Implement skill persistence service (codex)

[numnx]

🧩 Task: Implement skill persistence service

Automated task PR opened by Code UX for T02.

📋 Summary

Sprint: Sprint 149 · Self Reflection & Persisten Skills · Task: T02

Add disabled-by-default persistent skill storage for shareable agent self-enhancement and a configurable self-reflection improvement loop for planning and QA agents, with MCP management/search surfaces, runtime prompt injection, dashboard controls, tests, and docs.

📝 Full Task Prompt

## Objective
Build the backend persistence and retrieval layer for persistent skills, including markdown import/export and semantic search, using the contracts from T01.

## Scope
- `src/repositories/skill-repository.ts` (new)
- `src/services/skill-service.ts` (new)
- `src/services/skill-markdown-parser.ts` (new)
- `src/services/embedding-service.ts` and `src/services/memory-service.ts` only as reference or minimal shared helper extraction if needed
- `src/contracts/skill-types.ts` (new, if T01 did not create a dedicated file)
- `tests/backend/repositories/skill-repository.test.ts` (new)
- `tests/backend/services/skill-service.test.ts` (new)
- `docs/dashboard/memory.md`
- `docs/architecture/agent-preset-foundation.md`

## Implementation Requirements
1. Implement `SkillRepository` methods for creating/listing/updating/deleting skill storages, attaching/detaching storages to agent presets, and CRUD for skills within a storage with project ownership checks.
2. Implement `SkillService` methods for writing a skill from markdown/frontmatter, rendering a skill back to markdown, listing by storage or agent, and searching skill content with bounded vector ranking similar to `MemoryService.search`.
3. Store embeddings in `skill_embeddings` with model and dimension metadata; skip or mark unembedded skills when no embedding provider is available rather than failing ordinary CRUD.
4. Parse skill markdown with YAML-like frontmatter fields such as `title`, `description`, `tags`, `appliesTo`, and `version`, while preserving the body as the authoritative skill instruction content.
5. Enforce project boundaries for every storage, skill, attachment, and search operation; deleting a storage must cascade or explicitly remove its skills, embeddings, and agent bindings.
6. Add deterministic tests for CRUD, project isolation, markdown round-trip, embedding dimension filtering, top-K ordering, and agent-attached storage search.
7. Update docs to explain how persistent skills differ from project memory and how vector search ranks skills.

## Constraints
- Depends on T01 contracts and tables; do not redefine incompatible schema names.
- Do not expose MCP tools or dashboard UI in this task.
- Keep vector search bounded and deterministic; avoid full-history dashboard-side reconciliation or unbounded scans.
- Do not write skill files into the project workspace or `.code-ux/` sprint directories.

## Verification
- Run `pnpm run lint`.
- Run `pnpm run test:backend -- tests/backend/repositories/skill-repository.test.ts tests/backend/services/skill-service.test.ts`.
- Success means persistent skill CRUD, markdown parsing, attachment lookup, and vector search work with project isolation and deterministic ordering.

✅ QA Review

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

🤖 Model & Provider

Provider	Codex CLI
Model	gpt-5.5

⏱️ Timing

Started	2026-07-07 01:09:57 UTC
Finished	2026-07-07 01:20:03 UTC
Duration	10m 6s

📊 Token Usage

Metric	Value
Input tokens	222,351
Cached input tokens	2,799,872
Output tokens	23,412
Total tokens	3,045,635
Invocations	1
Tool calls	57
Active time	9m 41s

Included usage estimate (subscription/local login): $3.21

🌿 Branch Info

Base: feature/CODUX-149-self-reflection-persisten-skills
Head: task/feature-codux-149-self-reflection-persisten-skil-t02-codex-mr9yc28l

---

🤖 Generated by Code UX