Tools

Goal

Use this project-maintainer skill to keep skills aligned, healthy, and releasable. Primary maintainer outcome: improve local skill markdown quality (SKILL.md and references/*.md) through structured analysis and actionable optimization proposals. This skill orchestrates maintenance workflows; it does not replace domain skills.

User-facing Capability Summary

If the user asks what this skill can do, answer with these three capability groups:

1. Maintain local skills:
   • Keep skill docs and metadata aligned (SKILL.md, agents/openai.yaml, README.md, AGENTS.md).
   • Run structure/consistency checks and flag issues.
   • Propose meaningful optimization/refactor updates and report PASS/FAIL.
2. Benchmark local skills against upstream ecosystems (openai/skills and anthropics/skills) and propose meaningful structure improvements.
   • Download/update both upstream repos into .cache/upstream-skills/.
   • Study upstream SKILL.md structure patterns (frontmatter, sections, workflow/trigger clarity, and layout).
   • Audit local skills including hidden .agents/skills/*.
   • Generate markdown-focused optimization proposals for local skill docs (SKILL.md, references/*.md, and related maintainer docs) with no auto-applied refactors.
3. Bootstrap new skills:
   • Create skill scaffolds with required files and metadata assets.
   • Start by invoking $skill-creator to initialize the skill template and required metadata.
   • Align README/install prompts and maintainer metadata.
   • Run consistency and sync checks to avoid drift.

Available Tasks (User Menu)

When the user asks what this skill can do, offer this task list:

1. sync metadata/docs
   • Align SKILL.md, agents/openai.yaml, README.md, and AGENTS.md.
2. audit consistency
   • Run structure/rules checks across skills and report findings by severity.
3. bootstrap skill
   • Use $skill-creator to initialize the new skill scaffold, then add required files and run metadata/docs checks.
4. refresh postgres references
   • Execute Postgres best-practices refresh workflow defined in this skill references.
5. benchmark against upstream
   • Download/update openai/skills and anthropics/skills, study SKILL.md patterns, compare local skills, and propose markdown optimization updates (no auto-apply).
   • After artifacts are generated, review local skills one by one and state whether each skill needs changes (CHANGE) or not (NOOP), with concrete proposals when needed.

Trigger rules

Use this skill when users ask to:

• Maintain or clean up one or more skills
• Optimize one or more skills (quality, consistency, or maintainability)
• Refactor skill structure or instructions while preserving intent
• Bootstrap a new skill (reusable or maintainer)
• Benchmark local skills against upstream skill ecosystems (for example openai/skills and anthropics/skills)
• Sync SKILL.md, agents/openai.yaml, and repository docs
• Run a maintenance pass before release
• Refresh Postgres best-practices references

Workflow

1. Route the request with references/maintenance-router.md.
2. For metadata/docs alignment, follow references/metadata-sync.md.
3. For repository-wide structure and rules checks, follow references/doc-consistency.md.
4. For skill bootstrap, follow references/skill_openai_metadata.md then references/metadata-sync.md.
5. For upstream benchmarking and structure proposals, follow references/openai-skill-benchmark.md (clone/update upstream repos first, then analyze artifacts and propose per-skill updates one by one).
6. For Postgres best-practices refresh, follow references/postgres-refresh.md (self-contained workflow in this skill).
7. Before finishing, run references/release-checklist.md and report pass/fail with actionable findings.

Guardrails

• Keep this skill orchestration-only in v1.
• Prefer repeatable commands and documented checks in this skill before inventing ad-hoc flows.
• Do not depend on markdown guidance outside this skill's references/ folder.
• Keep changes scoped to requested maintenance outcomes.
• Keep task boundaries explicit: run only the requested task (sync, audit, bootstrap, refresh, benchmark) unless the user requests a mixed flow.
• If no meaningful updates are needed, return PASS (NOOP) and avoid persistent file edits.