL,
created_at REAL NOT NULL,
PRIMARY KEY(parent_task_id, relation_type, sibling_sequence)
)
"""
)
self.conn.execute(
"CREATE INDEX IF NOT EXISTS ix_receipts_task ON receipts(task_id)"
"""
CREATE TABLE IF NOT EXISTS task_terminal_proofs (
task_id TEXT PRIMARY KEY,
claim_id TEXT NOT NULL UNIQUE,
task_revision INTEGER NOT NULL,
result_digest TEXT NOT NULL,
evidence_refs_json TEXT NOT NULL,
created_at REAL NOT NULL
)
"""
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_root_reviews (
review_id TEXT PRIMARY KEY,
root_task_id TEXT NOT NULL,
trigger_child_task_id TEXT NOT NULL,
trigger_child_revision INTEGER NOT NULL,
root_revision_before INTEGER NOT NULL,
root_revision_after INTEGER NOT NULL,
record_json TEXT NOT NULL,
created_at REAL NOT NULL
)
"""
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_root_action_dispatches (
review_id TEXT PRIMARY KEY,
root_task_id TEXT NOT NULL,
action TEXT NOT NULL,
idempotency_key TEXT NOT NULL UNIQUE,
status TEXT NOT NULL,
created_at REAL NOT NULL
)
"""
)
self._ensure_columns(
"task_root_action_dispatches",
{
"mission_id": "TEXT NOT NULL DEFAULT ''",
"root_revision": "INTEGER NOT NULL DEFAULT 0",
"row_version": "INTEGER NOT NULL DEFAULT 0",
"plan_json": "TEXT",
"plan_digest": "TEXT",
"execution_key": "TEXT",
"claim_owner": "TEXT",
"claim_token": "TEXT",
"claim_fence": "INTEGER",
"claim_expires_at": "REAL",
"execution_receipt_json": "TEXT",
"execution_receipt_digest": "TEXT",
"rereview_json": "TEXT",
"rereview_digest": "TEXT",
"retry_at": "REAL",
"last_error_code": "TEXT",
"updated_at": "REAL",
},
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_root_action_effects (
execution_key TEXT PRIMARY KEY,
root_task_id TEXT NOT NULL,
target_task_id TEXT,
target_job_id TEXT,
request_digest TEXT NOT NULL,
state TEXT NOT NULL,
effect_phase TEXT NOT NULL DEFAULT 'LEGACY_UNKNOWN',
receipt_json TEXT,
receipt_digest TEXT,
retry_at REAL,
last_error_code TEXT,
created_at REAL NOT NULL,
updated_at REAL NOT NULL
)
"""
)
self._ensure_columns(
"task_root_action_effects",
{
# Existing v1/v2 EFFECT_UNKNOWN rows may already have sent
# an external request. They must never migrate to a phase
# that permits a new lease or commit.
"effect_phase": "TEXT NOT NULL DEFAULT 'LEGACY_UNKNOWN'",
},
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_root_action_fences (
root_task_id TEXT PRIMARY KEY,
next_fence INTEGER NOT NULL,
active_review_id TEXT,
active_fence INTEGER,
active_phase TEXT,
updated_at REAL NOT NULL
)
"""
)
self.conn.execute(
"CREATE INDEX IF NOT EXISTS ix_root_action_dispatch_ready "
"ON task_root_action_dispatches(status,retry_at,created_at)"
)
self.conn.execute(
"CREATE INDEX IF NOT EXISTS ix_root_action_dispatch_root "
"ON task_root_action_dispatches(root_task_id,root_revision)"
)
self.conn.execute(
"CREATE UNIQUE INDEX IF NOT EXISTS ux_root_review_trigger "
"ON task_root_reviews(trigger_child_task_id, trigger_child_revision)"
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_missions (
mission_id TEXT PRIMARY KEY,
root_task_id TEXT NOT NULL UNIQUE,
authenticated_identity TEXT NOT NULL,
owner_principal_id TEXT NOT NULL,
owner_alias TEXT NOT NULL,
goal_slug TEXT NOT NULL,
root_acceptance_ref TEXT NOT NULL UNIQUE,
acceptance_verifier_id TEXT NOT NULL,
original_acceptance_json TEXT NOT NULL,
request_digest TEXT NOT NULL,
idempotency_key TEXT NOT NULL,
created_at REAL NOT NULL,
UNIQUE(authenticated_identity, idempotency_key)
)
"""
)
self.conn.execute(
"CREATE INDEX IF NOT EXISTS ix_task_missions_owner "
"ON task_missions(authenticated_identity, mission_id)"
)
self.conn.execute(
"""
CREATE TABLE IF NOT EXISTS task_mission_delegations (
authenticated_identity TEXT NOT NULL,
idempotency_key TEXT NOT NULL,
mission_id TEXT NOT NULL,
parent_task_id TEXT NOT NULL,
child_task_id TEXT NOT NULL UNIQUE,
request_digest TEXT NOT NULL,
created_at REAL NOT NULL,
PRIMARY KEY(authenticated_identity, idempotency_key)
)
"""
)
def _ensure_columns(self, table: str, columns: dict[str, str]) -> None:
@@ -407,6 +527,8 @@ class TaskPlaneStore:
root_task_id=row["root_task_id"],
idempotency_key=row["idempotency_key"],
failure_reason=row["failure_reason"],
owner_principal_id=row["owner_principal_id"],
mission_id=row["mission_id"],
)
def _row_to_worker(self, row: sqlite3.Row | None) -> WorkerRegistration:
@@ -433,20 +555,32 @@ class TaskPlaneStore:
disabled_reason=row["disabled_reason"],
)
def create_task(self, task_plan: TaskPlan, idempotency_key: str | None = None) -> TaskRecord:
def create_task(
self,
task_plan: TaskPlan,
idempotency_key: str | None = None,
*,
owner_principal_id: str = "",
mission_id: str = "",
created_at: float | None = None,
) -> TaskRecord:
task_id = task_plan.task_id or f"task-{uuid.uuid4()}"
task_plan.task_id = task_id
task_json = redact_value(task_plan.to_dict(redact=False))
now = _now()
now = _now() if created_at is None else float(created_at)
with self._transaction():
if idempotency_key:
existing = self.conn.execute(
"SELECT task_id,task_json FROM tasks WHERE idempotency_key=?",
"SELECT task_id,task_json,owner_principal_id,mission_id FROM tasks WHERE idempotency_key=?",
(idempotency_key,),
).fetchone()
if existing:
if _canonical(json.loads(existing["task_json"])) != _canonical(task_json):
raise ConflictError("idempotency key payload mismatch")
if owner_principal_id and existing["owner_principal_id"] != owner_principal_id:
raise ConflictError("idempotency key owner mismatch")
if mission_id and existing["mission_id"] != mission_id:
raise ConflictError("idempotency key mission mismatch")
return self.get_task(existing["task_id"])
root_task_id = task_id
for dependency in task_plan.dependencies:
@@ -460,8 +594,8 @@ class TaskPlaneStore:
self.conn.execute(
"""
INSERT INTO tasks
(task_id, revision, status, task_json, task_type, root_task_id, parent_task_id, correlation_id, idempotency_key, created_at, updated_at, failure_reason)
VALUES (?, 1, ?, ?, ?, ?, ?, ?, ?, ?, ?, NULL)
(task_id, revision, status, task_json, task_type, root_task_id, parent_task_id, correlation_id, idempotency_key, created_at, updated_at, failure_reason, owner_principal_id, mission_id)
VALUES (?, 1, ?, ?, ?, ?, ?, ?, ?, ?, ?, NULL, ?, ?)
""",
(
task_id,
@@ -474,6 +608,8 @@ class TaskPlaneStore:
idempotency_key,
now,
now,
owner_principal_id,
mission_id,
),
)
self._set_budget(task_id, task_plan.budgets)
@@ -485,10 +621,22 @@ class TaskPlaneStore:
return self.get_task(task_id)
def create_task_with_event(
self, task_plan: TaskPlan, idempotency_key: str | None = None
self,
task_plan: TaskPlan,
idempotency_key: str | None = None,
*,
owner_principal_id: str = "",
mission_id: str = "",
created_at: float | None = None,
) -> TaskRecord:
with self._transaction():
task = self.create_task(task_plan, idempotency_key=idempotency_key)
task = self.create_task(
task_plan,
idempotency_key=idempotency_key,
owner_principal_id=owner_principal_id,
mission_id=mission_id,
created_at=created_at,
)
if not self.list_events(task.task_id):
self._fault("create.after_task_before_event")
self.append_event(
@@ -509,6 +657,26 @@ class TaskPlaneStore:
raise NotFoundError(f"task {task_id} not found")
return self._row_to_task(row)
def assert_root_open_for_child(self, root_task_id: str) -> TaskRecord:
"""Fence child admission against root completion/action finalization."""
root = self.get_task(root_task_id)
if root.task_id != root.root_task_id:
raise ConflictError("child admission authority is not a root task")
if root.status == "COMPLETED":
raise ConflictError("cannot add a child below a completed root")
fence = self.conn.execute(
"SELECT active_phase FROM task_root_action_fences WHERE root_task_id=?",
(root_task_id,),
).fetchone()
if fence is not None and fence["active_phase"] in {
"CLAIMED",
"EFFECT_UNKNOWN",
"EXECUTED_PENDING_REREVIEW",
"ROOT_CLOSING",
}:
raise ConflictError("root action fence blocks child admission")
return root
def list_tasks(self) -> list[TaskRecord]:
with self._lock:
rows = self.conn.execute(
@@ -1299,7 +1467,14 @@ class TaskPlaneStore:
return updated
def delegate_task_with_budget_and_events(
self, parent_task_id: str, child_plan: TaskPlan, *, caller: str
self,
parent_task_id: str,
child_plan: TaskPlan,
*,
caller: str,
owner_principal_id: str = "",
mission_id: str = "",
created_at: float | None = None,
) -> TaskRecord:
with self._transaction():
parent_budget = self.get_task_budget(parent_task_id)
@@ -1318,7 +1493,12 @@ class TaskPlaneStore:
self.consume_budget(parent_task_id, needed)
self._fault("delegate.after_budget_before_child")
child_plan.parent_task_id = parent_task_id
child = self.create_task(child_plan)
child = self.create_task(
child_plan,
owner_principal_id=owner_principal_id,
mission_id=mission_id,
created_at=created_at,
)
self.append_event(
parent_task_id,
"TASK_DELEGATED",
@@ -1641,394 +1821,6 @@ class TaskPlaneStore:
).fetchone()
return row["parent_task_id"] if row else None
def get_active_claim_count(self, worker_id: str) -> int:
"""Number of currently active (non-expired) claims for a worker."""
with self._lock:
row = self.conn.execute(
"SELECT COUNT(1) FROM claims WHERE worker_id=? AND status='ACTIVE' AND lease_until>?",
(worker_id, _now()),
).fetchone()
return int(row[0])
# -- governed proof / receipt persistence --------------------------------
def save_schedulability_proof(self, proof) -> str:
"""Persist an upfront PENDING proof snapshot (idempotent upsert)."""
if hasattr(proof, "to_dict"):
proof_dict = proof.to_dict()
else:
proof_dict = dict(proof)
envelope = {k: v for k, v in proof_dict.items() if k != "digest"}
digest = _hash_hex(_canonical(envelope))
now = _now()
with self._transaction():
self.conn.execute(
"""
INSERT INTO schedulability_proofs
(proof_id, task_id, worker_id, claim_id, status, envelope_json, digest, created_at, updated_at)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
ON CONFLICT(proof_id) DO UPDATE SET
claim_id=excluded.claim_id,
status=excluded.status,
envelope_json=excluded.envelope_json,
digest=excluded.digest,
updated_at=excluded.updated_at
""",
(
proof_dict["proof_id"],
proof_dict["task_id"],
proof_dict["worker_id"],
proof_dict.get("claim_id"),
proof_dict.get("status", "PENDING"),
json.dumps(envelope, ensure_ascii=False),
digest,
now,
now,
),
)
return proof_dict["proof_id"]
def _insert_proof(
self,
proof,
status: str,
*,
claim_id: str | None = None,
generation: dict | None = None,
observations: list | None = None,
now: float | None = None,
) -> None:
now = now if now is not None else _now()
envelope = proof.payload() if hasattr(proof, "payload") else dict(proof)
envelope = {k: v for k, v in envelope.items() if k != "digest"}
if claim_id is not None:
envelope["claim_id"] = claim_id
envelope["status"] = status
runtime = envelope.setdefault("runtime", {})
if generation is not None:
runtime["generation"] = generation
if observations is not None:
envelope["observations"] = observations
digest = _hash_hex(_canonical(envelope))
self.conn.execute(
"""
INSERT INTO schedulability_proofs
(proof_id, task_id, worker_id, claim_id, status, envelope_json, digest, created_at, updated_at)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
ON CONFLICT(proof_id) DO UPDATE SET
claim_id=excluded.claim_id,
status=excluded.status,
envelope_json=excluded.envelope_json,
digest=excluded.digest,
updated_at=excluded.updated_at
""",
(
envelope["proof_id"],
envelope["task_id"],
envelope["worker_id"],
envelope.get("claim_id"),
status,
json.dumps(envelope, ensure_ascii=False),
digest,
now,
now,
),
)
def get_schedulability_proof(self, proof_id: str) -> dict:
proof_id = _validated_id(proof_id, "proof_id")
with self._lock:
row = self.conn.execute(
"SELECT * FROM schedulability_proofs WHERE proof_id=?", (proof_id,)
).fetchone()
if row is None:
raise NotFoundError(f"proof {proof_id} not found")
envelope = json.loads(row["envelope_json"])
envelope["digest"] = row["digest"]
return envelope
def _insert_receipt(self, receipt) -> str:
if hasattr(receipt, "to_dict"):
receipt_id = receipt.receipt_id
kind = receipt.kind
task_id = receipt.task_id
claim_id = receipt.claim_id
proof_id = receipt.proof_id
worker_id = receipt.worker_id
digest = receipt.digest
payload = receipt.payload
created_at = receipt.created_at
else:
data = dict(receipt)
receipt_id = data["receipt_id"]
kind = data["kind"]
task_id = data["task_id"]
claim_id = data.get("claim_id")
proof_id = data.get("proof_id")
worker_id = data["worker_id"]
digest = data["digest"]
payload = data.get("payload", {})
created_at = data.get("created_at", _now())
self.conn.execute(
"INSERT INTO receipts(receipt_id, kind, task_id, claim_id, proof_id, worker_id, digest, payload_json, created_at) "
"VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)",
(
receipt_id,
kind,
task_id,
claim_id,
proof_id,
worker_id,
digest,
json.dumps(payload, ensure_ascii=False),
created_at,
),
)
return receipt_id
def save_receipt(self, receipt) -> str:
with self._transaction():
return self._insert_receipt(receipt)
def get_receipt(self, receipt_id: str) -> dict:
receipt_id = _validated_id(receipt_id, "receipt_id")
with self._lock:
row = self.conn.execute(
"SELECT * FROM receipts WHERE receipt_id=?", (receipt_id,)
).fetchone()
if row is None:
raise NotFoundError(f"receipt {receipt_id} not found")
result = dict(row)
result["payload"] = json.loads(row["payload_json"])
return result
def list_receipts(self, task_id: str) -> list[dict]:
task_id = _validated_id(task_id, "task_id")
with self._lock:
rows = self.conn.execute(
"SELECT * FROM receipts WHERE task_id=? ORDER BY created_at ASC",
(task_id,),
).fetchall()
result = []
for row in rows:
item = dict(row)
item["payload"] = json.loads(row["payload_json"])
result.append(item)
return result
# -- atomic governed claim ----------------------------------------------
def governed_claim(
self,
*,
task,
worker,
lease_seconds: int,
proof,
decision,
now: float | None = None,
):
"""Atomically claim via the governed path.
Within a single reentrant transaction: reuse the claim_task kernel
(CAS/lease/fencing/concurrency), persist the proof as CLAIMED with the
re-read evidence, write claim+proof receipts geared to read-back, and
append the CLAIM_RECHECK_ACCEPTED / CLAIM_GOVERNED_ACCEPTED events.
"""
now = now if now is not None else _now()
with self._transaction():
claim = self.claim_task(
task.task_id,
worker.worker_id,
worker.identity,
lease_seconds=lease_seconds,
expected_revision=task.revision,
operation_digest=task.task_plan.operation_digest(),
)
generation = {
"task_revision": claim.task_revision,
"attempt": claim.attempt,
"lease_generation": claim.attempt,
}
self._insert_proof(
proof,
"CLAIMED",
claim_id=claim.claim_id,
generation=generation,
observations=[c.to_dict() for c in decision.checks],
now=now,
)
claim_receipt = build_receipt(
kind="claim",
task_id=task.task_id,
worker_id=worker.worker_id,
claim_id=claim.claim_id,
proof_id=proof.proof_id,
extra={
"status": "CLAIMED",
"attempt": claim.attempt,
"task_revision": claim.task_revision,
"lease_until": claim.lease_until,
},
created_at=now,
)
proof_receipt = build_receipt(
kind="proof",
task_id=task.task_id,
worker_id=worker.worker_id,
claim_id=claim.claim_id,
proof_id=proof.proof_id,
extra={"status": "CLAIMED", "allowed": True, "recheck_id": decision.recheck_id},
created_at=now,
)
self._insert_receipt(claim_receipt)
self._insert_receipt(proof_receipt)
self.append_event(
task.task_id,
"CLAIM_RECHECK_ACCEPTED",
{
"proof_id": proof.proof_id,
"worker_id": worker.worker_id,
"claim_id": claim.claim_id,
"recheck_id": decision.recheck_id,
"allowed": True,
"generation": {"task_revision": task.revision, "attempt": None, "lease_generation": None},
},
subject=f"task/{task.task_id}",
)
self.append_event(
task.task_id,
"CLAIM_GOVERNED_ACCEPTED",
{
"claim_id": claim.claim_id,
"proof_id": proof.proof_id,
"worker_id": worker.worker_id,
"generation": generation,
},
subject=f"task/{task.task_id}",
)
return claim, claim_receipt.receipt_id
def record_tenant_skip_audit(
self,
task_id: str,
worker_id: str,
*,
task_tenant: str,
worker_tenant: str,
reason: str = "cross_tenant_skip",
now: float | None = None,
) -> dict[str, Any]:
"""Persist a structured 'enumeration=skip+audit' finding for a
cross-tenant task that a worker's enumeration was NOT allowed to see.
Contract (G4): enumeration of cross-tenant tasks must SKIP and record a
structured audit finding -- it must NOT raise / fail closed. Direct
claim remains fail-closed (403 cross_tenant_denied) elsewhere.
"""
now = now if now is not None else _now()
finding = {
"source": "tenant",
"kind": "cross_tenant_skip",
"decision": "skip+audit",
"task_tenant": task_tenant,
"worker_tenant": worker_tenant,
"task_id": task_id,
"worker_id": worker_id,
"reason": reason,
}
with self._transaction():
self.conn.execute(
"""
INSERT INTO tenant_skip_audit(task_id, worker_id, finding_json, created_at)
VALUES (?, ?, ?, ?)
ON CONFLICT(task_id, worker_id) DO UPDATE SET
finding_json=excluded.finding_json,
created_at=excluded.created_at
""",
(task_id, worker_id, json.dumps(finding, ensure_ascii=False), now),
)
self.append_event(
task_id,
"TASK_TENANT_SKIPPED",
finding,
subject=f"task/{task_id}",
)
return finding
def list_tenant_skip_audit(self, task_id: str | None = None) -> list[dict[str, Any]]:
"""Read-back of recorded cross-tenant skip findings (testability)."""
if task_id is None:
with self._lock:
rows = self.conn.execute(
"SELECT task_id, worker_id, finding_json, created_at FROM tenant_skip_audit ORDER BY created_at ASC"
).fetchall()
else:
task_id = _validated_id(task_id, "task_id")
with self._lock:
rows = self.conn.execute(
"SELECT task_id, worker_id, finding_json, created_at FROM tenant_skip_audit WHERE task_id=? ORDER BY created_at ASC",
(task_id,),
).fetchall()
out: list[dict[str, Any]] = []
for r in rows:
finding = json.loads(r["finding_json"])
out.append(
{
"task_id": r["task_id"],
"worker_id": r["worker_id"],
"finding": finding,
"created_at": r["created_at"],
}
)
return out
def record_proof_rejection(
self,
*,
proof,
decision,
task,
now: float | None = None,
) -> str:
"""Atomically persist a rejected proof + rejection receipt + event."""
now = now if now is not None else _now()
with self._transaction():
self._insert_proof(
proof,
"REJECTED",
observations=[c.to_dict() for c in decision.checks],
now=now,
)
proof_receipt = build_receipt(
kind="proof",
task_id=task.task_id,
worker_id=proof.worker_id,
proof_id=proof.proof_id,
extra={
"status": "REJECTED",
"allowed": False,
"reason": decision.reason,
"error_code": decision.error_code,
"recheck_id": decision.recheck_id,
},
created_at=now,
)
self._insert_receipt(proof_receipt)
self.append_event(
task.task_id,
"CLAIM_RECHECK_REJECTED",
{
"proof_id": proof.proof_id,
"worker_id": proof.worker_id,
"reason": decision.reason,
"error_code": decision.error_code,
"generation": {"task_revision": task.revision, "attempt": None, "lease_generation": None},
},
subject=f"task/{task.task_id}",
)
return proof_receipt.receipt_id
def _compute_depth(self, task_id: str) -> int:
depth = 0
visited = set()
@@ -1,168 +0,0 @@
"""toil_governor: Retry / Toil Governor 骨架。
实现 V4 §4 / V3 TrackA A2:在同 operation + 同 blocker、且无新证据的前提下抑制
重复重试;blocker 变化或 record_success 后重置降频;长期外部 blocker 自动降频。
UNKNOWN blocker 不直接放行,置 reconcile_required 标记由调用方先 reconcile。
本模块内存态(进程内),无持久化、无写操作、无网络。
"""
from __future__ import annotations
import hashlib
import json
import time
from dataclasses import dataclass, field
from typing import Optional
@dataclass(frozen=True)
class RetryGovernorDecision:
allow: bool
reason: str
next_eligible_at: Optional[float] = None
suppressed_count: int = 0
def to_dict(self) -> dict:
return {
"allow": self.allow,
"reason": self.reason,
"next_eligible_at": self.next_eligible_at,
"suppressed_count": self.suppressed_count,
"reconcile_required": "reconcile" in self.reason,
}
_UNKNOWN_BLOCKER_FP = "UNKNOWN"
class ToilGovernor:
"""内存中的重试抑制器。同指纹退化,blocker 变化重置。"""
def __init__(self, min_backoff: float = 2.0):
self._min_backoff = min_backoff
# operation -> 状态
self._state: dict[str, dict] = {}
# blocker 指纹 -> 连续失败次数(用于自动降频)
self._blocker_failures: dict[str, int] = {}
# ------------------------------------------------------------ key
def operation_key(self, fingerprint: str) -> str:
"""规范化 operation 指纹为稳定 key。"""
if not fingerprint:
raise ValueError("operation fingerprint must not be empty")
fp = fingerprint.strip()
if len(fp) <= 64:
return fp
digest = hashlib.sha256(fp.encode("utf-8")).hexdigest()[:16]
return f"{fp[:40]}..{digest}"
def blocker_fingerprint(self, blocker) -> str:
"""把 blocker(str / dict / object)规范化为指纹字符串。"""
if blocker is None:
return _UNKNOWN_BLOCKER_FP
if isinstance(blocker, str):
return _UNKNOWN_BLOCKER_FP if blocker.strip() in ("", "unknown") else blocker.strip()
if isinstance(blocker, dict):
try:
return json.dumps(blocker, sort_keys=True, ensure_ascii=False, default=str)
except Exception:
return _UNKNOWN_BLOCKER_FP
try:
return f"{type(blocker).__name__}:{hash(blocker)}"
except Exception: # pragma: no cover - 不可哈希对象兜底
return _UNKNOWN_BLOCKER_FP
# ------------------------------------------------------------ decide
def decide(
self,
operation,
blocker=None,
now: Optional[float] = None,
backoff_base: float = 60.0,
) -> RetryGovernorDecision:
"""判断本次重试是否放行。
- 同 operation 同 blocker 连续调用且无成功记录 → 抑制(allow=False),
下次可重试时间按 backoff_base * 2^(连续次数-1) 退避。
- blocker 指纹与上上次不同 → 视为新证据,放行并重置连续计数。
- blocker 为 UNKNOWN → 不直接放行,置 reconcile_required,返回 UNKNOWN 语义。
"""
when = now if now is not None else time.time()
op = self.operation_key(operation)
bfp = self.blocker_fingerprint(blocker)
state = self._state.setdefault(
op, {"blocker": None, "failures": 0, "last_allowed": None}
)
# —— UNKNOWN blocker:不直接放行,需 reconcile ——
if bfp == _UNKNOWN_BLOCKER_FP:
cur = self._blocker_failures.get(bfp, 0) + 1
self._blocker_failures[bfp] = cur
delay = backoff_base * (2 ** max(0, cur - 1))
return RetryGovernorDecision(
allow=False,
reason="reconcile_required: unknown blocker, do not bypass",
next_eligible_at=when + delay,
suppressed_count=cur - 1,
)
# —— blocker 变化(新证据)→ 放行并重置 ——
prev_bfp = state["blocker"]
if prev_bfp is not None and prev_bfp != bfp:
state["blocker"] = bfp
state["failures"] = 0
state["last_allowed"] = when
self._blocker_failures[bfp] = 1
return RetryGovernorDecision(allow=True, reason="blocker_changed")
# —— 同 blocker 连续 ——
if state["blocker"] is not None:
state["failures"] += 1
else:
state["blocker"] = bfp
state["failures"] = 0
state["last_allowed"] = when
self._blocker_failures[bfp] = 1
return RetryGovernorDecision(allow=True, reason="first_attempt")
failures = state["failures"]
# 长期外部 blocker 自动降频:连续失败次数越多,退避越久
delay = backoff_base * (2 ** max(0, failures - 1))
return RetryGovernorDecision(
allow=False,
reason="suppressed: same blocker, no new evidence",
next_eligible_at=when + delay,
suppressed_count=failures,
)
# ------------------------------------------------------------ record
def record_success(self, operation, now: Optional[float] = None) -> None:
"""记录一次成功,重置该 operation 的降频与连续计数。"""
op = self.operation_key(operation)
state = self._state.get(op)
if state is not None:
state["blocker"] = None
state["failures"] = 0
state["last_allowed"] = now if now is not None else time.time()
else:
self._state[op] = {
"blocker": None,
"failures": 0,
"last_allowed": now if now is not None else time.time(),
}
# ------------------------------------------------------------ stats
def stats(self) -> dict:
return {
"operations": len(self._state),
"blocker_fingerprints": len(self._blocker_failures),
"suppressed_by_blocker": sum(
1 for s in self._state.values() if s["failures"] > 0
),
"blocker_failure_counts": dict(self._blocker_failures),
}
-201
View File
@@ -1,201 +0,0 @@
"""Gate / Evidence Engine — verdict model and evaluation (WP5 / P0-5).
The verdict is the auditable result of checking whether a scope's required
evidence set is fully satisfied. The single governing rule is:
PASS <=> every required evidence for the scope is present.
Otherwise => NOT_VERIFIED. An explicit blocker => BLOCKED.
Proof / evidence entries are audit snapshots, NOT bearer credentials. A
service being "active", a file "existing", or a historical evidence entry
does NOT imply a PASS for the runtime/business scope. See
ARCH-AUTO-CLAIM-R2-20260914 §4.3.
"""
from __future__ import annotations
import time
from dataclasses import dataclass, field
from enum import StrEnum
from typing import Any, Iterable, Mapping
VALID_SCOPES: frozenset[str] = frozenset(
{"unit", "contract", "staging", "runtime", "business"}
)
class VerdictStatus(StrEnum):
"""Ordered fail-closed verdict states."""
PASS = "PASS"
FAIL = "FAIL"
NOT_VERIFIED = "NOT_VERIFIED"
BLOCKED = "BLOCKED"
DIVERGED = "DIVERGED"
@dataclass(frozen=True)
class EvidenceRequirement:
"""A single expected evidence entry for a scope."""
name: str
present: bool = False
detail: str = ""
@dataclass(frozen=True)
class Verdict:
"""Immutable audit record produced by evaluate_verdict."""
schema_version = "aios.verdict.v1"
scope: str
status: VerdictStatus
subject: str
checks: tuple[dict[str, Any], ...] = ()
evidence_refs: tuple[str, ...] = ()
missing_evidence: tuple[str, ...] = ()
reason: str = ""
decided_at: float = field(default_factory=time.time)
def to_dict(self) -> dict[str, Any]:
return {
"schema_version": self.schema_version,
"scope": self.scope,
"status": str(self.status.value),
"subject": self.subject,
"checks": list(self.checks),
"evidence_refs": list(self.evidence_refs),
"missing_evidence": list(self.missing_evidence),
"reason": self.reason,
"decided_at": self.decided_at,
}
# Marker names carried by an EvidenceRequirement that explicitly flag a blocker
# when present (fail-closed on blocked conditions even inside the evidence set).
BLOCKER_NAME_MARKERS = ("blocked", "blocker", "forbidden", "denied")
def _present_names(reqs: list[EvidenceRequirement]) -> set[str]:
return {req.name for req in reqs if req.present}
def _detect_blocker(reqs: list[EvidenceRequirement]) -> EvidenceRequirement | None:
for req in reqs:
lowered = req.name.lower()
if any(marker in lowered for marker in BLOCKER_NAME_MARKERS):
return req
return None
def evaluate_verdict(
*,
scope: str,
subject: str,
required_evidence: Iterable[EvidenceRequirement]
| Mapping[str, EvidenceRequirement]
| Iterable[str]
| None,
available_evidence: Iterable[str] | None = None,
reason: str = "",
blocked: bool = False,
now: float | None = None,
) -> Verdict:
"""Evaluate a scope's evidence closure into a Verdict.
PASS only when every required evidence is present. Missing evidence ->
NOT_VERIFIED. An explicit blocker (either `blocked=True` or a present
evidence name matching a blocker marker) -> BLOCKED. An unknown scope ->
FAIL (fail-closed / parameter error).
"""
now = now if now is not None else time.time()
if scope not in VALID_SCOPES:
return Verdict(
scope=scope,
status=VerdictStatus.FAIL,
subject=subject,
reason=reason or f"invalid scope: {scope!r}",
decided_at=now,
)
reqs: list[EvidenceRequirement] = _normalize_requirements(required_evidence)
available: set[str] = _normalize_available(available_evidence)
checks: list[dict[str, Any]] = []
present: set[str] = set()
missing: list[str] = []
for req in reqs:
is_present = req.present or (req.name in available)
checks.append(
{
"check": f"evidence:{req.name}",
"ok": is_present,
"detail": req.detail,
}
)
if is_present:
present.add(req.name)
else:
missing.append(req.name)
evidence_refs = tuple(
name for name in (req.name for req in reqs) if name in present
)
blocker = _detect_blocker(reqs)
if blocked or blocker is not None:
status = VerdictStatus.BLOCKED
effective_reason = reason or (
f"explicit blocked condition: {blocker.name}" if blocker else "blocked"
)
elif missing:
status = VerdictStatus.NOT_VERIFIED
effective_reason = reason or (
f"missing evidence: {', '.join(missing)}"
)
else:
status = VerdictStatus.PASS
effective_reason = reason or "all required evidence present"
return Verdict(
scope=scope,
status=status,
subject=subject,
checks=tuple(checks),
evidence_refs=evidence_refs,
missing_evidence=tuple(missing),
reason=effective_reason,
decided_at=now,
)
def _normalize_requirements(
required: Iterable[EvidenceRequirement]
| Mapping[str, EvidenceRequirement]
| Iterable[str]
| None,
) -> list[EvidenceRequirement]:
if required is None:
return []
if isinstance(required, Mapping):
return [
value if isinstance(value, EvidenceRequirement)
else EvidenceRequirement(name=key, present=bool(value), detail="")
for key, value in required.items()
]
out: list[EvidenceRequirement] = []
for item in required:
if isinstance(item, EvidenceRequirement):
out.append(item)
else:
out.append(EvidenceRequirement(name=str(item)))
return out
def _normalize_available(available: Iterable[str] | None) -> set[str]:
if available is None:
return set()
return {str(name) for name in available}
@@ -1,5 +1,10 @@
INSERT OR IGNORE INTO node_status(node_id, health, active, score, last_seen_at, note) VALUES ('server-hw','ok',1,100,datetime('now'),'primary logical node'),('server-local','ok',0,95,datetime('now'),'backup logical node');
INSERT OR IGNORE INTO routing_policy(role, primary_node, backup_node, active_node, failover_priority, updated_at) VALUES ('owner','server-hw','server-local','server-hw',1,datetime('now')),('admin','server-hw','server-local','server-hw',2,datetime('now')),('user','server-hw','server-local','server-hw',3,datetime('now'));
INSERT OR IGNORE INTO users(id, username, display_name, role, status, password_hash, created_at, updated_at) VALUES ('bootstrap-owner','owner','Bootstrap Owner','owner','approved','bootstrap-change-me',datetime('now'),datetime('now')),('bootstrap-admin','admin','Bootstrap Admin','admin','approved','bootstrap-change-me',datetime('now'),datetime('now')),('sample-user','sample-user','Sample User','user','approved','bootstrap-change-me',datetime('now'),datetime('now'));
-- SEC-P0-BOOTSTRAP-CREDENTIAL hardening: seed NO usable password. Empty hash
-- means "not set / forced reset" -> owner/admin must set a strong password via
-- the bootstrap flow (scripts/patch_runtime_owner_admin_user.py) over a secure channel.
INSERT OR IGNORE INTO users(id, username, display_name, role, status, password_hash, created_at, updated_at) VALUES ('bootstrap-owner','owner','Bootstrap Owner','owner','approved','',datetime('now'),datetime('now')),('bootstrap-admin','admin','Bootstrap Admin','admin','approved','',datetime('now'),datetime('now')),('sample-user','sample-user','Sample User','user','approved','',datetime('now'),datetime('now'));
-- One-time cleanup: neutralize any legacy known plaintext bootstrap seed already present.
UPDATE users SET password_hash='', updated_at=datetime('now') WHERE password_hash='bootstrap-change-me';
INSERT OR IGNORE INTO model_policy(role, default_model, allowed_models, updated_at) VALUES ('owner','deepseekv4-flash','["deepseekv4-flash","qwen3.6-35b","gpt-5.5"]',datetime('now')),('admin','deepseekv4-flash','["deepseekv4-flash","qwen3.6-35b"]',datetime('now')),('user','deepseekv4-flash','["deepseekv4-flash"]',datetime('now'));
INSERT OR IGNORE INTO quota_policy(username, daily_limit, used_today, updated_at) VALUES ('owner',10000,0,datetime('now')),('admin',2000,0,datetime('now')),('sample-user',200,0,datetime('now'));
+32 -4
View File
@@ -55,10 +55,36 @@ def seed_base_models():
if not one('SELECT * FROM gitops_versions WHERE version=?', ('V1R1C0',)):
exec_sql('INSERT INTO gitops_versions(version,tag,level,summary,status,created_by,created_at) VALUES(?,?,?,?,?,?,?)', ('V1R1C0','V1R1C0','R','MultiUser runtime selector + GitOps + custom LLM priority registry','ready','system',now()))
_ROLE_WHITELIST = {'owner', 'admin', 'user'}
def token_identity(token: str):
"""Resolve the server-authoritative identity bound to a token.
Format of each TOKEN_FILE line (one per line, ':'-separated):
<token> -> authenticated principal, role 'user'
<token>:<user>:<role> -> mapped principal; role restricted to owner/admin/user
Identity NEVER comes from a client-supplied header. In local dev mode the
DEV_ALLOW flag maps a fixed dev principal (must be explicitly enabled).
"""
if DEV_ALLOW:
return {'username': 'dev-owner', 'role': 'owner'}
if not token or not TOKEN_FILE.exists():
return None
for line in TOKEN_FILE.read_text().splitlines():
line = line.strip()
if not line:
continue
parts = [x.strip() for x in line.split(':')]
if not parts or parts[0] != token:
continue
if len(parts) >= 3:
role = parts[2] if parts[2] in _ROLE_WHITELIST else 'user'
return {'username': safe_name(parts[1]), 'role': role}
return {'username': 'user-%s' % token[:8], 'role': 'user'}
return None
def valid_token(token: str) -> bool:
if DEV_ALLOW: return True
if not token or not TOKEN_FILE.exists(): return False
return token.strip() in [x.strip() for x in TOKEN_FILE.read_text().splitlines() if x.strip()]
return token_identity(token) is not None
def git_cmd(args, check=False):
WORKSPACE.mkdir(parents=True, exist_ok=True); p = subprocess.run(['git'] + args, cwd=str(WORKSPACE), text=True, capture_output=True)
if check and p.returncode != 0: raise RuntimeError(p.stderr or p.stdout or 'git command failed')
@@ -88,8 +114,10 @@ class Handler(BaseHTTPRequestHandler):
server_version = 'AIOSGitOpsModelAPI/V1R1C0'
def actor(self):
auth = self.headers.get('Authorization',''); token = auth.replace('Bearer ', '').strip() if auth.startswith('Bearer ') else self.headers.get('X-AIOS-Token','')
ok = valid_token(token); user = self.headers.get('X-AIOS-User') or ('dev-owner' if DEV_ALLOW else 'anonymous'); role = self.headers.get('X-AIOS-Role') or ('owner' if DEV_ALLOW else 'user')
return {'username': safe_name(user), 'role': role, 'token_ok': ok}
ident = token_identity(token)
if ident is None:
return {'username': 'anonymous', 'role': 'user', 'token_ok': False}
return {'username': ident['username'], 'role': ident['role'], 'token_ok': True}
def require_write(self, roles=None):
a = self.actor()
if not a['token_ok']: self.send_json({'error':'unauthorized','hint':'write APIs require Bearer token or AIOS_GITOPS_DEV_ALLOW=1'}, 401); return None
+14 -29
View File
@@ -1,8 +1,8 @@
<<<<<<< HEAD
# AIOS Agent Command Plane Registry
# Git is the single source of truth for all agent commands.
version: 1.0.0
created_at: 2026-07-08T15:20:00+08:00
registry_version: 1.0.0
created_at: 2026-07-08T15:44:00+08:00
baseline: V1R3C1
commands:
@@ -12,6 +12,12 @@ commands:
completed_path: "agents/commands/completed/"
failed_path: "agents/commands/failed/"
directories:
commands/active: "New commands pending execution — Git is the only entry point"
commands/running: "Commands currently being executed by an agent"
commands/completed: "Commands that finished successfully — moved here on success"
commands/failed: "Commands that finished with errors — moved here on failure"
policies:
path: "agents/policies/"
source: "git"
@@ -31,6 +37,12 @@ agents:
type: reviewer
role: review
lifecycle:
new: active/
running: running/
success: completed/
failure: failed/
# Audit plane configuration
audit:
enabled: true
@@ -38,30 +50,3 @@ audit:
agent_runs: "audit/agent-runs/"
version_history: "audit/version-history/"
command_history: "audit/command-history/"
=======
# AIOS Agent Command Registry
# Git is the single source of truth for all agent commands.
registry_version: 1.0.0
created_at: "2026-07-08T15:44:00+08:00"
baseline: V1R3C1
directories:
commands/active: "New commands pending execution — Git is the only entry point"
commands/running: "Commands currently being executed by an agent"
commands/completed: "Commands that finished successfully — moved here on success"
commands/failed: "Commands that finished with errors — moved here on failure"
state: agents/state/
policies: agents/policies/
lifecycle:
new: active/
running: running/
success: completed/
failure: failed/
# Command lifecycle state machine:
# active/ → locked & moved to running/ → execute → success → completed/
# → failure → failed/
>>>>>>> 4d18ee0 (feat(cmd): enable Agent Command Plane, audit plane, and Web UI dashboard)
@@ -0,0 +1,68 @@
[Unit]
Description=AIOS GitOps Model API V2R2C2 Baseline1.1 isolated candidate (2852c2a) - control-plane bootstrap only
After=network-online.target
Wants=network-online.target
Requires=aios-trusted-candidate-builder.socket
[Service]
Type=simple
User=aiman
Group=gitops
WorkingDirectory=/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime
UMask=0077
Environment=PYTHONUNBUFFERED=1
Environment=PYTHONDONTWRITEBYTECODE=1
Environment=AIOS_GITOPS_ROOT=/var/lib/aios-gitops
Environment=AIOS_GITOPS_WORKSPACE=/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime
Environment=AIOS_GITOPS_REPO=/data/aios-gitops/repo.git
Environment=AIOS_GITOPS_BARE_REPO=/data/aios-gitops/repo.git
Environment=AIOS_GITOPS_AUTHORITATIVE_GIT_DIR=/data/aios-gitops/repo.git
Environment=AIOS_GITOPS_REPOSITORY_IDENTITY=aios/aios
Environment=AIOS_GITOPS_BASELINE_AUTHORITY=/var/lib/aios-gitops-runtime-anchor/baseline/current.json
Environment=AIOS_GITOPS_DURABLE_JOB_EXECUTION_ENABLED=false
Environment=AIOS_GITOPS_JOB_CONTROLLER_ENABLED=false
Environment=AIOS_GITOPS_WORKSPACE_ROLE=runtime-anchor
Environment=AIOS_GITOPS_RUNTIME_OID=2852c2a40394195192c6c02a215ba31764e2cdda
Environment=AIOS_GITOPS_ARTIFACT=/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime/governed-runtime-package.json
Environment=AIOS_GITOPS_ARTIFACT_SHA256=932431dbbdee55522342d9c94457f40d8f6f7e4fc7eaadc71c66d0a6bdf8472f
Environment=AIOS_GITOPS_DEPLOYMENT_STATE=CANDIDATE_ISOLATED
Environment=AIOS_GITOPS_MODEL_API_SERVICE=aios-gitops-model-api-v2r2c2-baseline11-candidate.service
Environment=AIOS_RUNTIME_CANDIDATE_TREE_OID=ccd67c2357e7a4cd74e06be61ab81bba820d7671
Environment=AIOS_RUNTIME_PROVENANCE_COMMIT=2852c2a40394195192c6c02a215ba31764e2cdda
Environment=AIOS_AUTHORITY_SOURCE_PUBLISHED=true
Environment=AIOS_GITOPS_WRITES_ENABLED=false
Environment=AIOS_GITOPS_LEGACY_GLOBAL_WRITES_ENABLED=false
Environment=AIOS_SILENT_RELEASE_EXECUTION_ENABLED=false
Environment=AIOS_SILENT_RELEASE_STATE_ROOT=/var/lib/aios-gitops/model-api-v2r2c2-baseline11-candidate/silent-release
Environment=AIOS_BRIDGE_V2_URL=http://127.0.0.1:18892
Environment=AIOS_GITOPS_HOST=127.0.0.1
Environment=AIOS_GITOPS_PORT=18897
Environment=PYTHONPATH=/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime:/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime/app:/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime/runtime-agent-os
Environment=AIOS_TRUSTED_CANDIDATE_BUILDER_ENABLED=true
Environment=AIOS_TRUSTED_CANDIDATE_BUILDER_SOCKET=/run/aios-trusted-candidate-builder.sock
Environment=AIOS_TRUSTED_CANDIDATE_BUILDER_POLICY=/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime/config/gitops/trusted-candidate-builder-policy.json
Environment=AIOS_TRUSTED_CANDIDATE_BUILDER_PUBLIC_KEYRING=%d/builder_public_keyring
Environment=AIOS_TRUSTED_CANDIDATE_BUILDER_CREDENTIALS_DIRECTORY=%d
Environment=AIOS_TRUSTED_BUILDER_SOURCE_ADMISSION_ROOT=/var/lib/aios-release-source-admission
Environment=AIOS_SILENT_RELEASE_TRUSTED_POLICY=/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime/config/gitops/trusted-release-policy.json
LoadCredentialEncrypted=builder_hmac:/etc/credstore.encrypted/aios-trusted-candidate-builder/builder_hmac
LoadCredentialEncrypted=source_admission_hmac:/etc/credstore.encrypted/aios-release-source-admission/source_admission_hmac
LoadCredential=builder_public_keyring:/etc/aios-gitops/trusted-builder/public-keyring.json
ExecStart=/usr/bin/python3 -B /opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime/app/aios_model_api_staging.py
Restart=on-failure
RestartSec=2
NoNewPrivileges=true
PrivateTmp=true
ProtectHome=true
ProtectSystem=strict
ProtectKernelTunables=true
ProtectKernelModules=true
ProtectControlGroups=true
LockPersonality=true
RestrictSUIDSGID=true
RestrictAddressFamilies=AF_UNIX AF_INET
ReadOnlyPaths=/opt/aios/gitops-model-api/2852c2a40394195192c6c02a215ba31764e2cdda/runtime /data/aios-gitops/repo.git /etc/aios-gitops/trusted-builder/public-keyring.json /var/lib/aios-release-source-admission /var/lib/aios-release-migration
ReadWritePaths=/var/lib/aios-gitops/model-api-v2r2c2-baseline11-candidate
[Install]
WantedBy=multi-user.target
-76
View File
@@ -1,76 +0,0 @@
{
"schema": "aios.system-truth-contract.v1",
"state_ownership": {
"TaskAdmission": {
"owns": [
"envelope",
"authorization",
"grant",
"admission_receipt"
],
"must_not_own": [
"claim",
"heartbeat",
"execution_status",
"commit_result"
]
},
"TaskPlane": {
"owns": [
"task_state",
"claim",
"lease",
"heartbeat",
"attempt",
"deadline"
],
"must_not_own": [
"owner_signature",
"git_commit_truth"
]
},
"GitAuthority": {
"owns": [
"branch_oid",
"commit_oid",
"file_bytes"
],
"must_not_own": [
"task_state"
]
},
"RuntimeProjection": {
"owns": [
"deployed_source_oid",
"served_routes",
"external_action_readback"
],
"must_not_own": [
"source_commit_status"
]
}
},
"completion_requires": [
"valid_claim_holder",
"fresh_worker_heartbeat",
"valid_writer_lease",
"commit_receipt",
"post_oid_differs_from_pre_oid",
"output_sha256_matches_plan",
"runtime_readback",
"external_projection_readback"
],
"queue_requires": [
"fresh_schedulability_proof",
"at_least_one_healthy_scope_matching_worker",
"grant_or_jit_principal_can_cover_claim_window"
],
"forbidden_equivalences": [
"ADMITTED == EXECUTED",
"QUEUED == CLAIMED",
"service_active == scenario_pass",
"drift_status_PASS == evidence_complete",
"hash_declared == artifact_exists",
"OpenAPI_operation == backend_route_exists"
]
}
@@ -0,0 +1,28 @@
{
"schema": "aios.unified-platform-program-contract.v1",
"goal_slug": "AIOS-UNIFIED-PLATFORM-CONVERGENCE",
"authority_base": "beabd053961047f170e87e1110f2636dbd50f5fb",
"existing_identity_lineage_dependency": {
"mission_id": "mission-f3af205b-7dd6-455f-a561-88f0294cc815",
"root_task_id": "task-133ad96e-e0d7-4202-934f-80c75053b11f",
"acceptance_mutation_allowed": false
},
"tracks": {
"A": "IDENTITY_LINEAGE_EXISTING_MISSION",
"B": "ANNEX_2_3_R2_AUTO_CLAIM",
"C": "PLATFORM_GENERATION_G0",
"D": "GOVERNED_MERGE_STAGING_RELEASE"
},
"program_acceptance": [
"U01",
"U02",
"U03",
"U04",
"U05",
"U06",
"U07",
"U08"
],
"deployment_requires_review_go": true,
"new_business_authority_created": false
}
+1
View File
@@ -0,0 +1 @@
{"bindings":[{"acceptance_key":"C01","acceptance_position":0,"lineage_request_digest":"e9946481e7c9fe8dee94aa42ffa0a9b41546ce9d46601ff221a958710c72f6cc","mission_id":"mission-f3af205b-7dd6-455f-a561-88f0294cc815","owner_principal_id":"local-owner-authority","parent_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","root_acceptance_ref":"mission-acceptance:mission-f3af205b-7dd6-455f-a561-88f0294cc815:identity-lineage-live-closure-v1","root_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","route":"GOVERNED_GIT","schema":"aios.autonomous-child-binding.v1","sibling_sequence":1,"task_id":"task-1668131d-7484-4c8e-83b4-2cf9927842fe","task_plan_digest":"3f0824f51d824de7d38821205c012c2a23ac11e9f5889b8b0aafb07cce85333f"},{"acceptance_key":"C02","acceptance_position":1,"lineage_request_digest":"84f28d350ef3a2dce47fb5f3163b61bfd329ce5fe85bd59c886d4813e806b602","mission_id":"mission-f3af205b-7dd6-455f-a561-88f0294cc815","owner_principal_id":"local-owner-authority","parent_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","root_acceptance_ref":"mission-acceptance:mission-f3af205b-7dd6-455f-a561-88f0294cc815:identity-lineage-live-closure-v1","root_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","route":"SEMANTIC_IMPORT","schema":"aios.autonomous-child-binding.v1","sibling_sequence":2,"task_id":"task-24fe3d9a-783e-4a68-bf9c-5b1a9d2b56c7","task_plan_digest":"e01680583b8cc2ec3d33026dda9f7d095692d0ae79fc43f380066511566e305e"},{"acceptance_key":"C03","acceptance_position":2,"lineage_request_digest":"148fa8b55224649ac80f4d9d5c344019dfef24b1d5ed614ef62ee71ccaf8a6d3","mission_id":"mission-f3af205b-7dd6-455f-a561-88f0294cc815","owner_principal_id":"local-owner-authority","parent_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","root_acceptance_ref":"mission-acceptance:mission-f3af205b-7dd6-455f-a561-88f0294cc815:identity-lineage-live-closure-v1","root_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","route":null,"schema":"aios.autonomous-child-binding.v1","sibling_sequence":3,"task_id":"task-78e0dff5-c1fe-491a-99d2-81ad5cf180da","task_plan_digest":"aef0c0cfee2686ad7bff122328408c5094370bf29fac37a49ddc50173fa60986"},{"acceptance_key":"C04","acceptance_position":3,"lineage_request_digest":"927dde8068b9dde4d57e0f9eb6bd52d95d761533974a7437d01707ea3c5c807d","mission_id":"mission-f3af205b-7dd6-455f-a561-88f0294cc815","owner_principal_id":"local-owner-authority","parent_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","root_acceptance_ref":"mission-acceptance:mission-f3af205b-7dd6-455f-a561-88f0294cc815:identity-lineage-live-closure-v1","root_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","route":null,"schema":"aios.autonomous-child-binding.v1","sibling_sequence":4,"task_id":"task-9b6e0cb1-c54a-4f91-8762-7c72112cf71d","task_plan_digest":"d1eae4e90a5892dd851fbe80c2ee0920e7060cd6ce701ae0522d110dcd9af638"},{"acceptance_key":"C05","acceptance_position":4,"lineage_request_digest":"4336de284be15d504af661c18a98c820da3c3518ef5df25d4607c19c3ba724b2","mission_id":"mission-f3af205b-7dd6-455f-a561-88f0294cc815","owner_principal_id":"local-owner-authority","parent_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","root_acceptance_ref":"mission-acceptance:mission-f3af205b-7dd6-455f-a561-88f0294cc815:identity-lineage-live-closure-v1","root_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","route":null,"schema":"aios.autonomous-child-binding.v1","sibling_sequence":5,"task_id":"task-bd8c44c5-7081-4a3f-a3b0-4de4ad6e3b29","task_plan_digest":"dd7e02ef51f601210295c9bda1061dcd7ff944bc7a5dcb2e813db2d601b138a4"},{"acceptance_key":"C06","acceptance_position":5,"lineage_request_digest":"54bdc5cf4414bc52105e966072009c9b0ce8235392f767b38e856072dd01148c","mission_id":"mission-f3af205b-7dd6-455f-a561-88f0294cc815","owner_principal_id":"local-owner-authority","parent_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","root_acceptance_ref":"mission-acceptance:mission-f3af205b-7dd6-455f-a561-88f0294cc815:identity-lineage-live-closure-v1","root_task_id":"task-133ad96e-e0d7-4202-934f-80c75053b11f","route":null,"schema":"aios.autonomous-child-binding.v1","sibling_sequence":6,"task_id":"task-591a87fa-e722-4647-a71d-80624b7e6b6a","task_plan_digest":"9b0d092f6a0a987844edae129eaedfc6bed87c135d8d3db390b0f01b43f896c8"}],"policy_version":"identity-lineage-autonomous-next-child-v1","schema":"aios.autonomous-child-binding-set.v1"}
@@ -0,0 +1,96 @@
# G7 — Security-Critical Closure 证据与判定
- **Gate**: G7_SECURITY_CRITICAL_CLOSURE
- **判定**: `G7_CHANGES_REQUIRED`(修复已就绪,等待上游治理合入 + staging_readback + ROOT_REVIEW 后升为 APPROVED_FOR_GOVERNED_SUBMISSION)
- **Authority HEAD (base OID)**: `beabd053961047f170e87e1110f2636dbd50f5fb`
- **工作分支**: `agent/g7-security-critical-closure-20260914`(基于 origin/main;本地修复分支,待 PR + Owner 合入 main,不绕过保护 main)
- **日期**: 2026-09-14
- **扫描源**: `aios-scan-full.json` (1.08MB, 1039 findings, 高召回含误报)
- **规模**: 729 文件 / 24 个指派文件修改 / 5 个新增负面测试套件
## 摘要
全仓 review 为高召回设计并含误报,故未机械修复全部 1039 条发现。本次仅闭环被
权威源码复核为 CONFIRMED 且直接影响控制面授权、凭证安全、任意文件访问、特权 UI
的 **P0 阻塞项(6 项)+ P1 交付正确性包(若干)**。DEAD_PATH/FALSE_POSITIVE 项已
按契约排除,并记录依据。
## 修复分账(S1–S5)
| 阻塞项 | 角色 | 分类 | 文件 | pre→post SHA | 修复要点 |
|---|---|---|---|---|---|
| SEC-P0-AUTH-ROLE-HEADER | S1 | CONFIRMED | adapters/gitops/aios_gitops_model_api.py | 60396e0→8990317 | role/user 由服务端 token→identity 映射派生,忽略 X-AIOS-User/Role 头 |
| SEC-P0-AUTH-ROLE-HEADER | S1 | CONFIRMED | active-runtime/app/aios_gitops_model_api.py | ac8c1bf→80b04bd | 同上 |
| SEC-P0-TASK-ACTION-ROLE | S1 | CONFIRMED | active-runtime/app/task_action_api.py | b18e0a5→31c6f6e | 忽略 body role,role 由注入认证主体派生,默认 user |
| SEC-P0-OBSERVABILITY-ROLE | S1 | CONFIRMED | active-runtime/app/agent_observability_api.py | f516ef9→6b065fb | 认证 tenant 决定查询范围,客户端 role=owner 被忽略 |
| SEC-P0-BOOTSTRAP-CREDENTIAL | S2 | CONFIRMED | active-runtime/migrations/002_seed_owner_admin_user.sql | 7fff9e5→4cb5a33 | 种子空哈希(强制重置),清除 bootstrap-change-me |
| SEC-P0-BOOTSTRAP-CREDENTIAL | S2 | CONFIRMED | active-runtime/app/aios_webops_app.py | 83387a8→7001142 | PBKDF2 随机盐;SECRET 缺失 fail-closed;拒绝明文回退;会话 HMAC+过期 |
| SEC-P0-BOOTSTRAP-CREDENTIAL | S2 | CONFIRMED | active-runtime/app/aios_v5_webops_app.py | 83387a8→7001142 | 同上 |
| SEC-P0-CONTROL-PLANE-FILE-READ | S3 | CONFIRMED | active-runtime/app/aios_chatgpt_actions_api.py | 47ae2c8→4f345a0 | 命令读取:认证+严格标识+包含校验+1MiB 上限 |
| SEC-P0-CONTROL-PLANE-FILE-READ | S3 | CONFIRMED | active-runtime/app/aios_model_api_staging.py | 1616b37→e94d0dc | 同上 |
| SEC-P1-PATH-CONTAINMENT | S3 | CONFIRMED | services/aios-event-ingress/server.py | c790d41→c4e93cb | ID 白名单+resolve 包含校验 |
| SEC-P1-PATH-CONTAINMENT | S3 | CONFIRMED | active-runtime/app/agent_observability_store.py | 40cc555→4746b3a | execution_id 白名单+包含校验 |
| SEC-P1-PATH-CONTAINMENT | S3 | CONFIRMED | active-runtime/app/aios_campaign_recovery.py | 0d09fd2→06e208f | 证据路径 resolve+realpath 包含校验 |
| SEC-P1-PATH-CONTAINMENT | S3 | CONFIRMED | scripts/gitops/task_worktree.sh | 72d33f4→19ea429 | TASK_ID 正则收紧,拒绝 .. ../ ..\ |
| SEC-P1-PATH-CONTAINMENT | S3 | CONFIRMED | scripts/init_gitops_baseline.sh | 472e60c→48f584e | zip-slip 解包前条目校验 |
| SEC-P1-PATH-CONTAINMENT | S3 | CONFIRMED | ops/gitops-migration/sync_gitea_owner_password.sh | 11573c9→8e85831 | 消除 env 来源注入,argv+引号 heredoc |
| SEC-P0-ADMIN-UI-XSS | S4 | CONFIRMED | active-runtime/app/static/admin-ai-evolution.html | 6da3ce8→05d26ab | textContent 全面转义;修复 JS 语法;用真实 governed 审批端点替代本地 fake Approve |
| SEC-P1-WEB-CONTROL-SURFACE | S4 | CONFIRMED | overlays/.../gitops-dashboard/index.html | 7f459a1→484447b | 客户端 CSRF token 传播;状态变更带 token;branch 服务端校验 |
| SEC-P1-RUNTIME-DELIVERY | S5 | CONFIRMED | agents/registry.yaml | 7b69d96→0037e73 | 解决阻塞性 merge conflict,恢复可解析 YAML |
| SEC-P1-RUNTIME-DELIVERY | S5 | CONFIRMED | active-runtime/app/aios_model_api_staging.py | 1616b37→e94d0dc | 消除 do_POST 遮蔽(合并 bridge 路由) |
| SEC-P1-RUNTIME-DELIVERY | S5 | CONFIRMED | docs/gitops/openapi-chatgpt-actions.yaml | 5278b91→9f339c0 | 补全 securitySchemes.taskPrincipalAuth |
| SEC-P1-RUNTIME-DELIVERY | S5 | CONFIRMED | deploy/install/install.sh | ab1f433→b90b759 | 移除 `\|\| true`,启动失败传播 |
| SEC-P1-RUNTIME-DELIVERY | S5 | CONFIRMED | scripts/silent_install_v5.sh | 59e9552→e26a3b3 | 缺失脚本显式报错 |
| SEC-P1-RUNTIME-DELIVERY | S5 | CONFIRMED | delta-fix/apply_delta_fix_only.sh | de1e2fd→d09d2ab | 修复未定义变量 |
| SEC-P1-RUNTIME-DELIVERY | S5 | **DEAD_PATH** | active-runtime/app/architecture_owner_gate_http.py | 未改 | 全库零引用,非活路径 |
| SEC-P1-RUNTIME-DELIVERY | S5 | **DEAD_PATH** | runtime/agent-os/api.py | 未改 | 全库零引用独立模块,facade 签名不符但无活路径 |
## 负面测试(S6 独立 QA 验证)
新增 5 个测试套件,均针对**真实已修复模块**(经 tests/conftest.py 的 active_runtime 命名空间映射到实际文件):
- `tests/test_g7_auth_negative.py` — 9 passed
- `tests/test_g7_credential_negative.py` — 4 passed
- `tests/test_g7_path_containment_negative.py` — 7 passed(+8 bash 用例如下)
- `tests/test_g7_runtime_delivery.py` — 4 passed
- `tests/test_g7_web_surface_negative.py` — 42 passed
**合计 66 passed / 0 failed**。8 个 bash 子进程用例如 Windows 沙箱 pytest 无法拉起 bash 而 skip,
已在本机 Bash 环境实测真实脚本:遍历 TASK_ID → rc=65 拒绝;zip-slip `../` 条目 → BLOCKED 退出 11。
### 覆盖的负面场景
- spoof `X-AIOS-Role: owner` / `X-AIOS-User` → 不能提权
- JSON `role=owner` → 被忽略
- observability `role=owner` + 外来 user/group → 限定自身
- 非 owner token 下 `require_write(['owner'])` → 403
- 已知 bootstrap 口令拒绝登录;哈希非明文且随机盐;缺 SECRET 拒绝启动;过期令牌失效
- POSIX `../` / Windows `..\` / 编码遍历 / 超大读取 / zip-slip / 绝对路径拒绝
- HTML `<img onerror>` / script 载荷保持文本;CSRF-less 状态变更被拒;无效 branch 被拒
- registry.yaml / openapi 可解析且引用完整;install.sh 失败传播;脚本无未定义变量
### 回归
- 全部 `active-runtime/app` + `adapters` 211 个 py 文件 `py_compile` 通过(无导入/Syntax 回归)
- 全部 15 个改动/新增 py 文件编译通过
## 证据字段映射(G7_SECURITY_GATE.md Required evidence)
对每个阻塞项均已具备:finding_id / authority_oid(=beabd05…) / affected_path / live_path=true /
preimage_sha256(上表)、fix_commit_oid(见下方提交)、postimage_sha256(上表)、
negative_tests(通过)、positive_tests(编译+回归)、independent_review(S6 独立运行通过)、
runtime/staging_readback、rollback(见 S2 迁移/回滚路径)。
## 判定与下一步
- **当前判定**: `G7_CHANGES_REQUIRED` → 修复就绪,待合入流程推进。
- **下一步**(需 Owner/治理,非本 Agent 越权执行):
1. 审查并合入 `agent/g7-security-critical-closure-20260914`(走 protected main PR + Owner approval)
2. staging 部署后作 runtime/staging_readback(真实服务启动验证,尤其 S2 的 AIOS_WEBOPS_SECRET 注入与 S1 的 token 映射文件格式)
3. 执行 S2 owner 引导设密流程(scripts/patch_runtime_owner_admin_user.py)设置真实 owner 强口令
4. 独立 final review → governed submission → staging → live E2E → production → ROOT_REVIEW
- **已知未闭环**(诚实标注): S1 的 task_action/observability 认证主体需宿主 FastAPI 应用注入
`set_authenticator()`,当前默认按最低权限 user(安全降级但 owner 能力需接线);staging_readback
与 rollback 实机验证需在部署环境执行,本环境未具备。
## 备注(Triage 纪律)
- 未机械处理 1000+ 条 medium/low/维护性/样式/历史死代码注释(契约明确非当前发布阻塞项)。
- DEAD_PATH 项未改动并给出全库引用证据。
+1 -1
View File
@@ -15,5 +15,5 @@ bash "$PKG_ROOT/scripts/smoke_test.sh" "$AIOS_RUNTIME_ROOT"
python3 "$PKG_ROOT/scripts/generate_release_manifest.py" "$AIOS_RUNTIME_ROOT"
echo "DELTA_FIX_ONLY_APPLY_OK release=$AIOS_RELEASE_ID runtime=$AIOS_RUNTIME_ROOT"
python3 "$PKG/fixes/patch_runtime_v5_guards.py" "$TARGET"
python3 "$PKG_ROOT/fixes/patch_runtime_v5_guards.py" "$AIOS_RUNTIME_ROOT"
echo "V5_DIGITAL_LIFE_DELTA_APPLIED"
+4 -4
View File
@@ -33,11 +33,11 @@ cp deploy/systemd/aios-autonomous-daemon.service /etc/systemd/system/
cp deploy/systemd/aios-v1r1-wechat-worker.service /etc/systemd/system/
systemctl daemon-reload
systemctl enable aios-autonomous-daemon.service || true
systemctl enable aios-v1r1-wechat-worker.service || true
systemctl enable aios-autonomous-daemon.service
systemctl enable aios-v1r1-wechat-worker.service
systemctl restart aios-autonomous-daemon.service || true
systemctl restart aios-v1r1-wechat-worker.service || true
systemctl restart aios-autonomous-daemon.service
systemctl restart aios-v1r1-wechat-worker.service
sleep 2
systemctl status aios-autonomous-daemon.service --no-pager || true
+51
View File
@@ -0,0 +1,51 @@
# ADR: R2 evidence digests reuse TaskPlane canonical JSON
Status: ACCEPTED FOR R2 REVIEW CANDIDATE
Date: 2026-09-14
## Context
R1 introduced local sorted/separator JSON encoders in schedulability proof and
Retry/Toil fingerprint code. Platform Baseline V4 §1.3/§4.4/§12 prohibits new
parallel canonical encoders unless a scoped ADR proves the existing contract is
not applicable.
The candidate already carries the established TaskPlane contract:
`app.task_plane.models.canonical_json(value)`
which serializes UTF-8 JSON deterministically with sorted keys and compact
separators.
## Decision
R2 reuses `task_plane.models.canonical_json` for:
1. schedulability observation/proof digests;
2. Retry/Toil operation fingerprint payloads.
Retry/Toil adds a domain schema wrapper
`aios.retry-toil-fingerprint.v1` before hashing so the digest domain is
separated without inventing a new byte encoder.
No generic canonical platform, second encoder, or new write authority is added.
## Bytes and algorithm
- canonical text producer: `task_plane.models.canonical_json`
- bytes: UTF-8 encoding of that exact text
- digest: SHA-256 hex
- schedulability schema: `aios.schedulability-proof.v2`
- retry fingerprint schema: `aios.retry-toil-fingerprint.v1`
## Migration / compatibility
These R2 modules are not yet live authorities. Their R1 digests are review
candidate evidence only and do not require backwards-compatible production
decoding. Any future persisted proof/fingerprint consumer must bind schema
version explicitly.
## Guard
R2 tests reject a reintroduction of local `json.dumps(sort_keys=True,
separators=...)` encoders in the two modules.
+232
View File
@@ -0,0 +1,232 @@
# AIOS Unified Platform Convergence — Architecture Change V2
Status: R2 REVIEW CANDIDATE
Date: 2026-09-14
Authority base for rebind: `beabd053961047f170e87e1110f2636dbd50f5fb`
## 1. Purpose
R2 preserves the R1 architectural direction but corrects the eight review
findings F01–F08. It does not authorize deployment.
The program still has four tracks:
1. existing Identity / Lineage S3;
2. ANNEX-2.3 R2 Auto-Claim / Execution Integrity;
3. Platform Generation G0 (WP0a–e/WP1/WP2 only);
4. governed merge, staging, release and root/program acceptance.
Existing Identity/Lineage mission/root/children remain authoritative and are
not recreated or redefined by this program.
## 2. Authority split
- Git authority main: `beabd053961047f170e87e1110f2636dbd50f5fb`
- release-control runtime: `3c320aeb89d9ba8f0ed516abeb57252682b18983`
- production V2R2C2: `4642b74af62c89a505ae49241e805a5a88ca897f`
A review candidate is not a governed commit. All target preimages must be
refreshed from live Git authority before submission.
## 3. R2 safety corrections
### 3.1 Caller evidence is fail-closed
`caller_scan v2` separates:
- static source caller evidence;
- active runtime caller evidence;
- test-only evidence;
- opaque/dynamic uncertainty;
- complete negative evidence.
Static source imports alone do not prove runtime LIVE.
`PROVEN_ORPHAN` requires all of:
- zero production static caller;
- zero active systemd/API/scheduler/runtime-loaded caller;
- relevant runtime host set non-empty and completely scanned;
- loaded bytes negative proof;
- systemd/API/scheduler scan coverage;
- complete observation window;
- no probe failures/opaque dynamic references;
- rollback reference;
- independent acceptance `GO`.
Incomplete negative evidence returns `LIKELY_ORPHAN`; probe failure, opaque
dynamic import, or unresolved production source caller returns `UNKNOWN`.
UNKNOWN can never drive Sunset.
### 3.2 Schedulability proof is diagnostic, final claim is authoritative
R2 checks task lifecycle state and treats explicit `max_concurrency=0` as zero.
Invalid capacity values are structured blockers.
The proof remains non-bearer. Claim has three stages:
1. preflight observation;
2. claim-time read-only recheck;
3. existing TaskPlane authority performs `claim_atomically` at the
linearization point.
The atomic adapter contract must revalidate task revision/generation/status,
worker generation/identity/health/capacity, authorization ref/revision,
scope, dependencies, freeze and conflicting lease inside the same
authoritative transaction/fenced operation that creates the claim.
The returned receipt must include claim id, lease id, fencing generation and
receipt revision.
R2 reference tests verify the adapter contract only. They do **not** prove a
production adapter is atomic; real staging A18–A23 remain mandatory.
### 3.3 Retry Governor is whitelist-only
- UNKNOWN/indeterminate → RECONCILE;
- governance/dependency boundary → BLOCK;
- proven success → SUPPRESS/REUSE;
- unrecognized state → BLOCK;
- only explicitly executable states may return EXECUTE.
A prior success for the same operation conflicting with a later NOT_APPLIED
observation must reconcile rather than re-execute.
### 3.4 Canonical bytes are reused
Schedulability and Retry/Toil reuse
`app.task_plane.models.canonical_json`; no second JSON canonical encoder is
introduced. See `ADR_CANONICAL_EVIDENCE_DIGEST_V1.md`.
### 3.5 Platform G0 discovery is not G0 acceptance
WP0a–d and WP1 now expose completeness/missing evidence. Lexical/static
discovery cannot by itself satisfy Generation-G0 Exit.
See `G0_EVIDENCE_COVERAGE_V2.md`.
## 4. Non-circular review/deploy gates
R1's production gate was circular because existing root C06 and Program
production evidence could be required before production itself.
R2 replaces it with:
```text
REVIEW_READY
↓
STAGING_ACTIVATION_READY
↓
STAGING_ACCEPTED
↓
PRODUCTION_ELIGIBLE
↓
PRODUCTION_ACCEPTED
↓
PROGRAM_COMPLETE
```
### REVIEW_READY
Requires:
- live authority rebind + exact preimages;
- ChangeEnvelope complete;
- complete affected repository regression;
- independent review GO.
### STAGING_ACTIVATION_READY
Requires:
- governed commit + authority readback;
- bounded staging activation scope;
- rollback/recovery plan ready;
- pre-activation lab/contract evidence.
It does **not** require evidence that only real staging activation can produce.
### STAGING_ACCEPTED
Requires actual staging:
- TaskPlane adapter/claim/lease/fencing;
- root action executor;
- ANNEX A01–A23;
- restart/response-loss/sibling concurrency/recovery evidence.
### PRODUCTION_ELIGIBLE
Requires:
- STAGING_ACCEPTED;
- existing release authorization valid;
- production rollback ready;
- freeze/NO-GO clear.
### PRODUCTION_ACCEPTED
Requires actual deployment:
- production deployment/runtime readback;
- authenticated business E2E;
- external projection readback.
### PROGRAM_COMPLETE
Only after production acceptance:
- existing Identity/Lineage original C01–C06 all accepted;
- Program final acceptance all satisfied;
- final root review proves no remaining acceptance gaps.
The existing root acceptance contract is unchanged. Its full completion is
consumed only by the final program-complete gate.
## 5. State authority rules
No new business authority is added.
- caller_scan: read-only evidence producer;
- WP0 inventories: read-only discovery;
- Environment Profile: read-only host profile;
- Retry Governor: decision/audit only;
- SchedulabilityProof: short-lived diagnostic evidence;
- SystemTruth: read-only derived view;
- TaskPlane / existing claim transaction: authoritative task/claim/lease state;
- existing mission/continuation path: root-action executor;
- existing release Saga: release authority.
No new Task DB, scheduler, mission store, release controller or authorization
authority is introduced in R2.
## 6. Migration
- Existing mission/root/children: preserve IDs and acceptance.
- Legacy durable task/job: attach metadata only; do not replace identity.
- Schedulability proof: persist only through existing evidence/journal if
persistence is required.
- No direct SQLite mutation outside existing TaskPlane repository.
- G0 outputs remain shadow/read-only.
## 7. Deployment status
R2 code may proceed to live-authority rebase and repository-native review after
its focused regressions pass.
Deployment remains NO-GO until:
- live preimages are refreshed;
- full repository regressions pass;
- atomic TaskPlane adapter is implemented and verified;
- ANNEX A18–A23 pass;
- real root-action executor consumes PENDING intents exactly once;
- independent staging review is GO.
## 8. Rollback
Review candidate: discard candidate.
Governed but undeployed commit: use existing successor/revert path; no shared
branch reset/force push.
Staging: disable feature/policy binding and preserve evidence.
Production: existing release Saga rollback only.
## 9. Deferred platform work
WP3–WP10 remain dependency-gated backlog from Platform Baseline V4. R2 does not
claim their acceptance and does not pre-build them.
+21
View File
@@ -0,0 +1,21 @@
# G0 Evidence Coverage V2
R2 corrects the R1 wording: WP0a–d and WP1 outputs are **discovery evidence**,
not complete Generation-G0 acceptance by themselves.
| WP | R2 output | What it proves | Still required for G0 Exit |
|---|---|---|---|
| WP0a | PARTIAL_DISCOVERY | authority/runtime/production OID relation | VERSION/current/baseline/README/versions/release metadata reader-writer-caller/live mapping |
| WP0b | PARTIAL_DISCOVERY | lexical Git-writer candidates | writer→auth→repo→ref→CAS→receipt→caller graph and live classification |
| WP0c | PARTIAL_DISCOVERY | authorization references by term/subdomain hint | owner/broker, TaskPrincipal, ClaimLease/TaskLease, WriterLease, release/recovery, signer chain + revocation/live caller |
| WP0d | PARTIAL_DISCOVERY | function-level canonical/digest candidates | producer/consumer, input schema, bytes rule, algorithm and duplicate semantic groups |
| WP0e | caller_scan v2 | fail-closed source/runtime caller evidence | five historical known-case acceptance + complete runtime evidence for any PROVEN_ORPHAN |
| WP1 | capability discovery / requirements evaluated | actual local capability probes | declared host requirements for each key execution host, missing requirements resolved |
| WP2 | conservative decision layer | retry/suppress/reconcile/block decision | audit baseline on real operations; no scheduler authority |
`g0_exit_eligible=false` in WP0a–d output is intentional. The G0 gate is a
separate acceptance aggregation step and must reference the missing evidence
listed above.
WP3–WP10 remain dependency-gated backlog exactly as Platform Baseline V4
defines; R2 does not silently redefine the long-term platform goal to U01–U09.
+65
View File
@@ -0,0 +1,65 @@
{
"schema": "aios.unified-review-gates.v2",
"gates": [
{
"id": "REVIEW_READY",
"requires": [
"BASE_REBOUND",
"CHANGE_ENVELOPE_COMPLETE",
"REPO_REGRESSION_PASS",
"INDEPENDENT_REVIEW_GO"
]
},
{
"id": "STAGING_ACTIVATION_READY",
"requires": [
"REVIEW_READY",
"GOVERNED_COMMIT_READBACK",
"LIMITED_SCOPE_PLAN",
"ROLLBACK_READY",
"LAB_CONTRACT_PASS"
]
},
{
"id": "STAGING_ACCEPTED",
"requires": [
"STAGING_ACTIVATION_READY",
"LIVE_ADAPTER_PASS",
"ROOT_EXECUTOR_PASS",
"ANNEX_A01_A23_PASS",
"FAILURE_RECOVERY_PASS"
]
},
{
"id": "PRODUCTION_ELIGIBLE",
"requires": [
"STAGING_ACCEPTED",
"RELEASE_AUTHORIZATION_VALID",
"PRODUCTION_ROLLBACK_READY",
"NO_GO_CLEAR"
]
},
{
"id": "PRODUCTION_ACCEPTED",
"requires": [
"PRODUCTION_ELIGIBLE",
"PRODUCTION_DEPLOYMENT_READBACK",
"BUSINESS_E2E_PASS",
"EXTERNAL_PROJECTION_READBACK"
]
},
{
"id": "PROGRAM_COMPLETE",
"requires": [
"PRODUCTION_ACCEPTED",
"EXISTING_ROOT_C01_C06_ACCEPTED",
"PROGRAM_U01_U09_ACCEPTED",
"FINAL_ROOT_REVIEW_COMPLETE"
]
}
],
"notes": [
"No gate may require evidence produced only by its own downstream action.",
"Existing Identity/Lineage root acceptance is unchanged; full C01-C06 is consumed only at PROGRAM_COMPLETE."
]
}
+56
View File
@@ -0,0 +1,56 @@
# AIOS Unified Platform Convergence — Task Contract V2
## Program goal
Preserve existing governance and tenant isolation while converging
Identity/Lineage, ANNEX-2.3 execution integrity, Platform Generation G0 and the
existing release/recovery path into one reviewable and recoverable program.
## Existing dependency
Identity/Lineage Mission:
- `mission-f3af205b-7dd6-455f-a561-88f0294cc815`
- root `task-133ad96e-e0d7-4202-934f-80c75053b11f`
Its C01–C06 acceptance is immutable.
## Program acceptance
| Key | Acceptance |
|---|---|
| U01 | authority/runtime/production baseline reconciled and candidate bound to current Git authority |
| U02 | existing Identity/Lineage mission identity and acceptance contract preserved; pre-production dependencies C01–C05 reconciled without claiming root complete |
| U03 | ANNEX-2.3 A01–A23 accepted in real staging/lab as specified |
| U04 | Platform Generation G0 Exit independently accepted; partial discovery output alone is insufficient |
| U05 | unified governed commit + authority readback |
| U06 | staging integrated E2E root→child→proof→ROOT_REVIEW→actual action→re-review |
| U07 | production deployment + runtime/business/external readback accepted |
| U08 | restart/response-loss/sibling concurrency/rollback/recovery accepted |
| U09 | existing Identity/Lineage original C01–C06 fully accepted and final Program ROOT_REVIEW proves no remaining gaps |
## Gate consumption
- REVIEW_READY consumes no staging/production evidence.
- STAGING_ACTIVATION_READY consumes governed commit/readback and lab evidence,
but no evidence that requires staging activation.
- STAGING_ACCEPTED consumes live staging evidence.
- PRODUCTION_ELIGIBLE consumes staging acceptance + release authorization.
- PRODUCTION_ACCEPTED consumes live production evidence.
- PROGRAM_COMPLETE consumes U01–U09 including final existing-root acceptance.
## Autonomous rule
Any child terminal:
`child evidence → ROOT_REVIEW → reload original acceptance → query descendants
→ next legal action`.
A child/local/gate PASS closes only that scope. It never closes the Program by
itself.
## Naming
New task display names retain:
`TASK-<SAFE_USER_ALIAS>-<UTC>-R<ROOT8>-P<PARENT8>-D<DEPTH>-C<SEQ>-<PURPOSE>`.
Authoritative lineage is persisted separately; existing durable IDs are not
renamed/recreated.
+66
View File
@@ -0,0 +1,66 @@
# Unified Workstream DAG & Acceptance Matrix V2
## Workstream DAG
```text
Baseline Reconciliation
├─ Track A Identity/Lineage existing mission
│ ├─ C01 Governed commit/readback
│ ├─ C02 Staging activation
│ ├─ C03 Root-action executor
│ ├─ C04 Legacy attach
│ ├─ C05 Live failure/replay/concurrency E2E
│ └─ C06 Production convergence
├─ Track B ANNEX-2.3 R2
│ ├─ B1 Schedulability proof
│ ├─ B2 Discovery diagnostics
│ ├─ B3 Claim-time read recheck
│ ├─ B4 Atomic TaskPlane claim/lease/fencing
│ ├─ B5 Execution/heartbeat
│ └─ B6 Evidence/readback A01–A23
└─ Track C Platform G0
├─ WP0a Version/Baseline discovery
├─ WP0b Git writer discovery
├─ WP0c Authorization discovery
├─ WP0d Canonical/Digest discovery
├─ WP0e caller_scan
├─ WP1 Environment Profile
└─ WP2 Retry/Toil Governor
A + B + C REVIEW_READY
↓
governed commit/readback
↓
STAGING_ACTIVATION_READY
↓
actual staging activation + A01–A23 + executor
↓
STAGING_ACCEPTED
↓
PRODUCTION_ELIGIBLE
↓
actual production deployment/readback
↓
PRODUCTION_ACCEPTED
↓
existing root C01–C06 + Program final review
↓
PROGRAM_COMPLETE
```
No gate depends on evidence generated only by its own downstream action.
## R2 matrix
| Gate/Area | R2 code/review status | Deployment evidence still required |
|---|---|---|
| caller_scan v2 | focused tests required | historical known-case + live runtime proof |
| schedulability proof | focused tests required | live adapter linearization/lease/fence |
| atomic claim contract | reference contract/tests only | TaskPlane adapter + race/revoke/freeze staging |
| Retry Governor v2 | focused tests required | real audit baseline |
| WP0a–d | PARTIAL_DISCOVERY by design | complete G0 mapping and independent acceptance |
| WP1 | capability/requirement probe | key-host requirement profiles |
| canonical reuse | ADR + guard test | producer/consumer compatibility in live rebase |
| governed commit | NOT RUN | writer receipt + authority readback |
| staging | NOT RUN | A01–A23, root executor, failure recovery |
| production | NOT RUN | release + business/runtime/external readback |
@@ -1,64 +0,0 @@
# ADR: AIOS Control-Plane System Truth and Execution Integrity
## Status
Proposed implementation baseline for ANNEX-2.3 hardening.
## Problem
Historical failures show that multiple planes can each report locally valid state while the end-to-end goal is false: Admission can be ADMITTED while TaskPlane has no claim; a task can be QUEUED until its grant expires; a worker can have matching capability names but an exact scope for another job; OpenAPI can advertise a route absent from the deployed backend; and PASS labels can be emitted without production evidence.
## Decision
Do not create a third persistence plane. Keep one authority per state dimension and derive `SystemTruth` read-only.
| State dimension | Authority |
|---|---|
| envelope / grant / admission receipt | Task Admission |
| task lifecycle / claim / attempt / heartbeat / deadline | TaskPlane |
| writer lease | governed writer / lease authority |
| source bytes / branch and commit OIDs | Git authority |
| deployed source / served routes | runtime |
| external ChatGPT Action operations | external projection readback |
### 1. Admission is not execution
`ADMITTED` means authorization has been accepted. It must never be presented as running, claimed, committed, or complete.
### 2. Schedulability before QUEUED
Before TaskPlane projection to `QUEUED`, produce a fresh `SchedulabilityProof` containing:
- required capabilities
- exact task/job/branch/execution-digest scope
- at least one healthy eligible worker
- current grant/principal viability
- dependencies/freeze state
If no eligible worker exists, keep execution projection in `WAITING_CAPACITY` (Admission may remain ADMITTED). Do not burn a short one-time grant while the task is waiting for capacity.
### 3. Mint runtime authority just-in-time
Separate long-lived Owner authorization from short-lived runtime TaskPrincipal. Mint the execution principal at claim time or immediately before claim, not at admission time. This prevents 15-minute grants expiring in a queue with no eligible worker.
### 4. Exact-scope is mandatory
Capabilities alone never establish eligibility. A worker is eligible only when `task_id`, `external_job_id`, `target_branch`, `generation`, and `execution_digest` are within its signed scope.
### 5. No fake execution
Transitions require evidence:
- CLAIMED: claim holder + fencing generation + lease readback
- RUNNING: CLAIMED + fresh heartbeat + valid lease
- SUCCEEDED: commit receipt + pre/post OID change + planned file hashes + runtime readback
- SCENARIO_ACCEPTED: SUCCEEDED + external projection readback
`service active`, `drift_status=PASS`, focused tests, or an Admission receipt are not substitutes.
### 6. Real bytes, not placeholders
Execution plans must reference content-addressed bytes that exist before admission. Hash-only plans with missing bytes, zero-length files, TODO/pass placeholders, or mismatched hashes are rejected.
### 7. One source for route contracts
Backend route registry and external OpenAPI must derive from one canonical route manifest, or be parity-checked in CI. Schema-only deployment is blocked if the matching backend route is absent; backend-only deployment is blocked if the external projection is absent when required.
### 8. Dual-plane reconciliation
Projection is idempotent and read-back verified. Admission writes an append-only projection intent/outbox in the same transaction as the grant; a projector writes TaskPlane, reads back exact task identity/revision, and records a receipt. Divergence is a first-class failure, never silently masked.
## Required release gates
- no BLOCKER from control-plane consistency audit
- C4 regression: legacy task creation path remains TaskPrincipal-gated
- admission -> CLAIMED -> RUNNING -> WriterLease -> governed commit -> readback integration test
- real activation + real SSH signature host test
- external projection readback test
- no placeholder artifact findings
@@ -0,0 +1,215 @@
# AIOS Unified Platform Convergence R1 — 架构变更说明
## 0. 决策摘要
本变更将 **Identity/Lineage S3、ANNEX-2.3 R2 Auto-Claim、平台化 Phase-0、
Governed Merge/Release** 收敛为一个新的上层 Program,而不是改写已存在
Identity/Lineage Mission 的原始验收合同。
核心原则:
1. **新 Program,不重写旧 root。** 已有 Mission `mission-f3af205b-7dd6-455f-a561-88f0294cc815` 和 root
`task-133ad96e-e0d7-4202-934f-80c75053b11f` 原样保留,C01–C06 的 acceptance 不变。
2. **唯一 Git authority 是 `beabd053961047f170e87e1110f2636dbd50f5fb`。** release-control runtime
`3c320aeb89d9ba8f0ed516abeb57252682b18983` 和 production `4642b74af62c89a505ae49241e805a5a88ca897f` 是不同平面,不得互相冒充。
3. **恢复语义,不恢复旧拓扑。** 从 `3c320ae…` 选择性移植 SystemTruth/Execution
Integrity 语义,但不把该后继分支整体推回 main。
4. **单一权威、只读聚合。** 新 inventory/caller_scan/Environment Profile/
Retry Governor 都是 evidence producer,不成为新持久化权威。
5. **先评审,后部署。** 本 R1 是 Review Candidate;禁止在 Review Gate 通过前
进行 staging/production activation。
## 1. 当前权威三平面
| 平面 | OID | 含义 |
|---|---|---|
| Git authority / main | `beabd053961047f170e87e1110f2636dbd50f5fb` | 新候选唯一 source base |
| release-control runtime | `3c320aeb89d9ba8f0ed516abeb57252682b18983` | 现有控制面运行字节,不是 main |
| production baseline | `4642b74af62c89a505ae49241e805a5a88ca897f` | V2R2C2 生产 evidence closure |
`beabd053961047f170e87e1110f2636dbd50f5fb -> 3c320aeb89d9ba8f0ed516abeb57252682b18983` 仅包含 10 个 SystemTruth/consistency 相关路径;
现有 Identity/Lineage 27 文件候选与这 10 个路径无交集。因此 27 文件候选可以
在 **重新读取 authority preimage 并通过 apply/check 后** 重新绑定
`beabd053961047f170e87e1110f2636dbd50f5fb`,但不能复用旧 `3c320ae…` approval/base 记录。
## 2. 新 Program 与既有 Mission 的关系
### 2.1 新上层 Program(尚未在 live TaskPlane 创建)
`goal_slug = AIOS-UNIFIED-PLATFORM-CONVERGENCE`
实际 `mission_id/root_task_id` 必须由 live TaskPlane authority 创建并返回,
本候选不预造 ID。
### 2.2 既有 Identity/Lineage Mission(保留)
- mission: `mission-f3af205b-7dd6-455f-a561-88f0294cc815`
- root: `task-133ad96e-e0d7-4202-934f-80c75053b11f`
- owner: `local-owner-authority`
- alias: `U9CA8400F7`
- legacy durable task: `TASK-P0-CHATGPT-BRIDGE-V2-CONTROL-PLANE-CONVERGENCE-INTEGRATE-20260913T184412Z-19b7`
- legacy job: `job-7457a322d5f9d6931c7cd99ce3ec6c18`
该 Mission 作为新 Program 的 Track A 依赖/子目标;其历史 task/job/idempotency
身份不可为了新命名而重建。
## 3. 四条 Track
### Track A — Identity / Lineage S3
目标保持:
`root → child → terminal proof → ROOT_REVIEW → actual next action → execution → root re-review`
包含:
- C01 governed candidate commit/readback
- C02 staging lineage policy activation
- C03 root-action executor binding
- C04 legacy attach
- C05 live failure/replay/concurrency E2E
- C06 production convergence
### Track B — ANNEX-2.3 R2 Auto-Claim / Execution Integrity
新链:
`admission/preflight → SchedulabilityProof → discovery → claim-time authoritative recheck
→ lease/fencing/CAS → execution → heartbeat/progress → result/evidence → readback/acceptance`
R1 已实现 review-candidate 代码:
- `task_plane/schedulability.py`
- `task_plane/auto_claim.py`
- A01 风格多 blocker、scope/capability/heartbeat/auth/capacity/freeze/dependency/lease
检查
- proof 仅为诊断证据,不是 bearer credential
- claim 前完整 authority recheck
**R1 不直接替换现有 claim 路径**;必须在 staging 评审时通过 adapter 绑定后才可启用。
### Track C — Platform Phase-0 / Generation G0
仅包含批准的第一 generation:
- WP0a Version/Baseline inventory
- WP0b Git writer inventory
- WP0c Authorization inventory
- WP0d Canonical/Digest inventory
- WP0e caller_scan
- WP1 Environment Capability Profile
- WP2 Retry/Toil Governor
这些组件只读或纯决策:
- 不新增 version authority
- 不新增 canonical authority
- 不写业务状态
- `UNKNOWN` 永远不能被 caller_scan 降级为 orphan
- Retry Governor 只返回决策,不执行 retry
禁止提前引入 WP8 MissionRun store / WP9 新 Release Saga 控制面。
### Track D — Governed Merge / Staging / Release
顺序:
`Review Gate → governed child commit/readback → staging activation → integrated live E2E
→ release candidate/acceptance → production convergence`
部署不是本 Review Candidate 的交付内容。
## 4. 状态权威边界
| 维度 | 权威 |
|---|---|
| Mission/root/lineage | TaskPlane/Mission authority |
| claim/heartbeat/task revision | TaskPlane |
| Owner/TaskPrincipal authorization | existing auth/Admission authority |
| Writer lease / governed write | existing writer/lease authority |
| source bytes/ref | Git authority |
| deployed runtime | runtime readback |
| external projection | external projection readback |
| caller liveness evidence | caller_scan evidence only |
| retry decision | Retry/Toil Governor decision only |
新模块不得建立第三份真值数据库。
## 5. 关键架构优化
### 5.1 动态 readiness 与不可变身份分离
Planner 选择身份(mission/root/child/route/authority digest)不可变;
`readiness_class` 是动态观察,可在未激活、未终态时通过 revision/CAS 单调向
“更可执行”方向推进。active claim/READY/DONE 后冻结。
### 5.2 Legacy reader 无副作用
旧 `GET /v1/tasks/{id}` 可能触发全局 lease expiry。R1 改为:
1. 通过 `/v1/tasks/{id}/events` 做 Bearer 认证;
2. Task row 从 authority SQLite 以 `mode=ro` + `PRAGMA query_only=ON` 精确读取;
3. 只允许终态;
4. readback ref 同时绑定 events SHA 和 DB revision;
5. 不暴露 `lease_token_hash`。
### 5.3 SchedulabilityProof 是短寿命诊断证据
Proof 绑定 task/worker/actor、revision/generation、capabilities、scope digest、
authorization revision、capacity/dependency/freeze/lease 观察;TTL ≤ 300s。
真正 claim 必须重读 authority 并完全一致,proof 本身不能授权。
### 5.4 Root Action Provider / Executor 分离
Provider 只把 root review 映射到准备计划;Executor 使用现有 continuation/recovery
路径执行。不能让 planner 把“PENDING”当成“已执行”。
## 6. 数据迁移策略
- 既有 Mission/root/children:不重建。
- legacy `TASK-P0-CHATGPT-BRIDGE-V2-CONTROL-PLANE-CONVERGENCE-INTEGRATE-20260913T184412Z-19b7` / `job-7457a322d5f9d6931c7cd99ce3ec6c18`:仅 attachment,不改 identity。
- 新 Platform inventory:无持久业务迁移。
- 新 schedulability proof:短寿命诊断记录;若持久化,必须复用现有 TaskPlane
evidence/journal,禁止新数据库。
- SystemTruth:只读派生,不能成为第三控制面。
## 7. Review Gate
### R0 — 架构与代码评审
必须同时满足:
- 新 Program 与旧 Mission 边界明确
- base 重新绑定 `beabd053961047f170e87e1110f2636dbd50f5fb`
- 变更路径均有 ChangeEnvelope
- 新代码定向测试 PASS
- 旧候选继承证据有效
- 无新增业务权威/数据库
- legacy reader 不再使用可变 GET
- planner readiness 动态规则经过测试
- R2 proof 不能直接授权 claim
### R1 — Governed Commit Gate
在 R0 GO 后:
- live reread authority HEAD
- 所有 preimage SHA 重新采集
- exact apply/diff
- governed writer commit
- authority branch/commit readback
### R2 — Staging Gate
- authenticated principal/tenant
- identity/lineage/read isolation
- auto-claim adapter
- root action real executor
- restart/response-loss/sibling race
- ANNEX A01–A23
### R3 — Production Gate
只有 original root acceptance + Program acceptance 全部证据闭环,
并完成 rollback/recovery 验证后才能生产收敛。
## 8. 回滚
- Review candidate:直接丢弃候选,无线上副作用。
- Governed commit 未发布:通过正式 successor/revert,不 reset/force push。
- Staging activation:关闭 feature/policy binding,保留 evidence。
- Production:只走既有 release Saga/rollback contract。
## 9. 明确不做
- 不把 `3c320aeb89d9ba8f0ed516abeb57252682b18983` 恢复为 main。
- 不新建任务数据库、第二 scheduler、第二 release authority。
- 不把 caller_scan/Retry Governor/SystemTruth 作为写权威。
- 不把局部测试 PASS 当 root/program 完成。
@@ -0,0 +1,56 @@
# Unified Workstream DAG & Acceptance Matrix V1
```text
U01 Baseline Reconciliation
├─ Track A Identity/Lineage existing mission
│ ├─ C01 Governed commit/readback
│ ├─ C02 Staging activation
│ ├─ C03 Root-action executor
│ ├─ C04 Legacy attach
│ ├─ C05 Live failure/replay/concurrency E2E
│ └─ C06 Production convergence
├─ Track B ANNEX-2.3 R2
│ ├─ B1 Schedulability proof
│ ├─ B2 Discovery
│ ├─ B3 Claim-time recheck
│ ├─ B4 Claim/lease/fencing
│ ├─ B5 Execution/heartbeat
│ └─ B6 Evidence/readback A01-A23
└─ Track C Platform G0
├─ WP0a Version/Baseline
├─ WP0b Git writers
├─ WP0c Authorization
├─ WP0d Canonical/Digest
├─ WP0e caller_scan
├─ WP1 Environment Profile
└─ WP2 Retry/Toil Governor
A + B + C review GO
↓
U05 Governed unified commit/readback
↓
U06 Staging integrated E2E
↓
U07/U08 Production + Recovery
```
## Gate matrix
| Gate | Required evidence | Status in R1 |
|---|---|---|
| Baseline authority | main/runtime/prod current readback | PASS |
| Identity/Lineage candidate | frozen 27-file evidence + merge review | INHERITED/PENDING REBASE |
| Legacy reader | non-mutating authenticated read + terminal-only | LOCAL PASS |
| Planner dynamic readiness | CAS monotonic readiness, active/terminal freeze | CODED; repo-native rerun pending |
| R2 schedulability | blocker matrix + proof replay + claim recheck | LOCAL PASS |
| WP0a–d | read-only inventories | CODED; live scan pending |
| WP0e caller_scan | UNKNOWN safety/orphan rules | LOCAL PASS |
| WP1 env profile | host capability evidence | LOCAL PASS |
| WP2 retry governor | reconcile/no-op/block/execute | LOCAL PASS |
| Governed commit | writer receipt + branch readback | NOT RUN |
| Staging E2E | actual executor/readback | NOT RUN |
| Production | release acceptance/readback | NOT RUN |
## Deployment ordering
No deployment starts until all Review Gate rows required for R0 are PASS/GO.
@@ -1,33 +0,0 @@
# Remediation Work Packages
## P0-A — Single System Truth (no new DB)
Implement the read-only SystemTruth derivation in the runtime explain/status surface. Remove or rename PASS labels that are configuration-only.
## P0-B — Schedulability Gate
Admission may authorize a task, but TaskPlane projection to QUEUED requires a fresh SchedulabilityProof. If no eligible worker exists, expose WAITING_CAPACITY and do not start the short runtime grant TTL.
## P0-C — Just-in-time TaskPrincipal
Keep Owner authorization durable; mint short-lived TaskPrincipal only when an eligible worker is ready to claim. Bind task/job/branch/generation/execution_digest exactly.
## P0-D — Evidence-required state transitions
Guard CLAIMED/RUNNING/SUCCEEDED with claim holder, heartbeat, lease, commit receipt, OID change, file hashes, runtime readback, external projection readback.
## P0-E — Artifact reality gate
Before admission, verify every planned file byte exists, is non-placeholder, and matches SHA256. Reject hash-only or placeholder plans.
## P0-F — Route/projection single source
Generate or verify ChatGPT Actions OpenAPI from the backend route manifest. Block dead endpoints and unprojected required operations.
## P0-G — Dual-plane durable projection
Use append-only projection intent + idempotent TaskPlane write + exact readback receipt. Admission status and execution status remain separate.
## P0-H — Worker scope/schedulability diagnostics
Eligibility = capability AND heartbeat AND exact signed scope. Never select a worker only because capability names match.
## P0-I — QA gaps
Add:
1. C4 legacy createGovernedTask/projectOnlineAgentExecution remains TaskPrincipal-gated.
2. `_verified_task_principal` body injection negative test.
3. bearer-only no-delegation negative test.
4. real CLAIMED/RUNNING/WriterLease/commit readback path.
5. real activation/SSH signature host test.
+5
View File
@@ -12,6 +12,11 @@ components:
type: http
scheme: bearer
bearerFormat: opaque-token
taskPrincipalAuth:
type: http
scheme: bearer
bearerFormat: task-principal-jwt
description: Server-side TaskPrincipal resolved by the Bridge from the authoritative admitted+claimed Task; caller-supplied tokens are never accepted.
schemas:
PublicationState:
type: string
@@ -0,0 +1,31 @@
# Review Gate V1
## Review disposition
This package is a **review candidate**, not a deployable release.
### Code review focus
- authority base = `beabd053961047f170e87e1110f2636dbd50f5fb`
- no direct DB mutation outside existing repository/service contracts
- no new control-plane database
- proof != authorization
- caller_scan cannot promote UNKNOWN to orphan
- retry governor cannot perform business effects
- legacy read path is non-mutating
- planner readiness is observation, not immutable identity
- existing mission/task/job IDs preserved
### Required test classes before GO
1. candidate syntax
2. new module focused unit tests
3. inherited Identity/Lineage regression suite
4. legacy core/runtime tests
5. ANNEX A01–A23 staging tests
6. root-action executor restart/response-loss tests
7. tenant isolation
8. governed Git apply/readback
### Current R1 decision rule
`ARCH_REVIEW_GO` may be issued if architecture/code review finds no blocker.
`DEPLOY_GO` is explicitly unavailable until governed commit + staging E2E pass.
+42 -57
View File
@@ -1,57 +1,42 @@
# AIOS Task Plane
该模块是与任务发起方、Worker 类型和通知后端解耦的持久化任务控制平面。当前交付仅用于隔离开发与 Shadow 验收;默认 `execution_enabled=false`,不会修改生产、tag、Baseline 或 Registry。
## 架构
- `TaskPlaneService`:统一服务边界和 HTTP 路由。
- `TaskPlaneStore`:SQLite WAL、revision/CAS、串行事务、租约、hash-chained journal、evidence 和 durable outbox。
- `WorkerRegistry` / `LeaseManager`:身份绑定、capability、并发额度、心跳、过期探测后恢复。
- `DelegationEngine`:父子 DAG、全局深度、capability/risk/approval/budget 单调约束和幂等重放。
- `PolicyEngine`:默认关闭执行;身份、证据、结果摘要和可插拔签名验签均 fail-closed。
- `WorkerAdapter`:OpenClaw、Codex、Hermes、A2A 和 systemd Worker 的统一契约。默认实现只返回 `SHADOW_VERIFIED`,不执行外部命令。
- `DurableOutboxDispatcher`:Webhook、NATS JetStream、GPT callback、Dashboard 和 WeChat 的回执式通知接口。没有真实回执时始终保留 `NOT_DELIVERED`。
## 状态与恢复
正常路径为 `CREATED → QUEUED/CLAIMED → RUNNING → VERIFYING/COMPLETED`;失败路径使用 `FAILED_RECOVERABLE`、`FAILED_FINAL`、`CANCELLED` 或 `MANUAL_REVIEW_REQUIRED`。每次状态事件都进入同一事务中的 append-only hash chain。
租约过期不会自动重放副作用。控制平面必须先调用 Worker probe,只有明确确认原执行已停止,才可将 claim 标记为 `EXPIRED` 并把任务转为 `FAILED_RECOVERABLE`。结果必须匹配 task、claim、Worker identity、未过期 lease 和 operation digest;重复结果返回冲突。
## 安全模型
- task/worker/claim/message ID 使用 allowlist pattern,拒绝 URL 编码差异、路径分隔符和控制字符。
- HTTP body 最大 1 MB,非对象 JSON、异常 Content-Length 和非法参数返回 400/413。
- 内部异常响应不回显路径、payload 或凭据;默认 HTTP access log 不记录原始 request target。
- secret-like 字段在 durable task、event、evidence 和 outbox 写入前脱敏。
- 结果签名存在但没有配置验签器时拒绝;摘要、evidence 或 event 被篡改时验收失败。
- SQLite 文件创建为 0600;新建状态目录使用 0700。
## 租户语义(cross-tenant)
任务与 Worker 各带租户标识:任务取自 `task_plan.metadata.tenant_id`(默认 `default`),
Worker 取自 `trust_zone`(默认 `default`)。跨租户(cross-tenant)处理语义为:
- **枚举 = skip + 审计**:`poll` / `long_poll` / `schedule`(`queued_count`)在遍历候选任务时,
对与 Worker 租户不一致的任务执行 SKIP,写入结构化审计 finding
`{source:"tenant", kind:"cross_tenant_skip", decision:"skip+audit", task_tenant, worker_tenant}`
(持久化于 `tenant_skip_audit` 表并在任务事件链追加 `TASK_TENANT_SKIPPED`),**不抛异常、不 fail-closed**。
仅当存在同租户候选时才会被选中。
- **直接读/认领 = fail-closed**:按 `task_id` 直接 `governed_claim` / `claim_governed` 跨租户任务时
保持拒绝,返回 `403 cross_tenant_denied`,不泄露任务内容。
即:Worker 只能“看到”并认领同租户任务;枚举路径让跨租户任务不可见且留有审计,直接认领路径则明确失败。
## 验收
```bash
PYTHONPATH=active-runtime \
bash scripts/task-plane/run_task_plane_acceptance.sh
PYTHONPATH=active-runtime \
python3 scripts/task-plane/create_shadow_tasks.py \
--database reports/task-plane-shadow.sqlite3 \
--report reports/task-plane-shadow.json
```
上述脚本只使用本地 SQLite 和 Shadow adapter,不触发真实 OpenClaw、Codex、Hermes、微信或生产执行。
# AIOS Task Plane
该模块是与任务发起方、Worker 类型和通知后端解耦的持久化任务控制平面。当前交付仅用于隔离开发与 Shadow 验收;默认 `execution_enabled=false`,不会修改生产、tag、Baseline 或 Registry。
## 架构
- `TaskPlaneService`:统一服务边界和 HTTP 路由。
- `TaskPlaneStore`:SQLite WAL、revision/CAS、串行事务、租约、hash-chained journal、evidence 和 durable outbox。
- `WorkerRegistry` / `LeaseManager`:身份绑定、capability、并发额度、心跳、过期探测后恢复。
- `DelegationEngine`:父子 DAG、全局深度、capability/risk/approval/budget 单调约束和幂等重放。
- `PolicyEngine`:默认关闭执行;身份、证据、结果摘要和可插拔签名验签均 fail-closed。
- `WorkerAdapter`:OpenClaw、Codex、Hermes、A2A 和 systemd Worker 的统一契约。默认实现只返回 `SHADOW_VERIFIED`,不执行外部命令。
- `DurableOutboxDispatcher`:Webhook、NATS JetStream、GPT callback、Dashboard 和 WeChat 的回执式通知接口。没有真实回执时始终保留 `NOT_DELIVERED`。
## 状态与恢复
正常路径为 `CREATED → QUEUED/CLAIMED → RUNNING → VERIFYING/COMPLETED`;失败路径使用 `FAILED_RECOVERABLE`、`FAILED_FINAL`、`CANCELLED` 或 `MANUAL_REVIEW_REQUIRED`。每次状态事件都进入同一事务中的 append-only hash chain。
租约过期不会自动重放副作用。控制平面必须先调用 Worker probe,只有明确确认原执行已停止,才可将 claim 标记为 `EXPIRED` 并把任务转为 `FAILED_RECOVERABLE`。结果必须匹配 task、claim、Worker identity、未过期 lease 和 operation digest;重复结果返回冲突。
## 安全模型
- task/worker/claim/message ID 使用 allowlist pattern,拒绝 URL 编码差异、路径分隔符和控制字符。
- HTTP body 最大 1 MB,非对象 JSON、异常 Content-Length 和非法参数返回 400/413。
- 内部异常响应不回显路径、payload 或凭据;默认 HTTP access log 不记录原始 request target。
- secret-like 字段在 durable task、event、evidence 和 outbox 写入前脱敏。
- 结果签名存在但没有配置验签器时拒绝;摘要、evidence 或 event 被篡改时验收失败。
- SQLite 文件创建为 0600;新建状态目录使用 0700。
## 验收
```bash
PYTHONPATH=active-runtime \
bash scripts/task-plane/run_task_plane_acceptance.sh
PYTHONPATH=active-runtime \
python3 scripts/task-plane/create_shadow_tasks.py \
--database reports/task-plane-shadow.sqlite3 \
--report reports/task-plane-shadow.json
```
上述脚本只使用本地 SQLite 和 Shadow adapter,不触发真实 OpenClaw、Codex、Hermes、微信或生产执行。
+131
View File
@@ -0,0 +1,131 @@
# TaskPlane task identity and lineage
TaskPlane can bind admitted tasks to an authenticated owner principal and mission, assign stable root/parent lineage, and persist a root review after a child reaches a proof-bearing terminal state. The integration is opt-in through `TaskPlaneService(..., lineage_policy=...)`; existing services without a lineage policy retain their existing behavior.
## Authority boundary
`PrincipalLineageBinding` is service configuration. It maps the identity returned by the HTTP authenticator to `owner_principal_id`, `mission_id`, a safe display alias, and a registered root-acceptance policy reference. HTTP admission overwrites `issuer` with that authenticated identity. Direct in-process callers must pass an identity already verified by their boundary through `actor_identity`.
Client task metadata is never an identity or terminal-proof source. `aios_task_lineage`, the derived display name, and the lineage request digest are server-owned and rejected on new lineage admission. Similar-looking client metadata such as `owner_principal_id`, `mission_id`, or `terminal_evidence_proven` remains ordinary opaque metadata and cannot override the dedicated TaskPlane authority columns or the validated terminal-proof tables.
Root and child creation use the existing TaskPlane SQLite connection and transaction fence. The store adds authority columns and lineage sequence, terminal proof, root review, and root action tables to that database. It does not create a second TaskPlane database. The original task ID, claim ID, idempotency key, budget transaction, event journal, and revision CAS remain authoritative.
## Admission and delegation
A root admission requires:
- an authenticated identity with both TaskPlane capability and a `PrincipalLineageBinding`;
- a registered acceptance evaluator for the configured reference;
- no client-supplied reserved lineage metadata.
The request digest is computed before derived metadata is added. An identical idempotency replay returns the original task; a changed request, principal, mission, or parent conflicts. The persisted TaskPlane `created_at` is the same timestamp serialized into lineage metadata.
Task detail and list responses expose the server-bound name as `display_name`, including list responses that omit the full plan. Durable task and job identifiers remain unchanged.
Delegation currently creates only `CHILD` relations. It preserves the root owner, mission, acceptance reference, budget constraints, capability/risk monotonicity, and existing delegation transaction. The caller must resolve to the same server-owned root authority. Concurrent sibling admission uses the same SQLite write transaction to reserve unique sibling sequences.
`SUCCESSOR` and `RECOVERY` are intentionally not admitted by this binding because the current TaskPlane service does not expose a formal continuation/supersession authority contract. Adding those relations requires a separate source-of-authority design; callers must not encode them in metadata.
## Proven terminal review
The proof-bearing result path first applies the existing claim identity, lease, operation digest, evidence, result digest, event-chain, and evidence-chain checks. In the same transaction, a final child result records a terminal proof and triggers one `ROOT_REVIEW`. The review CAS advances the root revision and persists the exact trigger child and terminal revision.
The resulting next legal action is stored idempotently with status `PENDING`. `PENDING` means a durable action request exists. It does not mean a planner completed the root, created another child, recovered work, or crossed an owner boundary.
The optional `TaskPlaneService.consume_one_root_action()` consumes one persisted action through a root claim and a durable effect ledger. Mission configuration must provide an authoritative plan builder and acceptance verifier, and the service must receive a real `root_action_executor`. Missing execution wiring returns `BLOCKED`; the loopback Mission bootstrap adapter alone does not execute root actions. Plans survive process restart, and an unknown effect must be reconciled against its existing authority Job before execution can proceed. Worker metadata cannot supply that Job identity or certify a completed effect.
`PublicBridgeRootActionExecutor` is an adapter for authenticated public Bridge operations, not a deployed worker. Runtime integration must supply its trusted authentication and transport, then prove the actual Job result and repository readback. After an execution receipt, the consumer rechecks the stored original root acceptance and descendants. Root completion and child admission share the store's transaction boundary; a child result alone cannot close the root.
Direct worker completion of a lineage root is rejected. Root completion is legal only when the registered evaluator returns satisfied conditions backed by durable evidence references and the root review selects `ROOT_COMPLETE`. This integration persists that legal decision and request; it does not itself change the root task to `COMPLETED`.
After restart, `resume_terminal_review(child_task_id=...)` reads the durable terminal proof. If the trigger already has a review, it validates and returns that exact review and action receipt without running the evaluator again. This freezes the decision against evaluator drift and prevents a lost response from producing a second review. Tests close and reopen the SQLite database before treating this state as durable; visibility through the writing connection inside its transaction is only transaction-local.
The legacy `/fail` and `/cancel` paths do not construct a validated `ResultEnvelope` and therefore do not create this terminal proof or claim root-review completion. Use the proof-bearing result endpoint when terminal lineage review is required.
## Minimal service configuration
```python
from app.aios_task_lineage import AcceptanceCondition, RootAcceptanceSnapshot
from app.task_plane.api import TaskPlaneService
from app.task_plane.lineage_store import PrincipalLineageBinding, TaskLineagePolicy
def evaluate_root(root, descendants):
refs = tuple(
ref
for child in descendants
if child.terminal_evidence_proven
for ref in child.evidence_refs
)
return RootAcceptanceSnapshot(
(
AcceptanceCondition(
key="original-root-acceptance-not-yet-evaluated",
# Replace with the original mission's authoritative acceptance
# checks. A completed child alone cannot prove root completion.
satisfied=False,
evidence_refs=refs,
),
)
)
lineage_policy = TaskLineagePolicy(
principal_bindings=(
PrincipalLineageBinding(
authenticated_identity="spiffe://aios/owner/example",
owner_principal_id="principal-example",
mission_id="mission-example",
preferred_owner_alias="OWNER",
root_acceptance_ref="policy://root/example-v1",
),
),
acceptance_evaluators={"policy://root/example-v1": evaluate_root},
)
service = TaskPlaneService(store, policy=policy, lineage_policy=lineage_policy)
```
An evaluator that is missing, returns no conditions, or marks a condition satisfied without evidence fails closed. It cannot be supplied by task metadata.
Existing tasks receive empty authority-column defaults during additive schema
initialization. This does not infer ownership from their metadata or attach old
jobs to a new root. Migrating an actual legacy task requires an authoritative
owner/mission binding; this change does not fabricate one or alter a Job DB.
## Canonical Mission issuance
When discovery establishes that no root exists, the optional `MissionAuthority`
can allocate the first root. `POST /missions` accepts only `goal_slug` and
`original_acceptance`, with an `Idempotency-Key`. The authenticated principal
must have a server-configured `MissionPrincipalBinding`. Mission/root IDs,
owner binding, safe alias, UTC creation time and acceptance reference are
allocated and persisted by the authority in the existing TaskPlane database,
in the same transaction as root creation. They are never request identity
fields. An exact retry returns the original identity; a changed contract with
the same principal/idempotency key conflicts. `GET /missions/{mission_id}`
requires the same owner binding and returns the persisted acceptance and root.
Acceptance JSON is data. A service-owned verifier must independently establish
its conditions; absence of a registered verifier keeps acceptance unmet.
Creating the root or attaching an arbitrary evidence string cannot satisfy it.
The Mission authority and ordinary worker execution are separate: enabling
Mission admission does not grant worker claim or result permissions.
`app.aios_task_mission_authority_server` is a narrow loopback staging adapter
for environments with the existing registered Owner signature verifier. It
accepts `POST /v1/signed-request` containing exactly `request`, `signature` and
`key_id`. The request schema is `aios.signed-mission-request.v1`, audience is
`aios.task-mission-authority.staging.v1`, and the signed fields bind method,
path, body, idempotency key, issued/expiry timestamps and nonce. The maximum
validity is 300 seconds. The existing `aios-owner-approval-v1` verifier checks
the signature, and its registered signer identity is the service actor.
Neither a bearer belonging to Shadow nor an unsigned owner name establishes
that identity. The adapter does not load private keys or sign requests.
Allowed signed operations are Mission create/read and
`POST /tasks/{authority-returned-parent}/delegate`. Mission delegation issues
child IDs on the server and uses a durable, principal-scoped request mapping
for retries. The adapter exposes no Git writer, worker claim, terminal result,
legacy recovery, production deployment or arbitrary execution route. Deployment
of this bootstrap adapter is staging evidence, not a governed commit receipt
or proof that `PENDING` root actions execute.
+1
View File
@@ -0,0 +1 @@
{"baseline_mutation":false,"bootstrap_recovery_mode":false,"git_ls_tree_sha256":"58689cd365bec5795c19c2e1e03fc8ae774f768ed03a42d501b4430c95f71a8b","production_mutation":false,"schema":"aios.governed-runtime-package.v1","source_commit":"2852c2a40394195192c6c02a215ba31764e2cdda","source_file_count":965,"tag_mutation":false,"tree_oid":"ccd67c2357e7a4cd74e06be61ab81bba820d7671"}
+22
View File
@@ -0,0 +1,22 @@
#!/usr/bin/env python3
import sqlite3, sys
dbs = [
("scoped-leases", "/var/lib/aios-gitops/bridge-v2-staging/scoped-leases.sqlite3"),
("task-admission", "/var/lib/aios-gitops/bridge-v2-staging/task-admission.sqlite3"),
]
for label, db in dbs:
print(f"=== {label}: {db} ===")
try:
c = sqlite3.connect(f"file:{db}?mode=ro", uri=True)
tabs = [r[0] for r in c.execute("select name from sqlite_master where type='table'")]
print("tables:", tabs)
for t in tabs:
try:
n = c.execute(f"select count(*) from \"{t}\"").fetchone()[0]
print(f" {t}: rows={n}")
except Exception as e:
print(f" {t}: count-err {e}")
c.close()
except Exception as e:
print("ERR", e)
+24
View File
@@ -0,0 +1,24 @@
#!/usr/bin/env python3
import sqlite3
# scoped leases detail
db = "/var/lib/aios-gitops/bridge-v2-staging/scoped-leases.sqlite3"
c = sqlite3.connect(f"file:{db}?mode=ro", uri=True)
cols = [r[1] for r in c.execute("PRAGMA table_info(scoped_leases)")]
print("scoped_leases cols:", cols)
rows = c.execute("select * from scoped_leases").fetchall()
print("scoped_leases rows:", len(rows))
for r in rows[:12]:
print(" ", dict(zip(cols, r)))
c.close()
# task-admission envelopes status
db2 = "/var/lib/aios-gitops/bridge-v2-staging/task-admission.sqlite3"
c2 = sqlite3.connect(f"file:{db2}?mode=ro", uri=True)
for t in ["task_admission_envelopes","task_admission_grants","task_admission_receipts"]:
cols2 = [r[1] for r in c2.execute(f"PRAGMA table_info({t})")]
print(f"\n=== {t} cols: {cols2}")
rows2 = c2.execute(f"select * from {t}").fetchall()
for r in rows2[:6]:
print(" ", dict(zip(cols2, r)))
c2.close()
@@ -24,12 +24,14 @@ if run_gitea admin user list 2>/dev/null | awk '{print $2}' | grep -Fxq "$OWNER_
else
run_gitea admin user create --username "$OWNER_USERNAME" --password "$OWNER_PASSWORD" --email "$OWNER_EMAIL" --admin --must-change-password=false >/dev/null
fi
python3 - <<PY
import sqlite3, time
con=sqlite3.connect('$DB')
cur=con.cursor()
cur.execute("update user set is_admin=1,is_active=1,prohibit_login=0,must_change_password=0,updated_unix=? where lower_name=lower(?)", (int(time.time()), '$OWNER_USERNAME'))
cur.execute("update user set prohibit_login=1,is_admin=0,passwd='',passwd_hash_algo='disabled',must_change_password=1,updated_unix=? where lower_name in ('aios','admin','gitea') and lower_name<>lower(?)", (int(time.time()), '$OWNER_USERNAME'))
python3 - "$DB" "$OWNER_USERNAME" <<'PY'
import sqlite3, sys, time
db = sys.argv[1]
username = sys.argv[2]
con = sqlite3.connect(db)
cur = con.cursor()
cur.execute("update user set is_admin=1,is_active=1,prohibit_login=0,must_change_password=0,updated_unix=? where lower_name=lower(?)", (int(time.time()), username))
cur.execute("update user set prohibit_login=1,is_admin=0,passwd='',passwd_hash_algo='disabled',must_change_password=1,updated_unix=? where lower_name in ('aios','admin','gitea') and lower_name<>lower(?)", (int(time.time()), username))
con.commit()
PY
printf 'GITEA_OWNER_SYNC_OK user=%s default_local_login_disabled=1\n' "$OWNER_USERNAME"
@@ -219,6 +219,81 @@
</section>
</main>
</div>
<!-- SEC-P1 WEB CONTROL SURFACE hardening: CSRF token propagation + server-authoritative input pre-validation.
Runs before app.js so every state-change fetch issued by the control plane carries the CSRF token,
and invalid branch/worktree identifiers are blocked client-side (server remains authoritative). -->
<script>
(function () {
'use strict';
// Double-submit cookie pattern: server sets a signed `aios_csrf` cookie; the client echoes its
// value in X-CSRF-Token on every state-changing (non-safe) request. A server-issued
// <meta name="csrf-token"> takes precedence when present.
function csrfToken() {
var meta = document.querySelector('meta[name="csrf-token"]');
if (meta && meta.content) return meta.content;
var m = document.cookie.match(/(?:^|;\\s*)aios_csrf=([^;]+)/);
return m ? decodeURIComponent(m[1]) : '';
}
// Server-authoritative identifier rules for git refs / worktrees. Kept in sync with the backend
// validator so the client gives fast feedback, but the REAL enforcement happens server-side.
var ID_CHARS = /^[A-Za-z0-9._/-]+$/;
var ID_FORBIDDEN = ['..', '//', '@{', '.lock', ' ', '\\t', ';', '&', '|', '$', '(', ')', '<', '>', '"', "'", '`', '\\\\'];
function validateBranch(name) {
if (typeof name !== 'string' || !name.trim()) return 'branch name is required';
if (name.trim().length > 255) return 'branch name too long';
if (name !== name.trim()) return 'branch name must not have leading/trailing whitespace';
for (var i = 0; i < ID_FORBIDDEN.length; i++) {
if (name.indexOf(ID_FORBIDDEN[i]) !== -1) return 'branch name contains forbidden characters: ' + ID_FORBIDDEN[i];
}
if (name.indexOf('..') !== -1) return 'branch name must not contain ".."';
if (name.indexOf('@{') !== -1) return 'branch name must not contain "@{"';
if (!ID_CHARS.test(name)) return 'branch name contains invalid characters';
if (name.indexOf('/') === 0 || name[name.length - 1] === '/') return 'branch name must not start/end with "/"';
return null;
}
function validateWorktree(name) {
if (typeof name !== 'string' || !name.trim()) return 'worktree id is required';
var e = validateBranch(name);
return e ? 'worktree id invalid: ' + e : null;
}
function stateHeaders() {
return { 'X-CSRF-Token': csrfToken() };
}
// Wrap fetch: inject CSRF header on any non-safe method (state change).
var nativeFetch = window.fetch;
window.fetch = function (url, opts) {
opts = opts || {};
var method = String(opts.method || 'GET').toUpperCase();
var safe = (method === 'GET' || method === 'HEAD' || method === 'OPTIONS');
if (!safe) {
opts.headers = Object.assign({}, opts.headers);
opts.headers['X-CSRF-Token'] = csrfToken();
}
return nativeFetch.call(this, url, opts);
};
// Client-side pre-validation for state-change forms (server re-validates authoritatively).
document.addEventListener('submit', function (e) {
var form = e.target;
if (!form || (form.id !== 'branch-form' && form.id !== 'plan-form')) return;
var fd = new FormData(form);
var branch = String(fd.get('branch') || '').trim();
var err = validateBranch(branch);
if (err) {
e.preventDefault();
e.stopPropagation();
window.alert('Blocked: ' + err);
}
}, true);
window.__gitopsSurface = { csrfToken: csrfToken, validateBranch: validateBranch, validateWorktree: validateWorktree, stateHeaders: stateHeaders };
})();
</script>
<script src="./app.js"></script>
</body>
</html>
+43
View File
@@ -0,0 +1,43 @@
#!/usr/bin/env python3
import sys, importlib.util
# Load the installed migration planner runtime module
app = "/opt/aios/release-migration/current/app"
sys.path.insert(0, app)
import release_migration_plan as rm
VERSION = "V2R2C2-baseline11-principal-recovery"
CANDIDATE = "rc-v2r2c2-baseline11-principal-recovery-2852c2a40394"
SOURCE = "2852c2a40394195192c6c02a215ba31764e2cdda"
resolver = rm.FileMigrationPlanResolver(
"/var/lib/aios-release-migration",
profile_root="/etc/aios-gitops/release-migration-profiles",
expected_owner_uid=0,
)
try:
plan = resolver.resolve(VERSION, CANDIDATE, SOURCE)
if plan is None:
print("RESOLVE_RESULT = NONE (not required?)")
sys.exit(3)
print("RESOLVE_RESULT = MIGRATION_PLAN_READY")
d = plan.to_dict()
print("migration_id =", d.get("migration_id"))
print("migration_plan_sha =", d.get("sha256"))
print("source_commit =", d.get("source_commit"))
print("tree_oid =", d.get("tree_oid"))
# confirm the migration sub-contract binds snapshot
for sc in d.get("state_contracts", []):
print("state_contract snapshot_name =", sc.get("snapshot_name"))
print("state_contract snapshot_sha =", sc.get("snapshot_sha256"))
print("DURABLE_PREREQUISITE_STATE = SATISFIED")
sys.exit(0)
except rm.MigrationPlanRequired as e:
print("RESOLVE_RESULT =", e)
print("DURABLE_PREREQUISITE_STATE = NOT_SATISFIED")
sys.exit(1)
except Exception as e:
print("RESOLVE_ERROR =", type(e).__name__, e)
sys.exit(2)
+73
View File
@@ -0,0 +1,73 @@
--- a/active-runtime/app/aios_trusted_candidate_builder.py
+++ b/active-runtime/app/aios_trusted_candidate_builder.py
@@ -367,15 +367,15 @@
raise BuilderError(code) from exc
if not stat.S_ISDIR(metadata.st_mode) or path.is_symlink():
raise BuilderError(code)
- if os.name != "nt" and metadata.st_mode & 0o022:
- raise BuilderError("COLLECTOR_EVIDENCE_ROOT_MODE_DENIED")
- if (
- path == self._evidence_root
- and os.name != "nt"
- and self._expected_owner_uid is not None
- and metadata.st_uid != self._expected_owner_uid
- ):
- raise BuilderError("COLLECTOR_EVIDENCE_ROOT_OWNER_DENIED")
+ if path == self._evidence_root:
+ if os.name != "nt" and metadata.st_mode & 0o022:
+ raise BuilderError("COLLECTOR_EVIDENCE_ROOT_MODE_DENIED")
+ if (
+ os.name != "nt"
+ and self._expected_owner_uid is not None
+ and metadata.st_uid != self._expected_owner_uid
+ ):
+ raise BuilderError("COLLECTOR_EVIDENCE_ROOT_OWNER_DENIED")
def _digest(self, path: Path) -> str:
digest = hashlib.sha256()
--- a/tests/test_trusted_candidate_builder.py
+++ b/tests/test_trusted_candidate_builder.py
@@ -323,9 +323,11 @@
policy = load_builder_policy(POLICY_PATH)
config = TrustedCandidateBuilderConfig.from_policy(policy)
evidence_root = tmp_path / "evidence"
- evidence_root.mkdir(mode=0o700)
+ evidence_root.mkdir()
+ evidence_root.chmod(0o700)
authority_git = tmp_path / "repo.git"
- authority_git.mkdir(mode=0o700)
+ authority_git.mkdir()
+ authority_git.chmod(0o2775)
source = ServerOwnedEvidenceSource(
config,
@@ -356,6 +358,29 @@
)
with pytest.raises(BuilderError, match="COLLECTOR_EVIDENCE_UNAVAILABLE"):
server.handle_request(build, connection=object())
+
+
+def test_readiness_keeps_evidence_root_permissions_strict_with_group_shared_authority(tmp_path) -> None:
+ config = TrustedCandidateBuilderConfig.from_policy(load_builder_policy(POLICY_PATH))
+ evidence_root = tmp_path / "evidence"
+ evidence_root.mkdir()
+ evidence_root.chmod(0o720)
+ authority_git = tmp_path / "repo.git"
+ authority_git.mkdir()
+ authority_git.chmod(0o2775)
+
+ source = ServerOwnedEvidenceSource(
+ config,
+ authority_git,
+ evidence_root,
+ expected_owner_uid=None,
+ git_runner=lambda *_args, **_kwargs: SimpleNamespace(
+ stdout="a" * 40 + "\n", returncode=0
+ ),
+ )
+
+ with pytest.raises(BuilderError, match="COLLECTOR_EVIDENCE_ROOT_MODE_DENIED"):
+ source.assert_ready()
def test_unix_client_applies_bounded_timeout_and_fails_closed(monkeypatch) -> None:
@@ -1,167 +0,0 @@
#!/usr/bin/env python3
from __future__ import annotations
from pathlib import Path
import argparse, json, sys, time, re
from aios_control_plane_consistency import *
def jload(value):
try:
return json.loads(value or "{}")
except Exception:
return {}
def rows(path, query, params=()):
try:
return sqlite_rows(path, query, params)
except Exception as exc:
return [{"_error": str(exc)}]
def main() -> int:
ap = argparse.ArgumentParser(description="AIOS read-only control-plane consistency auditor")
ap.add_argument("--task-id", required=True)
ap.add_argument("--admission-db", type=Path, default=Path("/var/lib/aios-gitops/bridge-v2-staging/task-admission.sqlite3"))
ap.add_argument("--task-db", type=Path, default=Path("/var/lib/aios-agent-task/state.sqlite3"))
ap.add_argument("--git-dir", type=Path, default=Path("/data/aios-gitops/repo.git"))
ap.add_argument("--branch")
ap.add_argument("--openapi", type=Path)
ap.add_argument("--server-source", type=Path)
ap.add_argument("--artifact-root", type=Path)
ap.add_argument("--now", type=int)
ap.add_argument("--out", type=Path, default=Path("aios-control-plane-audit.json"))
ns = ap.parse_args()
now = ns.now or int(time.time())
findings = []
admission = None
task = None
workers = []
lease = None
git = None
evidence = EvidenceSnapshot()
# Admission plane.
try:
receipt = sqlite_rows(ns.admission_db,
"SELECT * FROM task_admission_receipts WHERE task_id=? ORDER BY rowid DESC LIMIT 1", (ns.task_id,))
if receipt:
r = dict(receipt[0])
envelope_id = r.get("envelope_id")
env = sqlite_rows(ns.admission_db,
"SELECT * FROM task_admission_envelopes WHERE envelope_id=? LIMIT 1", (envelope_id,))
grant = sqlite_rows(ns.admission_db,
"SELECT * FROM task_admission_grants WHERE envelope_id=? ORDER BY rowid DESC LIMIT 1", (envelope_id,))
ed = dict(env[0]) if env else {}
gd = dict(grant[0]) if grant else {}
gj = jload(gd.get("grant_json"))
ej = jload(ed.get("envelope_json"))
admission = AdmissionSnapshot(
task_id=ns.task_id,
envelope_id=envelope_id,
admission_status=ed.get("status"),
grant_id=gd.get("grant_id"),
grant_expires_at=gj.get("expires_at"),
one_time_use=gj.get("one_time_use"),
authority_base_oid=r.get("authority_base_oid") or ej.get("authority_base_oid"),
execution_digest=r.get("execution_digest") or ej.get("execution_digest"),
requested_capabilities=tuple(ej.get("requested_capabilities") or ()),
file_allowlist=tuple(ej.get("file_allowlist") or ()),
admitted_at=r.get("created_at") or r.get("admitted_at"),
)
except Exception as exc:
findings.append(Finding("ADMISSION_PLANE_UNREADABLE", Severity.BLOCKER, str(exc)))
# TaskPlane.
try:
trows = sqlite_rows(ns.task_db, "SELECT * FROM tasks WHERE task_id=? LIMIT 1", (ns.task_id,))
if trows:
td = dict(trows[0])
meta = jload(td.get("metadata_json"))
required = jload(td.get("required_capabilities_json"))
if isinstance(required, dict):
required = required.get("capabilities") or []
task = TaskSnapshot(
task_id=ns.task_id,
status=td.get("status"),
claimed_by=td.get("claimed_by"),
lease_expires_at=td.get("lease_expires_at"),
attempt=td.get("attempt"),
revision=td.get("revision"),
created_at=td.get("created_at"),
updated_at=td.get("updated_at"),
deadline=td.get("deadline"),
required_capabilities=tuple(required or ()),
metadata=meta if isinstance(meta, dict) else {},
)
arows = sqlite_rows(ns.task_db, "SELECT * FROM agents")
for row in arows:
d = dict(row)
caps = jload(d.get("capabilities_json"))
if isinstance(caps, dict):
caps = caps.get("capabilities") or []
last = parse_iso_epoch(d.get("last_seen_at"))
workers.append(WorkerSnapshot(
worker_id=d.get("agent_id") or d.get("worker_id") or "unknown",
status=d.get("status"),
last_seen_epoch=last,
capabilities=tuple(caps or ()),
))
except Exception as exc:
findings.append(Finding("TASK_PLANE_UNREADABLE", Severity.BLOCKER, str(exc)))
branch = ns.branch
if not branch and task and isinstance(task.metadata, dict):
plan = task.metadata.get("execution_plan") or task.metadata.get("plan") or task.metadata
if isinstance(plan, dict):
branch = plan.get("target_branch")
if branch:
current = git_ref(ns.git_dir, f"refs/heads/{branch}")
git = GitSnapshot(branch=branch, current_oid=current, expected_oid=admission.authority_base_oid if admission else None)
if admission and current and admission.authority_base_oid and current != admission.authority_base_oid:
findings.append(Finding(
"BRANCH_BINDING_DRIFT",
Severity.BLOCKER,
"Current branch OID differs from Admission authority_base_oid.",
{"current": current, "bound": admission.authority_base_oid},
"Replan/rebind formally; do not reuse old execution_digest.",
))
truth, derived = derive_truth(admission, task, workers, lease, git, evidence, now)
findings.extend(derived)
if task:
findings.extend(check_capability_aliases(task.required_capabilities, workers))
# Plan/artifact integrity if the immutable plan points to files.
if task and ns.artifact_root:
plan = task.metadata.get("execution_plan") or task.metadata.get("plan") or task.metadata
if isinstance(plan, dict):
hashes = plan.get("file_sha256") or {}
for rel in plan.get("file_allowlist") or []:
findings.extend(inspect_artifact(ns.artifact_root / rel, hashes.get(rel)))
if ns.openapi and ns.server_source:
findings.extend(check_openapi_backend_parity(
ns.openapi, ns.server_source,
required_ops=("getTaskAdmissionEnvelope","submitTaskAdmissionAuthorization","getTaskAdmissionStatus"),
))
report = {
"schema": "aios.control-plane-consistency-audit.v1",
"observed_at": now,
"task_id": ns.task_id,
"truth_state": truth.value,
"admission": asdict(admission) if admission else None,
"task": asdict(task) if task else None,
"eligible_workers": [asdict(w) for w in eligible_workers(task, workers, now)] if task else [],
"git": asdict(git) if git else None,
"findings": [asdict(f) for f in findings],
"summary": {
"blockers": sum(f.severity == Severity.BLOCKER for f in findings),
"majors": sum(f.severity == Severity.MAJOR for f in findings),
"minors": sum(f.severity == Severity.MINOR for f in findings),
},
}
dump_report(ns.out, report)
print(json.dumps(report["summary"], sort_keys=True))
return 2 if report["summary"]["blockers"] else 1 if report["summary"]["majors"] else 0
if __name__ == "__main__":
raise SystemExit(main())
@@ -1,20 +0,0 @@
#!/usr/bin/env python3
from aios_control_plane_consistency import *
NOW=2_000_000_000
scenarios = {}
a=AdmissionSnapshot(task_id="T",admission_status="ADMITTED",grant_expires_at=NOW+60)
scenarios["dual_plane_missing_task"]=derive_truth(a,None,[],None,None,None,NOW)
t=TaskSnapshot(task_id="T",status="QUEUED",required_capabilities=("writer.lease",),created_at="2033-05-18T03:23:20+00:00")
scenarios["queued_no_worker"]=derive_truth(a,t,[],None,None,None,NOW)
t2=TaskSnapshot(task_id="T",status="RUNNING",claimed_by="ghost")
scenarios["fake_running"]=derive_truth(a,t2,[],None,None,None,NOW)
t3=TaskSnapshot(task_id="T",status="SUCCEEDED")
scenarios["fake_success"]=derive_truth(a,t3,[],None,GitSnapshot("agent/x",expected_oid="a"*40),EvidenceSnapshot(),NOW)
for name,(state,findings) in scenarios.items():
print(name,state.value,[f.code for f in findings])
+8 -1
View File
@@ -19,10 +19,17 @@ usage() {
[[ "$ACTION" =~ ^(create|remove)$ ]] || usage
[[ "$AGENT" =~ ^(openclaw|chatgpt|controller)$ ]] || usage
[[ "$TASK_ID" =~ ^[A-Za-z0-9._-]+$ ]] || {
[[ "$TASK_ID" =~ ^[A-Za-z0-9][A-Za-z0-9._-]*$ ]] || {
echo "ERROR: invalid task id" >&2
exit 65
}
# Reject path traversal: no dot-dot, no leading dot, no trailing dot.
case "$TASK_ID" in
*".."*|.*|*.)
echo "ERROR: invalid task id (path traversal rejected)" >&2
exit 65
;;
esac
[[ "$(id -un)" == "$RUNTIME_USER" ]] || {
echo "ERROR: must run as $RUNTIME_USER" >&2
exit 66
+10
View File
@@ -8,6 +8,16 @@ ZIP_PATH="$(find /data /opt /root /home -name "$ZIP_NAME" 2>/dev/null | head -1
if [ -z "$ZIP_PATH" ]; then echo "BLOCKED: missing $ZIP_NAME"; exit 10; fi
cp "$ZIP_PATH" "$ROOT/imports/"
cd "$WORKSPACE"
# zip-slip guard: reject any entry that would escape the workspace on extract.
command -v unzip >/dev/null 2>&1 || { echo "BLOCKED: unzip unavailable"; exit 11; }
while IFS= read -r entry; do
case "$entry" in
*/../*|../*|*/..|..|/*|*\\*|*:* )
echo "BLOCKED: unsafe zip entry: $entry"
exit 11
;;
esac
done < <(unzip -Z1 "$ROOT/imports/$ZIP_NAME" 2>/dev/null || true)
unzip -o "$ROOT/imports/$ZIP_NAME"
git init
git config user.name "AIOS GitOps Bot"
@@ -0,0 +1,52 @@
"""Read-only WP0c authorization inventory."""
from __future__ import annotations
from dataclasses import dataclass, asdict
from pathlib import Path
import argparse, json, re
TERMS={
"task_principal":re.compile(r"\bTaskPrincipal\b|\btask[_ -]?principal\b",re.I),
"owner_approval":re.compile(r"\bowner[_ -]?approval\b|\bOWNER_SIGNATURE_REQUIRED\b",re.I),
"capability":re.compile(r"\bcapabilit(?:y|ies)\b",re.I),
"signer":re.compile(r"\bsigner\b|\bsshsig\b",re.I),
"lease":re.compile(r"\b(?:ClaimLease|WriterLease|writer[_ -]?lease)\b",re.I),
"authorization":re.compile(r"\bauthori[sz]ation\b",re.I),
}
def scan(repo:Path)->dict:
repo=repo.resolve(); refs=[]
for p in sorted(repo.rglob("*")):
if not p.is_file() or p.suffix.lower() not in {".py",".json",".yaml",".yml",".md"}: continue
if any(x in p.parts for x in {".git",".venv","node_modules","__pycache__"}): continue
try: lines=p.read_text(encoding="utf-8",errors="strict").splitlines()
except (UnicodeDecodeError,OSError): continue
for i,line in enumerate(lines,1):
kinds=sorted(k for k,pat in TERMS.items() if pat.search(line))
if kinds:
refs.append({"path":p.relative_to(repo).as_posix(),"line":i,"kinds":kinds,"excerpt":line.strip()[:240]})
return {
"schema_version":"aios.authorization-inventory.v1",
"references":refs,
"reference_count":len(refs),
"classification":"EVIDENCE_ONLY_NOT_AUTHORITY",
"completeness":"PARTIAL_DISCOVERY",
"covered_scope":["lexical_authorization_references"],
"missing_scope":[
"owner_broker_chain",
"taskprincipal_admission_chain",
"claim_and_task_lease_chain",
"writerlease_chain",
"release_recovery_capability_chain",
"signer_credential_provisioning_chain",
"live_caller_and_revocation_evidence",
],
"g0_exit_eligible":False,
}
def main()->int:
ap=argparse.ArgumentParser(); ap.add_argument("--repo",type=Path,required=True); ap.add_argument("--out",type=Path)
ns=ap.parse_args(); result=scan(ns.repo); text=json.dumps(result,indent=2,sort_keys=True)+"\n"
if ns.out: ns.out.write_text(text,encoding="utf-8")
else: print(text,end="")
return 0
if __name__=="__main__": raise SystemExit(main())
+432
View File
@@ -0,0 +1,432 @@
"""Read-only caller scanner for AIOS platform convergence.
Key safety rules:
- static source references are evidence of source callers, not runtime liveness;
- runtime liveness requires active, artifact-bound readback evidence;
- PROVEN_ORPHAN requires a complete negative-evidence chain and independent GO;
- parse/runtime uncertainty is fail-closed to UNKNOWN;
- this module never imports scanned business modules and never mutates authority.
"""
from __future__ import annotations
from dataclasses import dataclass, asdict
from pathlib import Path
from typing import Iterable, Mapping, Any, Sequence
import argparse
import ast
import hashlib
import json
import keyword
@dataclass(frozen=True)
class Caller:
kind: str
identity: str
evidence_ref: str
@dataclass(frozen=True)
class ScanResult:
schema_version: str
subject: dict[str, str]
callers: tuple[Caller, ...]
runtime_loaded: tuple[dict[str, str], ...]
unknown_dynamic_refs: tuple[str, ...]
test_only_callers: tuple[Caller, ...]
caller_count: int
assessment: str
coverage: dict[str, object]
evidence_refs: tuple[str, ...]
def to_dict(self) -> dict[str, object]:
return {
"schema_version": self.schema_version,
"subject": self.subject,
"callers": [asdict(x) for x in self.callers],
"runtime_loaded": list(self.runtime_loaded),
"unknown_dynamic_refs": list(self.unknown_dynamic_refs),
"test_only_callers": [asdict(x) for x in self.test_only_callers],
"caller_count": self.caller_count,
"assessment": self.assessment,
"coverage": self.coverage,
"evidence_refs": list(self.evidence_refs),
}
def _sha(path: Path) -> str:
return hashlib.sha256(path.read_bytes()).hexdigest()
def _is_identifier_path(parts: Sequence[str]) -> bool:
return bool(parts) and all(p.isidentifier() and not keyword.iskeyword(p) for p in parts)
def _normalize_rel_path(value: object) -> str:
if not isinstance(value, str):
return ""
return value.replace("\\", "/").lstrip("./")
def _default_import_roots(repo: Path, profile: Mapping[str, Any]) -> tuple[Path, ...]:
roots: list[Path] = [repo]
explicit = profile.get("import_roots", ()) or ()
if isinstance(explicit, (list, tuple)):
for item in explicit:
if not isinstance(item, str) or not item.strip():
continue
path = (repo / item).resolve()
try:
path.relative_to(repo)
except ValueError:
continue
if path.exists():
roots.append(path)
for rel in ("active-runtime", "active-runtime/app", "src", "app"):
path = (repo / rel).resolve()
if path.exists():
roots.append(path)
unique: list[Path] = []
seen: set[str] = set()
for root in roots:
key = str(root)
if key not in seen:
seen.add(key)
unique.append(root)
return tuple(unique)
def _module_candidates(path: Path, roots: Sequence[Path]) -> tuple[str, ...]:
out: set[str] = set()
resolved = path.resolve()
for root in roots:
try:
rel = resolved.relative_to(root.resolve())
except ValueError:
continue
parts = list(rel.with_suffix("").parts)
if parts and parts[-1] == "__init__":
parts = parts[:-1]
if _is_identifier_path(parts):
out.add(".".join(parts))
return tuple(sorted(out))
def _is_test_path(path: Path) -> bool:
parts = {p.lower() for p in path.parts}
return "tests" in parts or "test" in parts or path.name.startswith("test_")
def _iter_python(repo: Path) -> Iterable[Path]:
for path in repo.rglob("*.py"):
if ".git" in path.parts:
continue
if path.is_file():
yield path
def _relative_imports(
*,
importer_modules: Sequence[str],
is_package_init: bool,
level: int,
module: str | None,
aliases: Sequence[str],
) -> set[str]:
resolved: set[str] = set()
for importer in importer_modules:
parts = importer.split(".") if importer else []
package = parts if is_package_init else parts[:-1]
ascend = max(level - 1, 0)
if ascend > len(package):
continue
base = package[: len(package) - ascend] if ascend else package
if module:
base = [*base, *module.split(".")]
if base:
resolved.add(".".join(base))
for alias in aliases:
if alias == "*":
continue
candidate = [*base, alias]
if candidate:
resolved.add(".".join(candidate))
return resolved
def _ast_imports(path: Path, importer_modules: Sequence[str]) -> tuple[set[str], bool]:
try:
source = path.read_text(encoding="utf-8", errors="strict")
tree = ast.parse(source)
except (SyntaxError, UnicodeError, OSError):
return set(), True
imports: set[str] = set()
dynamic = False
for node in ast.walk(tree):
if isinstance(node, ast.Import):
imports.update(alias.name for alias in node.names)
elif isinstance(node, ast.ImportFrom):
alias_names = [alias.name for alias in node.names]
if node.level:
imports.update(
_relative_imports(
importer_modules=importer_modules,
is_package_init=path.name == "__init__.py",
level=node.level,
module=node.module,
aliases=alias_names,
)
)
else:
if node.module:
imports.add(node.module)
imports.update(
f"{node.module}.{name}" for name in alias_names if name != "*"
)
elif alias_names:
# "from <unknown> import x" should not happen for absolute syntax;
# keep fail-closed if the AST is not informative.
dynamic = True
elif isinstance(node, ast.Call):
fn = node.func
name = None
if isinstance(fn, ast.Name):
name = fn.id
elif isinstance(fn, ast.Attribute) and isinstance(fn.value, ast.Name):
name = f"{fn.value.id}.{fn.attr}"
if name in {"__import__", "importlib.import_module"}:
if (
node.args
and isinstance(node.args[0], ast.Constant)
and isinstance(node.args[0].value, str)
and not node.args[0].value.startswith(".")
):
imports.add(node.args[0].value)
else:
dynamic = True
return imports, dynamic
def _match_module(imported: str, subject_modules: Sequence[str]) -> bool:
return any(
imported == subject
or imported.startswith(subject + ".")
or subject.startswith(imported + ".")
for subject in subject_modules
)
def _active(item: Mapping[str, Any]) -> bool:
if item.get("active") is True:
return True
state = str(item.get("state") or item.get("status") or "").upper()
return state in {"ACTIVE", "RUNNING", "LOADED", "READY"}
def _artifact_bound_match(
item: Mapping[str, Any],
*,
subject_rel: str,
subject_sha: str,
) -> bool:
if not _active(item):
return False
source_path = _normalize_rel_path(
item.get("source_path") or item.get("repo_path") or item.get("path")
)
artifact_sha = str(
item.get("source_sha256")
or item.get("artifact_sha256")
or item.get("sha256")
or ""
).lower()
return bool(source_path == subject_rel and artifact_sha == subject_sha)
def _probe_complete(profile: Mapping[str, Any]) -> tuple[bool, dict[str, object]]:
runtime_hosts = list(profile.get("runtime_hosts", ()) or ())
observation_seconds = profile.get("observation_window_seconds")
try:
observation_seconds = float(observation_seconds)
except (TypeError, ValueError):
observation_seconds = 0.0
coverage = {
"repo_static": True,
"systemd": bool(profile.get("systemd_scanned", False)),
"runtime_hosts": runtime_hosts,
"runtime_hosts_complete": bool(profile.get("runtime_hosts_complete", False)),
"loaded_bytes": bool(profile.get("loaded_bytes_scanned", False)),
"api_routes": bool(profile.get("api_routes_scanned", False)),
"scheduler": bool(profile.get("scheduler_scanned", False)),
"observation_window_complete": bool(profile.get("observation_window_complete", False)),
"observation_window_seconds": observation_seconds,
"rollback_ref": str(profile.get("rollback_ref") or ""),
"independent_acceptance": str(profile.get("independent_acceptance") or ""),
"probe_failures": list(profile.get("probe_failures", ()) or ()),
}
complete = all(
(
coverage["systemd"],
coverage["runtime_hosts_complete"],
bool(runtime_hosts), # empty relevant-host set never proves absence
coverage["loaded_bytes"],
coverage["api_routes"],
coverage["scheduler"],
coverage["observation_window_complete"],
observation_seconds > 0,
bool(coverage["rollback_ref"]),
coverage["independent_acceptance"] == "GO",
not coverage["probe_failures"],
)
)
return bool(complete), coverage
def scan(
*,
repo: Path,
subject: Path,
runtime_profile: Mapping[str, Any] | None = None,
authority_ref: str | None = None,
) -> ScanResult:
repo = repo.resolve()
profile = dict(runtime_profile or {})
subject = (repo / subject).resolve() if not subject.is_absolute() else subject.resolve()
if not subject.is_file():
raise FileNotFoundError(subject)
roots = _default_import_roots(repo, profile)
subject_modules = _module_candidates(subject, roots)
production_static: list[Caller] = []
runtime_callers: list[Caller] = []
tests: list[Caller] = []
unknown_dynamic: list[str] = []
subject_sha = _sha(subject)
subject_rel = subject.relative_to(repo).as_posix()
evidence = [f"sha256:{subject_sha}"]
for path in _iter_python(repo):
if path.resolve() == subject:
continue
importer_modules = _module_candidates(path, roots)
imports, dynamic = _ast_imports(path, importer_modules)
matched = bool(subject_modules and any(_match_module(x, subject_modules) for x in imports))
rel = path.relative_to(repo).as_posix()
if matched:
caller = Caller(
"STATIC_TEST_ONLY" if _is_test_path(path) else "STATIC_PRODUCTION",
rel,
f"repo:{rel}",
)
(tests if _is_test_path(path) else production_static).append(caller)
if dynamic:
unknown_dynamic.append(rel)
runtime_loaded: list[dict[str, str]] = []
for item in profile.get("runtime_loaded", ()) or ():
if not isinstance(item, Mapping):
continue
if _artifact_bound_match(item, subject_rel=subject_rel, subject_sha=subject_sha):
loaded_path = _normalize_rel_path(
item.get("source_path") or item.get("repo_path") or item.get("path")
)
runtime_loaded.append({"path": loaded_path, "sha256": subject_sha})
runtime_callers.append(
Caller("RUNTIME_LOADED", loaded_path, str(item.get("evidence_ref") or "runtime_profile"))
)
for key, kind, identity_key in (
("systemd", "SYSTEMD", "unit"),
("api_routes", "API_ROUTE", "route"),
("scheduler", "SCHEDULER", "identity"),
):
for item in profile.get(key, ()) or ():
if not isinstance(item, Mapping):
continue
if _artifact_bound_match(item, subject_rel=subject_rel, subject_sha=subject_sha):
runtime_callers.append(
Caller(
kind,
str(item.get(identity_key) or ""),
str(item.get("evidence_ref") or f"runtime_profile:{key}"),
)
)
def _dedup(items: Sequence[Caller]) -> list[Caller]:
unique = {(c.kind, c.identity, c.evidence_ref): c for c in items}
return sorted(unique.values(), key=lambda c: (c.kind, c.identity, c.evidence_ref))
production_static = _dedup(production_static)
runtime_callers = _dedup(runtime_callers)
tests = _dedup(tests)
production = _dedup([*production_static, *runtime_callers])
negative_complete, coverage = _probe_complete(profile)
opaque = bool(unknown_dynamic) or bool(coverage["probe_failures"])
if runtime_callers:
assessment = "LIVE"
elif production_static or opaque:
# Source caller != runtime liveness; opaque/probe failures are also unknown.
assessment = "UNKNOWN"
elif negative_complete:
assessment = "PROVEN_ORPHAN"
else:
assessment = "LIKELY_ORPHAN"
coverage["import_roots"] = [
root.relative_to(repo).as_posix() if root != repo else "."
for root in roots
]
coverage["subject_modules"] = list(subject_modules)
coverage["negative_evidence_complete"] = negative_complete
coverage["static_production_callers"] = len(production_static)
coverage["runtime_callers"] = len(runtime_callers)
return ScanResult(
schema_version="aios.caller-scan.v2",
subject={
"path": subject_rel,
"source_commit": authority_ref or "",
"sha256": subject_sha,
},
callers=tuple(production),
runtime_loaded=tuple(runtime_loaded),
unknown_dynamic_refs=tuple(sorted(set(unknown_dynamic))),
test_only_callers=tuple(tests),
caller_count=len(production),
assessment=assessment,
coverage=coverage,
evidence_refs=tuple(evidence),
)
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--repo", type=Path, required=True)
ap.add_argument("--subject", type=Path, required=True)
ap.add_argument("--runtime-host-profile", type=Path)
ap.add_argument("--authority-ref")
ap.add_argument("--output", type=Path, required=True)
ns = ap.parse_args()
profile: dict[str, Any] = {}
if ns.runtime_host_profile:
profile = json.loads(ns.runtime_host_profile.read_text(encoding="utf-8"))
result = scan(
repo=ns.repo,
subject=ns.subject,
runtime_profile=profile,
authority_ref=ns.authority_ref,
)
ns.output.parent.mkdir(parents=True, exist_ok=True)
ns.output.write_text(
json.dumps(result.to_dict(), ensure_ascii=False, indent=2) + "\n",
encoding="utf-8",
)
print(json.dumps({"assessment": result.assessment, "caller_count": result.caller_count}, sort_keys=True))
return 0
if __name__ == "__main__":
raise SystemExit(main())
@@ -0,0 +1,63 @@
"""Read-only WP0d canonical/digest inventory."""
from __future__ import annotations
from pathlib import Path
import argparse, ast, json
CANON_NAMES=("canonical","digest","sha","fingerprint","execution_digest","policy_digest")
def _functions(path:Path):
try: tree=ast.parse(path.read_text(encoding="utf-8"))
except (SyntaxError,UnicodeDecodeError,OSError): return []
out=[]
for node in ast.walk(tree):
if isinstance(node,(ast.FunctionDef,ast.AsyncFunctionDef)):
name=node.name.lower()
source=ast.get_source_segment(path.read_text(encoding="utf-8"),node) or ""
signals=[]
if any(k in name for k in CANON_NAMES): signals.append("NAME")
if "sort_keys=True" in source and "json.dumps" in source: signals.append("SORTED_JSON")
if "hashlib.sha256" in source or "sha256(" in source: signals.append("SHA256")
if signals:
out.append({"name":node.name,"line":node.lineno,"signals":sorted(set(signals))})
return out
def scan(repo:Path)->dict:
repo=repo.resolve(); defs=[]
for p in sorted(repo.rglob("*.py")):
if any(x in p.parts for x in {".git",".venv","node_modules","__pycache__"}): continue
funcs=_functions(p)
for item in funcs:
defs.append({"path":p.relative_to(repo).as_posix(),**item})
groups={}
for item in defs:
key="+".join(item["signals"])
groups.setdefault(key,[]).append(f'{item["path"]}:{item["line"]}:{item["name"]}')
return {
"schema_version":"aios.canonical-digest-inventory.v1",
"definitions":defs,
"duplicate_groups":groups,
"definition_count":len(defs),
"classification":"EVIDENCE_ONLY_NOT_CANONICAL_AUTHORITY",
"completeness":"PARTIAL_DISCOVERY",
"covered_scope":["function_level_name_and_syntax_signals"],
"missing_scope":[
"producer_consumer_mapping",
"input_schema",
"bytes_rule",
"digest_algorithm_contract",
"duplicate_semantics_review",
"live_caller_evidence",
],
"g0_exit_eligible":False,
}
def main()->int:
ap=argparse.ArgumentParser(); ap.add_argument("--repo",type=Path,required=True); ap.add_argument("--out",type=Path)
ns=ap.parse_args(); result=scan(ns.repo); text=json.dumps(result,indent=2,sort_keys=True)+"\n"
if ns.out: ns.out.write_text(text,encoding="utf-8")
else: print(text,end="")
return 0
if __name__=="__main__": raise SystemExit(main())
+60
View File
@@ -0,0 +1,60 @@
"""ChangeEnvelope v1 generator.
The envelope records a proposed change; it neither authorizes nor executes it.
"""
from __future__ import annotations
from pathlib import Path
from typing import Iterable, Mapping
import argparse, hashlib, json, time
def sha256(path:Path)->str:
return hashlib.sha256(path.read_bytes()).hexdigest()
def build_change_envelope(
*,
change_id:str,
work_package:str,
source_commit:str,
candidate_root:Path,
target_paths:Iterable[str],
preimage_sha256:Mapping[str,str|None],
contract_versions:Mapping[str,str],
live_caller_evidence_ref:str|None,
tests:list[dict],
rollback_ref:str|None,
independent_review_ref:str|None,
)->dict:
paths=sorted(dict.fromkeys(str(p) for p in target_paths))
post={}
for rel in paths:
path=candidate_root/rel
if not path.is_file(): raise FileNotFoundError(rel)
post[rel]=sha256(path)
return {
"schema_version":"aios.change-envelope.v1",
"change_id":change_id,
"work_package":work_package,
"source_commit":source_commit,
"target_paths":paths,
"preimage_sha256":{p:preimage_sha256.get(p) for p in paths},
"postimage_sha256":post,
"contract_versions":dict(contract_versions),
"live_caller_evidence_ref":live_caller_evidence_ref,
"tests":tests,
"rollback_ref":rollback_ref,
"independent_review_ref":independent_review_ref,
"created_at_epoch":int(time.time()),
"authorization_effect":"NONE",
}
def main()->int:
ap=argparse.ArgumentParser()
ap.add_argument("--spec",type=Path,required=True); ap.add_argument("--candidate-root",type=Path,required=True); ap.add_argument("--out",type=Path,required=True)
ns=ap.parse_args(); spec=json.loads(ns.spec.read_text(encoding="utf-8"))
result=build_change_envelope(candidate_root=ns.candidate_root,**spec)
ns.out.write_text(json.dumps(result,indent=2,sort_keys=True)+"\n",encoding="utf-8")
return 0
if __name__=="__main__": raise SystemExit(main())
@@ -0,0 +1,129 @@
"""Environment capability profile with secret-safe, read-only probes.
The profile distinguishes "tool present" from "capability proven". Missing
requirements are computed from an explicit requirement set; an empty
requirement set never implies Generation-G0 completeness.
"""
from __future__ import annotations
from dataclasses import dataclass, asdict
from typing import Iterable
import argparse
import os
import platform
import shutil
import subprocess
from pathlib import Path
@dataclass(frozen=True)
class EnvironmentProfile:
schema_version: str
host_id: str
os_info: dict[str, str]
shells: tuple[str, ...]
tools: dict[str, str | None]
capabilities: dict[str, bool]
credential_mechanisms: tuple[str, ...]
filesystem: dict[str, object]
requirements: tuple[str, ...]
missing_requirements: tuple[str, ...]
completeness: str
g0_exit_eligible: bool
def to_dict(self): return asdict(self)
def _version(command: str) -> str | None:
exe = shutil.which(command)
if not exe:
return None
try:
cp = subprocess.run(
[exe, "--version"], text=True, capture_output=True, timeout=3
)
except Exception:
return None
first = (cp.stdout or cp.stderr).splitlines()
return first[0][:200] if first else None
def _systemd_running() -> bool:
if os.name != "posix" or not shutil.which("systemctl"):
return False
# Presence of systemctl alone is not evidence that systemd is PID1/runtime.
return Path("/run/systemd/system").is_dir()
def collect(host_id: str, requirements: Iterable[str] = ()) -> EnvironmentProfile:
requirements = tuple(dict.fromkeys(str(x) for x in requirements if str(x)))
shells = tuple(x for x in ("bash", "sh", "pwsh", "powershell", "cmd") if shutil.which(x))
tools = {name: _version(name) for name in ("git", "python3", "python", "node")}
posix = os.name == "posix"
systemd = _systemd_running()
sudo_noninteractive = False
if posix and shutil.which("sudo"):
try:
sudo_noninteractive = (
subprocess.run(
["sudo", "-n", "true"],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
timeout=2,
).returncode
== 0
)
except Exception:
sudo_noninteractive = False
load_credential = bool(systemd and shutil.which("systemd-creds"))
capabilities = {
"posix": posix,
"systemd": systemd,
"systemd_load_credential": load_credential,
"sudo_noninteractive": sudo_noninteractive,
"git": bool(tools["git"]),
"python": bool(tools["python3"] or tools["python"]),
"node": bool(tools["node"]),
}
credential = ("systemd.LoadCredential",) if load_credential else ()
missing = tuple(sorted(req for req in requirements if not capabilities.get(req, False)))
completeness = "REQUIREMENTS_EVALUATED" if requirements else "CAPABILITY_DISCOVERY_ONLY"
return EnvironmentProfile(
schema_version="aios.environment-profile.v2",
host_id=host_id,
os_info={"name": platform.system(), "version": platform.version()},
shells=shells,
tools=tools,
capabilities=capabilities,
credential_mechanisms=credential,
filesystem={
"case_sensitive": os.path.normcase("A") != os.path.normcase("a"),
"line_ending_default": "LF" if os.linesep == "\n" else "CRLF",
},
requirements=requirements,
missing_requirements=missing,
completeness=completeness,
g0_exit_eligible=bool(requirements and not missing),
)
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--host-id", required=True)
ap.add_argument("--require", action="append", default=[])
ap.add_argument("--out", type=Path)
ns = ap.parse_args()
result = collect(ns.host_id, ns.require)
import json
text = json.dumps(result.to_dict(), indent=2, sort_keys=True) + "\n"
if ns.out:
ns.out.write_text(text, encoding="utf-8")
else:
print(text, end="")
return 0
if __name__ == "__main__":
raise SystemExit(main())
@@ -0,0 +1,63 @@
"""Read-only WP0b Git writer inventory."""
from __future__ import annotations
from dataclasses import dataclass, asdict
from pathlib import Path
import argparse, json, re
PATTERNS={
"git_push":re.compile(r"\bgit\s+push\b"),
"git_commit":re.compile(r"\bgit\s+commit\b"),
"git_update_ref":re.compile(r"\bgit\s+update-ref\b"),
"governed_commit":re.compile(r"\b(?:commitFilesGoverned|commit_files_governed|commit-files)\b"),
"tag_publish":re.compile(r"\b(?:requestOrCreateTag|createReleaseCandidate|tag_publish)\b"),
"ref_mutation_api":re.compile(r"\b(?:update_ref|create_commit|push_branch|receive-pack)\b"),
}
SKIP_PARTS={".git",".venv","venv","node_modules","__pycache__"}
@dataclass(frozen=True)
class WriterEvidence:
path:str; line:int; kind:str; excerpt:str
def scan(repo:Path)->dict:
repo=repo.resolve()
evidence=[]
for p in sorted(repo.rglob("*")):
if not p.is_file() or any(x in SKIP_PARTS for x in p.parts):
continue
if p.suffix.lower() not in {".py",".sh",".bash",".service",".json",".yaml",".yml",".md"}:
continue
try: lines=p.read_text(encoding="utf-8",errors="strict").splitlines()
except (UnicodeDecodeError,OSError): continue
for number,line in enumerate(lines,1):
for kind,pattern in PATTERNS.items():
if pattern.search(line):
evidence.append(WriterEvidence(p.relative_to(repo).as_posix(),number,kind,line.strip()[:240]))
unique={(e.path,e.line,e.kind,e.excerpt):e for e in evidence}
ordered=sorted(unique.values(),key=lambda e:(e.path,e.line,e.kind))
return {
"schema_version":"aios.git-writer-inventory.v1",
"evidence":[asdict(e) for e in ordered],
"writer_candidate_count":len(ordered),
"classification":"EVIDENCE_ONLY_NOT_AUTHORITY",
"completeness":"PARTIAL_DISCOVERY",
"covered_scope":["lexical_writer_candidates"],
"missing_scope":[
"writer_to_authorization_mapping",
"repository_and_ref_resolution",
"cas_semantics",
"receipt_contract",
"live_caller_and_runtime_identity",
],
"g0_exit_eligible":False,
}
def main()->int:
ap=argparse.ArgumentParser(); ap.add_argument("--repo",type=Path,required=True); ap.add_argument("--out",type=Path)
ns=ap.parse_args(); result=scan(ns.repo); text=json.dumps(result,indent=2,sort_keys=True)+"\n"
if ns.out: ns.out.write_text(text,encoding="utf-8")
else: print(text,end="")
return 0
if __name__=="__main__": raise SystemExit(main())
@@ -0,0 +1,120 @@
"""Retry / Toil Governor V2.
Pure decision layer. It never performs the business operation.
The governor is deliberately conservative:
- unknown side effects reconcile;
- proven success is reused/suppressed, never executed again;
- unrecognized states block;
- only explicitly executable states may return EXECUTE.
Fingerprint bytes reuse the existing TaskPlane ``canonical_json`` contract.
"""
from __future__ import annotations
from dataclasses import dataclass
from enum import StrEnum
from pathlib import Path
import hashlib
import sys
# Reuse the existing in-repo canonical JSON contract. The platform tool is
# intentionally not a second canonical implementation.
try:
from app.task_plane.models import canonical_json
except ModuleNotFoundError: # standalone in-repo CLI execution
_repo_root = Path(__file__).resolve().parents[2]
_runtime = _repo_root / "active-runtime"
if str(_runtime) not in sys.path:
sys.path.insert(0, str(_runtime))
from app.task_plane.models import canonical_json
RETRY_FINGERPRINT_SCHEMA = "aios.retry-toil-fingerprint.v1"
UNKNOWN_STATES = frozenset(
{"UNKNOWN", "RECONCILIATION_REQUIRED", "EXECUTION_OUTCOME_UNKNOWN", "REMOTE_SYNCING"}
)
BLOCKED_STATES = frozenset({"BLOCKED", "OWNER_BOUNDARY", "NO_GO"})
SUCCESS_STATES = frozenset(
{"SUCCEEDED", "REMOTE_VERIFIED", "COMPLETED", "APPLIED", "ACCEPTED"}
)
EXECUTABLE_STATES = frozenset(
{"NOT_APPLIED", "READY", "FAILED_RETRYABLE", "RETRYABLE"}
)
class Decision(StrEnum):
EXECUTE = "EXECUTE"
RECONCILE = "RECONCILE"
SUPPRESS_NOOP = "SUPPRESS_NOOP"
BLOCK = "BLOCK"
@dataclass(frozen=True)
class RetryObservation:
operation_fingerprint: str
blocker_fingerprint: str
evidence_revision: str
side_effect_state: str
attempt: int
@dataclass(frozen=True)
class RetryDecision:
decision: Decision
reason: str
def fingerprint(value: object) -> str:
payload = {
"schema": RETRY_FINGERPRINT_SCHEMA,
"value": value,
}
return hashlib.sha256(canonical_json(payload).encode("utf-8")).hexdigest()
def decide(previous: RetryObservation | None, current: RetryObservation) -> RetryDecision:
state = str(current.side_effect_state or "").upper()
if state in UNKNOWN_STATES:
return RetryDecision(
Decision.RECONCILE,
"unknown/indeterminate side effect must reconcile before retry",
)
if state in BLOCKED_STATES:
return RetryDecision(Decision.BLOCK, "governance or dependency boundary")
if state in SUCCESS_STATES:
return RetryDecision(
Decision.SUPPRESS_NOOP,
"current effect is already proven successful; reuse receipt/readback",
)
if state not in EXECUTABLE_STATES:
return RetryDecision(
Decision.BLOCK,
f"unrecognized or non-executable side effect state: {state or '<empty>'}",
)
if previous is not None:
prev_state = str(previous.side_effect_state or "").upper()
same_effect = previous.operation_fingerprint == current.operation_fingerprint
if same_effect and prev_state in SUCCESS_STATES:
return RetryDecision(
Decision.RECONCILE,
"current executable observation conflicts with prior proven success",
)
if (
same_effect
and previous.blocker_fingerprint == current.blocker_fingerprint
and previous.evidence_revision == current.evidence_revision
):
return RetryDecision(
Decision.SUPPRESS_NOOP,
"no state/evidence change since prior attempt",
)
return RetryDecision(
Decision.EXECUTE,
"explicitly executable state with new evidence/effect context",
)
@@ -0,0 +1,133 @@
"""Read-only WP0a version/baseline discovery.
This is a *partial discovery producer*, not a version authority. It compares
already-discovered Git/runtime/production OIDs and records which Generation-G0
inventory dimensions are still missing.
Generation G0 must not be declared complete from this result alone.
"""
from __future__ import annotations
from dataclasses import dataclass, asdict
from pathlib import Path
import argparse, json, re, subprocess
OID_RE = re.compile(r"^[0-9a-f]{40}$")
REQUIRED_SCOPE = (
"git_authority_ref",
"runtime_source_identity",
"production_release_identity",
"architecture_current_projection",
"baseline_current_projection",
"VERSION_projection",
"README_projection",
"versions_directory",
"release_metadata_readers_writers_callers",
)
@dataclass(frozen=True)
class VersionInventory:
schema_version: str
authority_oid: str
runtime_oid: str
production_oid: str
runtime_descends_from_authority: bool | None
production_descends_from_authority: bool | None
authority_runtime_equal: bool
authority_production_equal: bool
conflicts: tuple[str, ...]
classification: str
completeness: str
covered_scope: tuple[str, ...]
missing_scope: tuple[str, ...]
g0_exit_eligible: bool
def to_dict(self): return asdict(self)
def _oid(value: str, name: str) -> str:
value = str(value or "").strip().lower()
if not OID_RE.fullmatch(value):
raise ValueError(f"invalid {name} oid")
return value
def _ancestor(repo: Path, left: str, right: str) -> bool | None:
cp = subprocess.run(
["git", "-C", str(repo), "merge-base", "--is-ancestor", left, right],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
)
if cp.returncode == 0:
return True
if cp.returncode == 1:
return False
return None
def collect(*, repo: Path, authority_oid: str, runtime_oid: str, production_oid: str) -> VersionInventory:
repo = repo.resolve()
authority = _oid(authority_oid, "authority")
runtime = _oid(runtime_oid, "runtime")
production = _oid(production_oid, "production")
rd = _ancestor(repo, authority, runtime) if repo.exists() else None
pd = _ancestor(repo, authority, production) if repo.exists() else None
conflicts = []
if authority != runtime:
conflicts.append("AUTHORITY_RUNTIME_DIVERGENCE")
if authority != production:
conflicts.append("AUTHORITY_PRODUCTION_DIVERGENCE")
if rd is False:
conflicts.append("RUNTIME_NOT_DESCENDANT_OF_AUTHORITY")
if pd is False:
conflicts.append("PRODUCTION_NOT_DESCENDANT_OF_AUTHORITY")
covered = (
"git_authority_ref",
"runtime_source_identity",
"production_release_identity",
)
missing = tuple(item for item in REQUIRED_SCOPE if item not in covered)
return VersionInventory(
"aios.version-baseline-inventory.v2",
authority,
runtime,
production,
rd,
pd,
authority == runtime,
authority == production,
tuple(conflicts),
"EVIDENCE_ONLY_NOT_AUTHORITY",
"PARTIAL_DISCOVERY",
covered,
missing,
False,
)
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--repo", type=Path, required=True)
ap.add_argument("--authority-oid", required=True)
ap.add_argument("--runtime-oid", required=True)
ap.add_argument("--production-oid", required=True)
ap.add_argument("--out", type=Path)
ns = ap.parse_args()
result = collect(
repo=ns.repo,
authority_oid=ns.authority_oid,
runtime_oid=ns.runtime_oid,
production_oid=ns.production_oid,
)
text = json.dumps(result.to_dict(), indent=2, sort_keys=True) + "\n"
if ns.out:
ns.out.write_text(text, encoding="utf-8")
else:
print(text, end="")
return 0
if __name__ == "__main__":
raise SystemExit(main())
+6 -1
View File
@@ -1,7 +1,12 @@
#!/usr/bin/env bash
set -euo pipefail
ROOT="$(cd "$(dirname "$0")/.." && pwd)"
bash "$ROOT/scripts/silent_install.sh"
SILENT_INSTALL="$ROOT/scripts/silent_install.sh"
if [ ! -f "$SILENT_INSTALL" ]; then
echo "[V5] ERROR: missing required installer: $SILENT_INSTALL" >&2
exit 1
fi
bash "$SILENT_INSTALL"
echo "V5_SILENT_INSTALL_OK"
echo "WebOps: https://ops.cnnetmkt.top"
echo "Digital Life: https://ops.cnnetmkt.top/life"
+19 -3
View File
@@ -9,11 +9,17 @@ from http.server import BaseHTTPRequestHandler, HTTPServer
import json
import os
import pathlib
import re
import time
EVENT_DIR = pathlib.Path(os.environ.get("AIOS_EVENT_DIR", "runtime/gitops/events"))
EVENT_DIR = pathlib.Path(os.environ.get("AIOS_EVENT_DIR", "runtime/gitops/events")).resolve()
EVENT_DIR.mkdir(parents=True, exist_ok=True)
# Strictly-constrained event ID: a single safe path segment. Rejects absolute
# paths, separators, and any traversal (`..`, backslashes) before path join.
_SAFE_EVENT_ID = re.compile(r"^[A-Za-z0-9][A-Za-z0-9_-]*$")
MAX_EVENT_FILE_BYTES = 8 * 1024 * 1024
class Handler(BaseHTTPRequestHandler):
def do_POST(self):
length = int(self.headers.get('Content-Length', 0))
@@ -26,8 +32,18 @@ class Handler(BaseHTTPRequestHandler):
return
event_id = event.get("id", f"evt-{int(time.time())}")
path = EVENT_DIR / f"{event_id}.json"
path.write_text(json.dumps(event, indent=2), encoding='utf-8')
if not isinstance(event_id, str) or not _SAFE_EVENT_ID.fullmatch(event_id):
self.send_response(400)
self.end_headers()
return
target = (EVENT_DIR / f"{event_id}.json").resolve()
if target.parent != EVENT_DIR:
# Defensive: resolved path must remain directly inside the event dir.
self.send_response(400)
self.end_headers()
return
target.write_text(json.dumps(event, indent=2), encoding='utf-8')
self.send_response(200)
self.end_headers()
-522
View File
@@ -1,522 +0,0 @@
"""ANNEX-2.3 R2 governed auto-claim full-chain and A23 positive/negative tests.
Covers the acceptance line:
SchedulabilityProof -> authoritative_claim_recheck -> TaskPlane claim
-> Agent automatic pull -> receipt/readback
and the A23 negative checks with their error codes. Uses the governed
authorization profile (execution_enabled=True + {"*": ALL_ACTION_CAPABILITIES})
per ARCH-AUTO-CLAIM-R2-20260914 §8.
"""
from __future__ import annotations
import json
import threading
import time
import urllib.error
import urllib.request
from http.server import ThreadingHTTPServer
from unittest import mock
import pytest
from agent_pull_runner import AgentPullRunner, RunnerConfig
from task_plane.api import TaskPlaneRequestHandler, TaskPlaneService
from task_plane.errors import ConflictError, ForbiddenError, LockedError
from task_plane.http_transport import HttpRunnerTransport
from task_plane.policy import ALL_ACTION_CAPABILITIES, PolicyEngine
from task_plane.receipts import receipt_digest
from task_plane.recheck import Check, RecheckDecision, RecheckEngine
from task_plane.schedulability import ProofBuilder, compute_proof_digest
from task_plane.schedule_service import SchedulingService
from task_plane.store import TaskPlaneStore
IDENTITY = "spiffe://aios/agent/r2"
WORKER = "r2-worker"
REG = {
"worker_id": WORKER,
"worker_type": "test",
"identity": IDENTITY,
"capabilities": ["code.read"],
"trust_zone": "tenant-A",
"max_concurrency": 1,
"heartbeat_interval_seconds": 15,
"supported_protocols": ["aios-task-v1"],
}
def make_service() -> tuple[TaskPlaneStore, TaskPlaneService]:
store = TaskPlaneStore(":memory:")
policy = PolicyEngine(
execution_enabled=True,
identity_capabilities={"*": set(ALL_ACTION_CAPABILITIES)},
)
service = TaskPlaneService(store, policy=policy)
return store, service
def register_worker(service, **overrides) -> None:
payload = dict(REG)
payload.update(overrides)
service.register_worker(payload)
def _assert_proof_ok(service, identity, proof_id):
proof = service.get_proof(proof_id, identity)
assert proof["status"] == "CLAIMED"
payload = {k: v for k, v in proof.items() if k != "digest"}
assert compute_proof_digest(payload) == proof["digest"]
return proof
def _assert_receipt_ok(service, identity, receipt_id, kind=None):
receipt = service.get_receipt(receipt_id, identity)
assert receipt_digest(receipt["payload"]) == receipt["digest"]
if kind:
assert receipt["kind"] == kind
return receipt
# -- REQ: full chain ---------------------------------------------------------
def test_r2_full_chain_poll_claim_and_readback():
store, service = make_service()
register_worker(service)
task = service.create_task(
{
"task_id": "task-r2-a",
"required_capabilities": ["code.read"],
"metadata": {"tenant_id": "tenant-A"},
}
)
result = service.poll(WORKER, IDENTITY)
assert result["task"]["task_id"] == "task-r2-a"
claim_id = result["claim_id"]
proof_id = result["proof_id"]
receipt_id = result["receipt_id"]
claim = store.get_claim(claim_id)
assert claim.status == "ACTIVE"
assert result["generation"]["task_revision"] == claim.task_revision
assert result["generation"]["attempt"] == claim.attempt
proof = _assert_proof_ok(service, IDENTITY, proof_id)
assert proof["claim_id"] == claim_id
assert proof["binding"]["worker_identity"] == IDENTITY
_assert_receipt_ok(service, IDENTITY, receipt_id, kind="claim")
receipts = service.list_task_receipts("task-r2-a", IDENTITY)
kinds = {r["kind"] for r in receipts}
assert kinds == {"claim", "proof"}
event_types = {e["type"] for e in store.list_events("task-r2-a")}
assert "CLAIM_RECHECK_ACCEPTED" in event_types
assert "CLAIM_GOVERNED_ACCEPTED" in event_types
def test_r2_governed_claim_atomic_single_transaction_fault_free():
store, service = make_service()
register_worker(service)
task = service.create_task(
{"task_id": "task-r2-atomic", "metadata": {"tenant_id": "tenant-A"}}
)
with mock.patch("task_plane.store._now", return_value=time.time() + 0.01):
result = service.poll(WORKER, IDENTITY)
assert result["claim_id"]
# Proving atomicity: proof/receipt/event all present after one claim.
proof = service.get_proof(result["proof_id"], IDENTITY)
assert proof["status"] == "CLAIMED"
receipts = service.list_task_receipts("task-r2-atomic", IDENTITY)
assert len(receipts) == 2
assert store.verify_event_chain("task-r2-atomic")
# -- A23 negative cases ------------------------------------------------------
def test_a23_enumeration_skips_cross_tenant_with_audit():
# G4: enumeration of a cross-tenant task must SKIP + record a structured
# audit finding (NOT fail closed). A tenant-B worker polling with only a
# tenant-A task sees idle/no_task and records a skip+audit finding.
store, service = make_service()
register_worker(service, trust_zone="tenant-B") # worker in tenant-B
service.create_task(
{"task_id": "task-r2-x", "metadata": {"tenant_id": "tenant-A"}}
)
result = service.poll(WORKER, IDENTITY)
assert result["status"] == "idle"
assert result["reason"] == "no_task"
audit = store.list_tenant_skip_audit("task-r2-x")
assert any(
a["worker_id"] == WORKER
and a["finding"]["source"] == "tenant"
and a["finding"]["kind"] == "cross_tenant_skip"
and a["finding"]["decision"] == "skip+audit"
and a["finding"]["task_tenant"] == "tenant-A"
and a["finding"]["worker_tenant"] == "tenant-B"
for a in audit
)
def test_a23_enumeration_prefers_same_tenant_over_cross_tenant():
# G4: when a same-tenant task exists, enumeration picks it and skips the
# cross-tenant task (with audit), never selecting cross-tenant work.
store, service = make_service()
register_worker(service) # tenant-A worker (default trust_zone from REG is tenant-A)
service.create_task(
{"task_id": "task-r2-xcross", "metadata": {"tenant_id": "tenant-B"}}
)
service.create_task(
{"task_id": "task-r2-own", "metadata": {"tenant_id": "tenant-A"}}
)
result = service.poll(WORKER, IDENTITY)
assert result["task"]["task_id"] == "task-r2-own"
cross_audit = store.list_tenant_skip_audit("task-r2-xcross")
assert any(a["finding"]["decision"] == "skip+audit" for a in cross_audit)
def test_a23_direct_cross_tenant_claim_fail_closed():
# G4: a DIRECT claim (by task_id) of a cross-tenant task remains FAIL-CLOSED
# (403 cross_tenant_denied) -- in contrast to enumeration which skips+audits.
_, service = make_service()
register_worker(service, trust_zone="tenant-B") # worker in tenant-B
task = service.create_task(
{"task_id": "task-r2-x", "metadata": {"tenant_id": "tenant-A"}}
)
with pytest.raises(ForbiddenError) as exc:
service.governed_claim(task["task_id"], {"worker_id": WORKER}, IDENTITY)
assert exc.value.error_code == "cross_tenant_denied"
assert exc.value.status_code == 403
def test_a23_capacity_exhausted():
store, service = make_service()
register_worker(service) # max_concurrency=1
task_a = service.create_task(
{"task_id": "task-r2-cap-a", "metadata": {"tenant_id": "tenant-A"}}
)
# Worker holds one active (non-expired) claim already.
service.claim(
task_a["task_id"],
{"worker_id": WORKER, "actor_identity": IDENTITY, "lease_seconds": 60},
)
service.create_task(
{"task_id": "task-r2-cap-b", "metadata": {"tenant_id": "tenant-A"}}
)
with pytest.raises(LockedError) as exc:
service.poll(WORKER, IDENTITY)
assert exc.value.error_code == "capacity_exhausted"
assert exc.value.status_code == 423
assert store.get_active_claim_count(WORKER) == 1
def test_a23_generation_stale():
store, service = make_service()
register_worker(service)
task = service.create_task(
{"task_id": "task-r2-gen", "metadata": {"tenant_id": "tenant-A"}}
)
worker = store.get_worker(WORKER)
fresh_task = store.get_task(task["task_id"])
proof = ProofBuilder(store).build(fresh_task, worker)
# Task revision advances after the proof was built -> stale generation.
updated = store.update_task_status(task["task_id"], fresh_task.revision, "QUEUED")
assert updated.revision == fresh_task.revision + 1
engine = RecheckEngine(store)
decision = engine.authoritative_claim_recheck(
proof=proof, task=store.get_task(task["task_id"]), worker=worker
)
assert decision.allowed is False
matching = [c for c in decision.checks if c.check == "generation_ok"]
assert matching and matching[0].ok is False
assert decision.error_code == "generation_stale"
def test_a23_lease_valid_mapping_rejects_overlap():
# lease_valid (already actively claimed / closed lease) maps to
# lease_expired_or_closed -- verified deterministically via to_error.
_, service = make_service()
decision = _make_decision("lease_valid")
error = RecheckEngine(service.store).to_error(decision)
assert error.error_code == "lease_expired_or_closed"
assert error.status_code == 423
def test_a23_reclaim_flight_rejected():
_, service = make_service()
register_worker(service)
task = service.create_task(
{"task_id": "task-r2-lease", "metadata": {"tenant_id": "tenant-A"}}
)
service.claim(
task["task_id"],
{"worker_id": WORKER, "actor_identity": IDENTITY, "lease_seconds": 60},
)
# Re-claiming an in-flight (CLAIMED) task is rejected by the governed path.
with pytest.raises(ConflictError) as exc:
service.scheduling.claim_governed(WORKER, task["task_id"])
assert exc.value.error_code == "task_binding_mismatch"
# -- A23 error mapping (deterministic) ---------------------------------------
def _make_decision(failed_check: str) -> RecheckDecision:
all_checks = [
"tenant_match",
"principal_match",
"authorization_ok",
"task_match",
"generation_ok",
"lease_valid",
"capacity_ok",
"worker_online",
"capabilities_match",
]
checks = [Check(c, c != failed_check, "detail") for c in all_checks]
return RecheckDecision(
recheck_id="rc-x",
proof_id="proof-x",
task_id="task-x",
worker_id=WORKER,
claim_id=None,
allowed=False,
reason=next((c.detail for c in checks if not c.ok), None),
checks=checks,
observed_at=123.0,
)
@pytest.mark.parametrize(
"failed_check,error_code,status",
[
("tenant_match", "cross_tenant_denied", 403),
("principal_match", "task_binding_mismatch", 409),
("task_match", "task_binding_mismatch", 409),
("generation_ok", "generation_stale", 409),
("lease_valid", "lease_expired_or_closed", 423),
("capacity_ok", "capacity_exhausted", 423),
],
)
def test_a23_error_mapping(failed_check, error_code, status):
_, service = make_service()
decision = _make_decision(failed_check)
error = RecheckEngine(service.store).to_error(decision)
assert error.error_code == error_code
assert error.status_code == status
# -- long_poll / poll / schedule semantics -----------------------------------
def test_long_poll_times_out_to_idle():
store, service = make_service()
register_worker(service)
scheduling = SchedulingService(
store, long_poll_seconds=15, long_poll_max=30, poll_interval_seconds=0.05
)
result = scheduling.long_poll(WORKER, timeout_seconds=1)
assert result == {"status": "idle", "reason": "timeout"}
def test_long_poll_delivers_when_task_appears():
store, service = make_service()
register_worker(service)
scheduling = SchedulingService(
store, poll_interval_seconds=0.05
)
# Seed a task after the poll loop has started, out of a timer thread.
service.create_task(
{"task_id": "task-r2-lp", "metadata": {"tenant_id": "tenant-A"}}
)
result = scheduling.long_poll(WORKER, timeout_seconds=3)
assert result["task"]["task_id"] == "task-r2-lp"
assert result["claim_id"]
def test_schedule_endpoint_report():
_, service = make_service()
register_worker(service)
report = service.schedule(WORKER, IDENTITY)
assert report["online"] is True
assert report["capacity_ok"] is True
assert report["queued_count"] == 0
def test_poll_idle_when_no_task():
_, service = make_service()
register_worker(service)
result = service.poll(WORKER, IDENTITY)
assert result["status"] == "idle"
# -- receipt readback integrity ----------------------------------------------
def test_receipt_tamper_detected():
_, service = make_service()
register_worker(service)
service.create_task({"task_id": "task-r2-tamper", "metadata": {"tenant_id": "tenant-A"}})
result = service.poll(WORKER, IDENTITY)
receipt = service.get_receipt(result["receipt_id"], IDENTITY)
# Mutating payload breaks the digest.
mutated = dict(receipt)
mutated["payload"] = dict(receipt["payload"])
mutated["payload"]["attempt"] = 999
assert receipt_digest(mutated["payload"]) != mutated["digest"]
# -- HTTP handler routes -----------------------------------------------------
@pytest.fixture
def http_env():
store, service = make_service()
handler = type("H", (TaskPlaneRequestHandler,), {})
handler.service = service
def authenticator(token):
return IDENTITY if token == "secret-token" else None
handler.authenticator = authenticator
server = ThreadingHTTPServer(("127.0.0.1", 0), handler)
thread = threading.Thread(target=server.serve_forever, daemon=True)
thread.start()
base = f"http://127.0.0.1:{server.server_address[1]}"
yield service, base
server.shutdown()
server.server_close()
def _http(method, url, token, body=None):
request = urllib.request.Request(url, method=method)
if token:
request.add_header("Authorization", "Bearer " + token)
data = None
if body is not None:
data = json.dumps(body).encode("utf-8")
request.add_header("Content-Type", "application/json")
request.add_header("Content-Length", str(len(data)))
try:
with urllib.request.urlopen(request, data=data, timeout=10) as response:
return response.status, json.loads(response.read().decode("utf-8"))
except urllib.error.HTTPError as exc:
try:
payload = json.loads(exc.read().decode("utf-8"))
except Exception:
payload = {}
return exc.code, payload
def test_http_claim_governed_and_readback(http_env):
service, base = http_env
service.register_worker(REG)
service.create_task({"task_id": "task-r2-http", "metadata": {"tenant_id": "tenant-A"}})
status, result = _http(
"POST", f"{base}/tasks/task-r2-http/claim-governed", "secret-token",
{"worker_id": WORKER, "lease_seconds": 30},
)
assert status == 200
assert result["task"]["task_id"] == "task-r2-http"
status, proof = _http("GET", f"{base}/proofs/{result['proof_id']}", "secret-token")
assert status == 200
assert proof["status"] == "CLAIMED"
status, receipt = _http("GET", f"{base}/receipts/{result['receipt_id']}", "secret-token")
assert status == 200
assert receipt_digest(receipt["payload"]) == receipt["digest"]
status, receipts = _http("GET", f"{base}/tasks/task-r2-http/receipts", "secret-token")
assert status == 200
assert len(receipts) == 2
def test_http_poll_and_no_auth_rejected(http_env):
service, base = http_env
service.register_worker(REG)
service.create_task({"task_id": "task-r2-pollhttp", "metadata": {"tenant_id": "tenant-A"}})
status, result = _http("GET", f"{base}/workers/{WORKER}/poll", "secret-token")
assert status == 200
assert result["claim_id"]
assert result["proof_id"]
# No bearer -> 401 across governed endpoints.
status, body = _http("GET", f"{base}/workers/{WORKER}/poll", None)
assert status == 401
def test_http_transport_register_poll_chain(http_env):
service, base = http_env
service.register_worker(REG)
service.create_task({"task_id": "task-r2-tr", "metadata": {"tenant_id": "tenant-A"}})
transport = HttpRunnerTransport(
base, token="secret-token", actor_identity=IDENTITY
)
reg = transport.register_worker("r2-worker-b", ["code.read"])
assert reg["worker_id"] == "r2-worker-b"
# The primary registered worker can long_poll and obtain the claim.
claimed = transport.long_poll(WORKER, 1)
assert claimed and claimed["claim_id"]
assert transport.claim(WORKER, claimed)["claim_id"] == claimed["claim_id"]
# End-to-end renew (REQ-03b) succeeds with the claim_id from the pull.
renewed = transport.renew(WORKER, "task-r2-tr", claim_id=claimed["claim_id"])
assert renewed["claim_id"] == claimed["claim_id"]
assert renewed["lease_until"]
# renew without an explicit claim_id self-resolves the worker's active claim.
renewed2 = transport.renew(WORKER, "task-r2-tr")
assert renewed2["claim_id"] == claimed["claim_id"]
assert renewed2["lease_until"] > renewed["lease_until"]
def test_agent_pull_runner_driven_renew_chain(http_env):
service, base = http_env
service.register_worker(REG)
service.create_task(
{"task_id": "task-r2-runner", "metadata": {"tenant_id": "tenant-A"}}
)
transport = HttpRunnerTransport(
base, token="secret-token", actor_identity=IDENTITY
)
runner = AgentPullRunner(
RunnerConfig(worker_id=WORKER, long_poll_seconds=1), transport
)
claimed = runner.discover_once()
assert claimed and claimed["claim_id"]
# renew pulls claim_id from the runner's inflight state (no 400).
renewed = runner.renew("task-r2-runner")
assert renewed["claim_id"] == claimed["claim_id"]
assert renewed["lease_until"]
def test_agent_pull_runner_discover_idle_returns_none_without_crash():
"""REQ-03c: an idle/no-task response must not crash discover_once."""
class IdleTransport:
def long_poll(self, worker_id, timeout_seconds):
return {"status": "idle", "reason": "timeout"}
def poll(self, worker_id):
return {"status": "idle", "reason": "no_task"}
def claim(self, worker_id, task):
return task
runner = AgentPullRunner(
RunnerConfig(worker_id=WORKER, long_poll_seconds=1), IdleTransport()
)
# A transport returning idle mappings (no task_id) yields None, no exception.
assert runner.discover_once() is None
assert runner.discover_once() is None
+5 -5
View File
@@ -4,10 +4,10 @@ from unittest import mock
import pytest
from task_plane.api import TaskPlaneService
from task_plane.errors import ConflictError, LockedError
from task_plane.policy import ALL_ACTION_CAPABILITIES, PolicyEngine
from task_plane.store import TaskPlaneStore
from app.task_plane.api import TaskPlaneService
from app.task_plane.errors import ConflictError, LockedError
from app.task_plane.policy import ALL_ACTION_CAPABILITIES, PolicyEngine
from app.task_plane.store import TaskPlaneStore
IDENTITY = "spiffe://aios/agent/heartbeat"
@@ -66,7 +66,7 @@ def test_lease_at_current_time_is_expired_and_cannot_be_renewed():
store, task, claim = claimed_task()
boundary = store.get_claim(claim["claim_id"]).lease_until
with mock.patch("task_plane.store._now", return_value=boundary):
with mock.patch("app.task_plane.store._now", return_value=boundary):
with pytest.raises(LockedError, match="lease expired"):
store.heartbeat_claim(task["task_id"], claim["claim_id"], 30)
+2
View File
@@ -568,6 +568,8 @@ def test_subprocess_crash_after_durable_sink_success_recovers_without_repeat(tmp
environment["PYTHONDONTWRITEBYTECODE"] = "1"
environment["ACTION"] = "execute"
environment["CRASH_AFTER_EFFECT"] = "1"
_repo_root = Path(__file__).resolve().parents[2]
environment["PYTHONPATH"] = str(_repo_root / "active-runtime" / "app") + os.pathsep + environment.get("PYTHONPATH", "")
crashed = subprocess.run([sys.executable, "-c", script], env=environment, check=False)
assert crashed.returncode == 77
assert json.loads(sink.read_text()) == {"count": 1, "oid": NEW}
+1 -1
View File
@@ -11,7 +11,7 @@ SERVICE_PACKAGE_ROOT = REPOSITORY_ROOT / "services" / "aios-autonomous-release-c
ACTIVE_RUNTIME_ROOT = REPOSITORY_ROOT / "active-runtime"
ACTIVE_RUNTIME_APP = ACTIVE_RUNTIME_ROOT / "app"
for path in (TEST_ROOT, SERVICE_PACKAGE_ROOT, ACTIVE_RUNTIME_APP):
for path in (ACTIVE_RUNTIME_ROOT, TEST_ROOT, SERVICE_PACKAGE_ROOT, ACTIVE_RUNTIME_APP):
value = str(path)
if value not in sys.path:
sys.path.insert(0, value)
+202
View File
@@ -0,0 +1,202 @@
from __future__ import annotations
from pathlib import Path
import importlib.util
import sys
def load(path: Path, name: str):
spec = importlib.util.spec_from_file_location(name, path)
mod = importlib.util.module_from_spec(spec)
sys.modules[name] = mod
spec.loader.exec_module(mod)
return mod
ROOT = Path(__file__).resolve().parents[2]
MOD = load(ROOT / "scripts/platform/aios_caller_scan.py", "aios_caller_scan")
def write(repo, rel, text):
p = repo / rel
p.parent.mkdir(parents=True, exist_ok=True)
p.write_text(text, encoding="utf-8")
return p
def complete_negative_profile():
return {
"runtime_hosts_complete": True,
"runtime_hosts": ["h1"],
"loaded_bytes_scanned": True,
"systemd_scanned": True,
"api_routes_scanned": True,
"scheduler_scanned": True,
"observation_window_complete": True,
"observation_window_seconds": 3600,
"rollback_ref": "runbook:rollback-1",
"independent_acceptance": "GO",
"probe_failures": [],
}
def bound_runtime_item(subject, *, repo, identity_key, identity_value):
return {
identity_key: identity_value,
"source_path": subject.relative_to(repo).as_posix(),
"source_sha256": MOD._sha(subject),
"active": True,
"evidence_ref": "ev:runtime",
}
def test_static_production_import_is_source_evidence_not_runtime_live(tmp_path):
write(tmp_path, "pkg/target.py", "VALUE=1\n")
write(tmp_path, "pkg/caller.py", "from pkg import target\nprint(target.VALUE)\n")
result = MOD.scan(
repo=tmp_path,
subject=Path("pkg/target.py"),
runtime_profile={"runtime_hosts_complete": False},
)
assert result.assessment == "UNKNOWN"
assert any(c.kind == "STATIC_PRODUCTION" for c in result.callers)
def test_relative_import_is_resolved(tmp_path):
write(tmp_path, "app/__init__.py", "")
write(tmp_path, "app/foo.py", "VALUE=1\n")
write(tmp_path, "app/consumer.py", "from . import foo\nprint(foo.VALUE)\n")
result = MOD.scan(
repo=tmp_path,
subject=Path("app/foo.py"),
runtime_profile={"runtime_hosts_complete": False},
)
assert result.assessment == "UNKNOWN"
assert any(c.identity == "app/consumer.py" for c in result.callers)
def test_active_runtime_import_root_is_supported(tmp_path):
write(tmp_path, "active-runtime/app/foo.py", "VALUE=1\n")
write(tmp_path, "active-runtime/app/consumer.py", "import foo\nprint(foo.VALUE)\n")
result = MOD.scan(
repo=tmp_path,
subject=Path("active-runtime/app/foo.py"),
runtime_profile={"runtime_hosts_complete": False},
)
assert any(c.identity == "active-runtime/app/consumer.py" for c in result.callers)
def test_test_only_caller_does_not_make_live(tmp_path):
write(tmp_path, "pkg/target.py", "VALUE=1\n")
write(tmp_path, "tests/test_target.py", "import pkg.target\n")
result = MOD.scan(
repo=tmp_path,
subject=Path("pkg/target.py"),
runtime_profile={"runtime_hosts_complete": False},
)
assert result.caller_count == 0
assert result.test_only_callers
assert result.assessment == "LIKELY_ORPHAN"
def test_unknown_dynamic_reference_never_degrades_to_orphan(tmp_path):
write(tmp_path, "pkg/target.py", "VALUE=1\n")
write(
tmp_path,
"pkg/dynamic.py",
"import importlib\nname=input()\nimportlib.import_module(name)\n",
)
result = MOD.scan(
repo=tmp_path,
subject=Path("pkg/target.py"),
runtime_profile=complete_negative_profile(),
)
assert result.unknown_dynamic_refs
assert result.assessment == "UNKNOWN"
def test_runtime_hosts_complete_alone_cannot_prove_orphan(tmp_path):
write(tmp_path, "foo.py", "x=1\n")
result = MOD.scan(
repo=tmp_path,
subject=Path("foo.py"),
runtime_profile={"runtime_hosts_complete": True},
)
assert result.assessment == "LIKELY_ORPHAN"
assert result.coverage["negative_evidence_complete"] is False
def test_empty_runtime_host_set_cannot_prove_orphan(tmp_path):
write(tmp_path, "foo.py", "x=1\n")
profile = complete_negative_profile()
profile["runtime_hosts"] = []
result = MOD.scan(repo=tmp_path, subject=Path("foo.py"), runtime_profile=profile)
assert result.assessment == "LIKELY_ORPHAN"
def test_probe_failure_is_unknown(tmp_path):
write(tmp_path, "foo.py", "x=1\n")
profile = complete_negative_profile()
profile["probe_failures"] = ["host:h1 timeout"]
result = MOD.scan(repo=tmp_path, subject=Path("foo.py"), runtime_profile=profile)
assert result.assessment == "UNKNOWN"
def test_complete_negative_evidence_proves_orphan(tmp_path):
write(tmp_path, "pkg/target.py", "VALUE=1\n")
result = MOD.scan(
repo=tmp_path,
subject=Path("pkg/target.py"),
runtime_profile=complete_negative_profile(),
)
assert result.assessment == "PROVEN_ORPHAN"
assert result.coverage["negative_evidence_complete"] is True
def test_runtime_loaded_requires_exact_path_sha_and_active(tmp_path):
subject = write(tmp_path, "pkg/target.py", "VALUE=1\n")
digest = MOD._sha(subject)
good = {
"source_path": "pkg/target.py",
"source_sha256": digest,
"active": True,
"evidence_ref": "ev:loaded",
}
result = MOD.scan(
repo=tmp_path,
subject=Path("pkg/target.py"),
runtime_profile={"runtime_loaded": [good]},
)
assert result.assessment == "LIVE"
assert any(c.kind == "RUNTIME_LOADED" for c in result.callers)
wrong_path = dict(good, source_path="other/target.py")
result2 = MOD.scan(
repo=tmp_path,
subject=Path("pkg/target.py"),
runtime_profile={"runtime_loaded": [wrong_path]},
)
assert result2.assessment != "LIVE"
def test_systemd_api_scheduler_require_bound_active_artifact(tmp_path):
subject = write(tmp_path, "pkg/target.py", "VALUE=1\n")
profile = {
"systemd": [bound_runtime_item(subject, repo=tmp_path, identity_key="unit", identity_value="svc.service")],
"api_routes": [bound_runtime_item(subject, repo=tmp_path, identity_key="route", identity_value="/x")],
"scheduler": [bound_runtime_item(subject, repo=tmp_path, identity_key="identity", identity_value="sched")],
}
result = MOD.scan(repo=tmp_path, subject=Path("pkg/target.py"), runtime_profile=profile)
assert result.assessment == "LIVE"
assert {"SYSTEMD", "API_ROUTE", "SCHEDULER"} <= {c.kind for c in result.callers}
def test_scanner_does_not_import_business_subject(tmp_path):
marker = tmp_path / "EXECUTED"
subject = write(tmp_path, "pkg/target.py", f"open({str(marker)!r},'w').write('bad')\n")
result = MOD.scan(
repo=tmp_path,
subject=Path("pkg/target.py"),
runtime_profile={"runtime_hosts_complete": False},
)
assert not marker.exists()
assert result.subject["sha256"] == MOD._sha(subject)
@@ -0,0 +1,30 @@
from __future__ import annotations
from pathlib import Path
import importlib.util
import sys
ROOT=Path(__file__).resolve().parents[2]
spec=importlib.util.spec_from_file_location("envprof",ROOT/"scripts/platform/aios_environment_profile.py")
M=importlib.util.module_from_spec(spec); sys.modules[spec.name]=M; spec.loader.exec_module(M)
def test_environment_profile_schema_is_secret_safe_and_structured():
p=M.collect("test-host")
d=p.to_dict()
assert d["schema_version"]=="aios.environment-profile.v2"
assert d["host_id"]=="test-host"
assert set(d["capabilities"]) >= {
"posix","systemd","systemd_load_credential","sudo_noninteractive","git","python","node"
}
assert set(d["filesystem"]) == {"case_sensitive","line_ending_default"}
assert d["completeness"]=="CAPABILITY_DISCOVERY_ONLY"
assert d["g0_exit_eligible"] is False
assert all("token" not in x.lower() and "secret=" not in x.lower() for x in d["credential_mechanisms"])
def test_missing_requirements_are_computed(monkeypatch):
monkeypatch.setattr(M, "_systemd_running", lambda: False)
p=M.collect("h", requirements=("systemd","git"))
d=p.to_dict()
assert "systemd" in d["missing_requirements"]
assert d["completeness"]=="REQUIREMENTS_EVALUATED"

Some files were not shown because too many files have changed in this diff Show More