fix-claim-lifecycle — Workgraph live mirror

Metadata

Status	done
Assigned	`agent-994`
Agent identity	`f51439356729d112a6c404803d88015d5b44832c6c584c62b96732b63c2b0c7e`
Model	`claude:opus`
Created	2026-04-28T22:24:06.307397955+00:00
Started	2026-04-28T22:45:29.371319831+00:00
Completed	2026-04-28T23:09:41.615301385+00:00
Tags	`bug,fix,claims`, `eval-scheduled`
Eval score	0.76
└ blocking impact	0.78
└ completeness	0.75
└ coordination overhead	0.80
└ correctness	0.78
└ downstream usability	0.75
└ efficiency	0.78
└ intent fidelity	0.85
└ style adherence	0.80

Description

Implement the approach chosen in design-claim-lifecycle. Read that task's log first via wg show design-claim-lifecycle for the chosen approach, file paths, and schema notes.

Bugs being fixed:

bug-reset-leaves-stale-claims.md: wg reset leaves stale claimed_by → dispatcher silently skips
bug-retry-doesnt-clear-stale-downstream-claims.md: wg retry doesn't propagate to downstream → silent stall on fan-out workflows

Also fix the misleading hint in wg service status output (currently says "check agent configuration" when the real cause is stale claims).

Validation

Failing tests written first (TDD), one per bug: - test_wg_reset_clears_stale_claims: tasks reset → claimed_by is null → dispatcher spawns fresh agents - test_wg_retry_clears_downstream_claims_on_dead_agents: retry upstream → transitive downstream claims to dead agents are cleared
(If heartbeat option chosen) test_dispatcher_heartbeat_unclaims_dead_agents: claim with non-running agent_id → dispatcher unclaims at next tick
Repro from bug-reset doc steps 1–6 produces fresh agent spawns (NOT "spawned=0")
Repro from bug-retry doc steps 1–6 produces downstream agent spawn after upstream completes
wg service status hint is updated to mention stale claims as a possibility
cargo build + cargo test pass
Permanent smoke scenarios added under tests/smoke/scenarios/ with owners list including this task id
cargo install --path . was run before claiming done

## Description
Implement the approach chosen in design-claim-lifecycle. Read that task's log first via `wg show design-claim-lifecycle` for the chosen approach, file paths, and schema notes.

Bugs being fixed:
- bug-reset-leaves-stale-claims.md: `wg reset` leaves stale `claimed_by` → dispatcher silently skips
- bug-retry-doesnt-clear-stale-downstream-claims.md: `wg retry` doesn't propagate to downstream → silent stall on fan-out workflows

Also fix the misleading hint in `wg service status` output (currently says "check agent configuration" when the real cause is stale claims).

## Validation
- [ ] Failing tests written first (TDD), one per bug:
      - test_wg_reset_clears_stale_claims: tasks reset → claimed_by is null → dispatcher spawns fresh agents
      - test_wg_retry_clears_downstream_claims_on_dead_agents: retry upstream → transitive downstream claims to dead agents are cleared
- [ ] (If heartbeat option chosen) test_dispatcher_heartbeat_unclaims_dead_agents: claim with non-running agent_id → dispatcher unclaims at next tick
- [ ] Repro from bug-reset doc steps 1–6 produces fresh agent spawns (NOT "spawned=0")
- [ ] Repro from bug-retry doc steps 1–6 produces downstream agent spawn after upstream completes
- [ ] `wg service status` hint is updated to mention stale claims as a possibility
- [ ] cargo build + cargo test pass
- [ ] Permanent smoke scenarios added under tests/smoke/scenarios/ with owners list including this task id
- [ ] cargo install --path . was run before claiming done

Depends on

Required by

done verify-end-to

Log

2026-04-28T22:24:06.293578670+00:00 Task paused
2026-04-28T22:28:28.639765008+00:00 Task resumed
2026-04-28T22:28:53.593263591+00:00 Lightweight assignment: agent=Careful Programmer (f5143935), exec_mode=full, context_scope=task, reason=Careful Programmer is the only agent with developer capabilities; its Careful tradeoff aligns with correctness-critical claim lifecycle fixes that require TDD, exhaustive testing, and binary verification before marking done.
2026-04-28T22:45:29.371324419+00:00 Spawned by coordinator --executor claude --model opus
2026-04-28T22:45:43.369170235+00:00 Starting implementation per design-claim-lifecycle. Reading design + relevant source files.
2026-04-28T23:08:34.450190339+00:00 Implementation done: extracted compute_closure to commands::claim_lifecycle, added eager downstream-claim clearing to wg retry (both Failed/Incomplete and InProgress branches), extended reconcile_orphaned_tasks + find_orphaned_tasks to cover Status::Open with stale claims. 5 new TDD tests added (reset/retry/sweep). 3 smoke scenarios added. All claim_lifecycle/retry/sweep/reset unit tests pass; pre-existing failures in spawn_task / tui / provenance_coverage are unrelated.
2026-04-28T23:09:18.130111273+00:00 Committed: c8044423b — pushed to remote
2026-04-28T23:09:41.615306574+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
2026-04-28T23:12:36.850070140+00:00 PendingEval → Done (evaluator passed; downstream unblocks)