Files
Assistant/dev/tasks/DEV-000214-agent-execution-recovery.json
m c9eb398857
CI / test (push) Successful in 59s
CI / test (pull_request) Successful in 1m8s
chore(governance): declare canary scope for recovery
2026-08-23 13:12:25 +03:00

24 lines
2.9 KiB
JSON

{
"task_id": "DEV-000214",
"task_revision": 1,
"architecture_revision": "ADR-0018",
"base_repo_revision": "f9b9765100df02bead16280d2db9c1af478a4589",
"risk_tier": "E",
"intent": "Make Agent Execution recovery restart-safe: renew active leases with managed heartbeats, fence stale workers, cap retries, and requeue transient execution failures without losing durable dispatch state.",
"requirement_ids": ["durable heartbeat", "lease fencing", "bounded retry", "restart recovery", "transient failure requeue"],
"allowed_change_scope": {"modules": ["work", "agent-worker"], "paths": ["contracts/work/**", "modules/work/**", "adapters/postgres/work_repository.go", "cmd/agent-worker/**", "cmd/agent-canary/**", "database/migrations/042_agent_execution_recovery.sql", "tests/agent_execution/**", "dev/ACTIVE_TASK.json", "dev/tasks/DEV-000214-agent-execution-recovery.json"]},
"forbidden_change_scope": {"paths": ["architecture/**", "security/**", "deploy/**", "cmd/devctl/**", "modules/identity/**", "modules/capability/**", "LLM-local/**"]},
"allowed_architecture_delta": {"new_cross_module_dependencies": 0, "new_canonical_owners": 0, "new_privileges": 0, "new_third_party_dependencies": 0, "new_database_owner_roles": 1},
"contracts": {"consume": ["existing AgentExecutor", "existing lease/callback contract", "durable dispatch queue"], "may_change": ["WorkerStore retry/recovery port"]},
"semantic_owner": "work owns dispatch, lease, callback and retry lifecycle",
"canonical_state": ["work agent dispatch queue and lease tables remain canonical; heartbeat is durable evidence of liveness"],
"state_transitions": ["CLAIMED -> QUEUED on transient failure or stale recovery", "CLAIMED -> FAILED after retry limit", "ACTIVE lease -> EXPIRED before stale claim requeue", "heartbeat renews lease and callback sequence"],
"side_effects_and_privileges": ["work_role writes only work-owned queue and lease state; no external mutation"],
"persistence_and_migrations": ["042 adds bounded retry limit to dispatch queue"],
"failure_retry_restart": ["stale recovery expires the old lease before requeue", "transient execution failure uses durable delayed retry", "retry limit prevents hot-loop poison dispatch", "heartbeat failure cancels execution and leaves a retryable dispatch"],
"assumptions": ["operational heartbeat ticker is lifecycle-managed by the bounded worker call and durable state remains authoritative"],
"acceptance_scenarios": ["long execution renews lease and emits ordered heartbeats", "stale claim is fenced and requeued", "third failed attempt becomes FAILED", "waiting/error dispatch is not left CLAIMED", "legacy dispatch behavior remains compatible"],
"required_tests": ["go test", "go vet", "preflight", "check-diff", "check-architecture", "independent", "git diff --check", "migration/schema checks", "CI", "production public health check"],
"terminal_gap_states": ["BLOCKED_BY_SPEC_GAP", "BLOCKED_BY_ARCHITECTURE_GAP", "BLOCKED_BY_MISSING_CAPABILITY"]
}