quality-pass-20260428t215000

Metadata

Description

## Quality Pass: Post-Triage Review

Review and optimize task metadata for the poietic-bugs batch before execution.

## Tasks to review
- design-claim-lifecycle (research/design)
- design-pdf-binary (research/design)
- fix-claim-lifecycle (fix)
- fix-failed-upstream (fix)
- fix-pdf-binary (fix)
- verify-end-to (test)

## What to do

For EACH task above:

### 1. Classify task type
`wg show <task-id>` to read. Classify:
- design tasks → **design**
- fix tasks → **fix**
- verify-end-to → **test**

### 2. Assign agent identity
`wg agency stats --by-task-type` → use recommended role.
JSON: `wg agency stats --by-task-type --json` → `.task_type_breakdown.recommendations[].best_role`
Apply: `wg assign <task-id> <agent-hash>`

### 3. Select model tier
`wg agency stats --by-task-type` → **Best Model by Task Type**.
Heuristics if data is sparse:
| Signal | Model |
|--------|-------|
| design tasks (multi-option reasoning) | opus |
| fix-failed-upstream (well-specified, 1-line conceptually) | sonnet |
| fix-claim-lifecycle (touches dispatcher + multiple commands) | sonnet or opus |
| fix-pdf-binary (parsing + tests) | sonnet |
| verify-end-to (test execution + evidence gathering) | sonnet |

Apply: `wg edit <task-id> --model <tier>`

### 4. Release for execution
`wg resume <task-id>`

## Validation
- [ ] All 6 batch tasks have an agent assigned
- [ ] All 6 have a model set
- [ ] All 6 are un-paused (status: open)
- [ ] Assignments justified by evaluation data, not arbitrary

Depends on

• done .assign-quality-pass-20260428t215000

Required by

• done design-claim-lifecycle
• done design-pdf-binary
• done fix-claim-lifecycle
• done fix-failed-upstream
• done fix-pdf-binary
• done verify-end-to

Log

• 2026-04-28T22:25:28.733309256+00:00 Lightweight assignment: agent=Default Assigner (eea940a6), exec_mode=full, context_scope=task, reason=Default Assigner agent is purpose-built for assignment decisions; this quality-pass task reviews metadata and assigns agents to 6 batch tasks—exactly the Assigner's domain.
• 2026-04-28T22:25:30.011089964+00:00 Spawned by coordinator --executor claude --model opus
• 2026-04-28T22:25:38.151089646+00:00 Starting quality-pass review for 6 batch tasks
• 2026-04-28T22:27:03.512470113+00:00 Plan: assign f5143935 (Careful Programmer, top role per agency stats) to all 6; opus for designs + fix-claim-lifecycle (multi-option/multi-component); sonnet for the narrower fix/test tasks
• 2026-04-28T22:31:07.641070374+00:00 Final assignments: design-claim-lifecycle opus Default Evaluator (31847164) [LLM .assign-* override; design=option-eval is Evaluator's specialty per agency stats] design-pdf-binary opus Default Evaluator (31847164) [LLM .assign-* override; same reasoning] fix-claim-lifecycle opus Careful Programmer (f5143935) [touches dispatcher + multiple commands] fix-failed-upstream sonnet Careful Programmer (f5143935) [well-specified, single concept] fix-pdf-binary sonnet Careful Programmer (f5143935) [parsing + tests] verify-end-to sonnet Careful Programmer (f5143935) [test execution + evidence] All 6 un-paused, status=open, agent assigned, model set.
• 2026-04-28T22:31:14.843311393+00:00 Validation complete: [x] All 6 batch tasks have an agent assigned [x] All 6 have a model set [x] All 6 are un-paused (status=open, paused=null) [x] Assignments justified: my Careful Programmer (role 52335de1, 0.68 avg/131 tasks, recommended by --by-task-type) for fix/verify; LLM .assign-* re-routed designs to Default Evaluator (role 75d2fab8, 0.77 avg/187 tasks) with rationale logged on each design task. Models picked per task heuristic table: opus for designs + multi-component fix-claim-lifecycle, sonnet for narrower work.
• 2026-04-28T22:31:30.836373011+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
• 2026-04-28T22:32:48.858627567+00:00 PendingEval → Done (evaluator passed; downstream unblocks)