| name | skill-maintainer |
| description | Use this skill when creating, reviewing, restoring, or changing repository skills. It keeps skill behavior self-contained, validates manifests, and coordinates required project documentation updates for skill-system changes. |
Skill Maintainer
This skill maintains the repository's skill system.
Scope
Use it when:
- adding a new
skills/<name>/ directory
- changing how skills are discovered, invoked, or validated
- restoring skills after backup loss
- moving behavior between project docs and skill-specific docs
- preparing a public template that should keep only generic skills
Skill Structure
Every skill must have:
SKILL.md
- YAML-style frontmatter with
name and description
- a concise scope section explaining when to use the skill
Optional files:
scripts/ for deterministic commands the agent should run instead of retyping large logic
references/ for supporting docs that are too detailed for the main skill file
tests/ when scripts or parsing logic have behavior worth locking down
Maintenance Rules
- Keep skill-specific instructions in that skill's
SKILL.md.
- Keep project-wide conventions in
AGENTS.md or docs/.
- Do not copy private project instructions into generic skills.
- If a change affects task lifecycle, task artifacts, skill discovery, skill execution, agent orchestration, restore behavior, or resume behavior, update project docs in the same source change.
- If a skill writes durable data, document whether that data belongs in
tasks/, data/, or runtime state such as .state/.
Verification
After changing skills, run:
.venv/bin/python skills/repo-health/scripts/check_repo_health.py
Run skill-specific tests when present, for example:
PYTHONPATH=skills/task-runner/scripts .venv/bin/python -m pytest skills/task-runner/tests
Publishing A Generic Template
When preparing a reusable public copy:
- keep core skills that implement task lifecycle and delegation
- remove or clearly separate personal integrations
- do not publish local
tasks/, data/, .state/, logs, credentials, or private runtime config
- include sanitized examples only when they teach the generic workflow