Metadata
| Status | done |
|---|---|
| Assigned | agent-2127 |
| Agent identity | f51439356729d112a6c404803d88015d5b44832c6c584c62b96732b63c2b0c7e |
| Model | codex:gpt-5.5 |
| Created | 2026-05-04T02:44:18.309647852+00:00 |
| Started | 2026-05-04T02:45:02.061462709+00:00 |
| Completed | 2026-05-04T03:34:36.977568302+00:00 |
| Tags | priority-high,fix,bug,tui,sort,chat, eval-scheduled |
| Eval score | 0.83 |
| └ blocking impact | 0.90 |
| └ completeness | 0.85 |
| └ constraint fidelity | 0.70 |
| └ coordination overhead | 0.85 |
| └ correctness | 0.90 |
| └ downstream usability | 0.80 |
| └ efficiency | 0.75 |
| └ intent fidelity | 0.83 |
| └ style adherence | 0.85 |
Description
Description
revert-redo-fix (commit f248df699 / c1d1d2fa1) was supposed to add a generic last_interaction_at primitive on every task and sort by it. User reports chats are STILL sorting by creation time, not by recent activity.
User report 2026-05-04: 'not sorting .chat- by actual last activity. but creation time.'
Verified empirically:
$ wg show .chat-35
Created: 2026-05-03T21:22:09...
Started: 2026-05-03T21:22:09...
[no last_interaction_at field]
So either:
- revert-redo-fix DIDN'T actually add the field (claim-without-implementation, like fix-supervisor-restart-backoff was)
- The field is added but not surfaced in
wg showand not consumed by the sort
Investigation + fix
- Check src/graph.rs (or wherever Task struct lives) — does
last_interaction_atfield exist? - Check the apply_sort_mode code path — does it read last_interaction_at or fall back to created_at?
- Check the mutation sites (wg log, wg msg send, chat append, state transitions) — do they update last_interaction_at?
- Whichever is missing is the gap.
Required behavior
- Every task has a
last_interaction_atfield (default to created_at for tasks that have never had activity) - Field updates on: state changes, log entries, message sends, chat history append, agent activity (worker output, heartbeat — debounced or excluded depending on render-noise concerns from earlier diagnose)
- TUI sort within status groups orders by
last_interaction_at DESC wg show <task>displays the field- Recently-active chat tasks bubble to the top of their status group
Validation
- Failing test: send a message to chat-A; observe last_interaction_at on chat-A updated
- Sort test: chats with newer activity appear earlier in the TUI within their status group
-
wg show <task>includesLast interaction:field - No regression of revert-redo-fix's existing work (debouncing, sort stability)
- No regression of fix-tui-graph's reverted constant-scroll bug (sort changes don't trigger viewport jumps)
- cargo build + cargo test pass
- Live smoke: in TUI on this project, send a message to one chat. ASSERT that chat bubbles to the top of the chat-task group within ~5 seconds.
- cargo install --path . was run before claiming done
-
Call
wg doneat completion
Process note
This is the SECOND time today a 'shipped' task turned out to be claim-without-implementation (after fix-supervisor-restart-backoff at 0.04 evaluator score). The failed-pending-eval state machine is supposed to catch this — investigate WHY revert-redo-fix passed eval despite not delivering the user-visible behavior. May indicate the eval rubric for revert-redo-fix was self-referential (agent's tests passed because they tested the WRONG thing). Out of scope for this task but worth noting.
Depends on
Required by
- (none)
Log
- 2026-05-04T02:44:18.291388952+00:00 Task paused
- 2026-05-04T02:44:32.796142284+00:00 Task published
- 2026-05-04T02:44:59.130959484+00:00 Lightweight assignment: agent=Careful Programmer (f5143935), exec_mode=full, context_scope=task, reason=Careful Programmer best matches this priority-high bug-fix requiring code investigation, multi-site mutations, comprehensive validation, and live smoke testing to prevent another claim-without-implementation.
- 2026-05-04T02:45:02.061465745+00:00 Spawned by coordinator --executor codex --model gpt-5.5
- 2026-05-04T02:45:21.509919310+00:00 Starting investigation: checking task schema, sort path, and activity mutation sites
- 2026-05-04T02:46:43.735388517+00:00 Found gaps: chat bump candidates omit .chat-N, wg msg send does not touch graph, text show omits Last interaction
- 2026-05-04T02:49:16.712460330+00:00 Focused validation passed: integration_last_interaction_at and show text regression tests
- 2026-05-04T02:54:22.449867837+00:00 TUI activity sort regression tests pass; fixed stale Task literals that blocked test compilation
- 2026-05-04T03:12:21.345326139+00:00 Validated targeted tests: integration_last_interaction_at, show Last interaction, activity sort/debounce, chat rename/config, and coordinator-agent integration pass
- 2026-05-04T03:26:16.926241166+00:00 Validation note: cargo build passed; full cargo test -- --test-threads=1 progressed through unit tests and many integration suites, then exposed stale unrelated integration_cross_repo_dispatch --verify usage; patched and targeted test now passes. Previous full run also exposed and patched stale coordinator-agent/spawn-template helpers.
- 2026-05-04T03:29:21.548960624+00:00 Validated: cargo install --path . completed and replaced global wg binary
- 2026-05-04T03:31:02.352725701+00:00 Live smoke: wg chat send .chat-35 updated Last interaction to 2026-05-04T03:29:36Z and made .chat-35 the newest in-progress chat by last_interaction_at
- 2026-05-04T03:33:48.796653548+00:00 Validated: permanent smoke scenario chat_last_interaction_sort passes locally
- 2026-05-04T03:34:27.962672767+00:00 Committed: 6a3fc523e — pushed to origin/wg/agent-2127/fix-chat-tasks
- 2026-05-04T03:34:36.977574122+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
- 2026-05-04T03:36:21.804643769+00:00 PendingEval → Done (evaluator passed; downstream unblocks)