(CODUX-69) Capture self-reflection ratings

[numnx]

🧩 Task: Capture self-reflection ratings

Automated task PR opened by Code UX for T01.

📋 Summary

Sprint: Sprint 180 · Self Reflection Rating · Task: T01

Expose each task agent's self-reflection as a persisted 5-star rating, surface the latest rating on Tasks and Live task cards, and show per-section ratings in an accessible hover/focus overlay.

🤖 Model & Provider

Provider	Codex CLI
Model	gpt-5.5

⏱️ Timing

Started	2026-07-07 12:40:08 UTC
Finished	2026-07-07 12:49:15 UTC
Duration	9m 7s

📊 Token Usage

Metric	Value
Input tokens	142,024
Cached input tokens	3,134,464
Output tokens	20,065
Total tokens	3,296,553
Invocations	1
Tool calls	80
Active time	8m 19s

Included usage estimate (subscription/local login): $2.88

✅ QA Review

⏳ QA review runs after this PR is opened. Check the task card in the dashboard for the outcome once it lands.

📝 Full Task Prompt

## Objective
Add a backend capture path for structured self-reflection ratings so each completed task run can persist an overall 0-5 rating plus per-section ratings.

## Scope
- Create `src/contracts/task-self-reflection-types.ts`
- Create `src/repositories/task-self-reflection-rating-repository.ts`
- Edit `src/repositories/db/app-db-schema.ts`
- Edit `src/repositories/db/app-db-migrations.ts`
- Edit `src/services/cli-workflow/pipeline/memory-capture-stage.ts`
- Edit `src/services/cli-workflow/pipeline/pipeline-context.ts`
- Edit `src/app/dependency-factory/core-factory.ts`
- Edit `src/repositories/settings-defaults.ts`
- Add or update tests under `tests/backend/services/cli-workflow/pipeline/memory-capture-stage.test.ts` and `tests/backend/repositories/task-self-reflection-rating-repository.test.ts`

## Implementation Requirements
1. Define `TaskSelfReflectionRating` and `TaskSelfReflectionSectionRating` contracts with `overallRating`, `sections`, `sourceTaskRunId`, `capturedAt`, and normalized section labels/ratings.
2. Add a `task_self_reflection_ratings` SQLite table with project, sprint, task, task run, overall rating, section JSON, timestamps, cascade foreign keys, and indexes for latest-by-task lookup.
3. Implement a repository that upserts one rating per task run and returns the latest rating map for a list of task ids.
4. Extend `.task-learnings.md` capture to parse an optional `## Self Reflection Rating` section, accepting an `Overall: N/5` line and bullet lines such as `- Implementation: 4/5 - note`.
5. Wire the repository through the CLI workflow pipeline context and core dependency factory so task runs capture ratings when memory capture reads `.task-learnings.md`.
6. Extend the default `workerLearningsInstruction` to ask workers for a self-reflection rating section without breaking existing category learning capture.

## Constraints
- Do not require a rating section for successful memory capture; absence of ratings must be a no-op.
- Clamp or reject invalid ratings safely so malformed worker output cannot crash task finalization.
- Keep memory entries and self-reflection ratings as separate persisted concepts; do not store UI-only rating data in `memories`.
- Do not change provider execution behavior or task status transitions.
- Preserve existing `.task-learnings.md` parsing for `## Category:` sections.

## Verification
- Run `pnpm run test:backend -- tests/backend/services/cli-workflow/pipeline/memory-capture-stage.test.ts tests/backend/repositories/task-self-reflection-rating-repository.test.ts`.
- Run `pnpm run lint`.
- Success means valid rating sections create one latest rating per task run, missing or malformed rating sections do not fail memory capture, and existing learning bullets still capture as before.

🌿 Branch Info

Base: feature/CODUX-180-self-reflection-rating
Head: task/feature-codux-180-self-reflection-rating-t01-codex-mramzpd4

---

🤖 Generated by Code UX