This skill turns repeated project judgment into reusable working rules. It is proposal-first and must not write repo files without approval.
Inputs
- Recent raw entries.
- Hard stuff radar findings.
- Lifecycle derivation.
- Artifact index entries.
- Existing
AGENTS.md, skill files, or checklist files when present.
Use the deterministic candidate helper when a vault path and slug are available:
python <this-skill>/scripts/distill_candidates.py --vault <vault> --slug <project-slug>
Output Contract
Separate:
- Facts supported by raw entries.
- Inferred patterns.
- Proposed reusable artifacts.
Possible proposed artifacts:
- AGENTS.md rule additions.
- Checklist items.
- Design review rubric updates.
- Debugging playbook entries.
- Prompt/schema contract reminders.
- Candidate new skill or existing skill update.
Safeguards
- Do not propose a hard rule from a single weak entry.
- Do not propose updating global skills when the pattern is repo-specific.
- Do not propose a broad AGENTS rule when a checklist item is enough.
- Flag privacy-sensitive or project-specific details before copying them into reusable skills.
- Ask for approval before modifying any repo file.