super-skill

Use when the user wants to manage, version, audit, or roll back their Agent Skills, or to review skill candidates mined from past sessions. super-skill is a git-backed package manager for the skills in ~/.agents/skills (and ~/.claude/skills) — it seeds existing skills under version control, explains where each came from, rolls any skill back one command, checks registry integrity, and surfaces recurring task families worth turning into a skill. Triggers include "version my skills", "roll back that skill", "why do I have this skill", "check my skill registry", "what could become a skill".

Invoke when a solo developer asks an agent to cut a release of a library or CLI — bumping the version, tagging, pushing, publishing to a registry (npm or PyPI), and creating a GitHub release. Trigger phrases include "ship it", "cut a release", "publish to npm/PyPI", "tag and release", "bump the version", or "make a GitHub release". Use it whenever the task is turning committed code into a published, versioned artifact. Do NOT invoke for routine commits, opening a normal PR, CI config changes, deploying a running service (that is deployment, not package release), or pre-release exploratory coding. If no publishable package manifest (package.json, pyproject.toml, setup.py) exists, stop and say so instead of guessing.

Use when the user wants to manage, version, audit, or roll back their Claude Code Agent Skills, or to review skill candidates mined from past sessions. super-skill is a git-backed package manager for the skills in ~/.claude/skills — it seeds existing skills under version control, explains where each came from, rolls any skill back one command, checks registry integrity (doctor), and surfaces recurring task families worth turning into a skill. Triggers include "version my skills", "roll back that skill", "why do I have this skill", "check my skill registry", "what could become a skill", "mine my sessions".

Fix a failing function by reproducing the failure, localizing the smallest wrong expression, making the minimal change, and re-verifying — for small bug-fix tasks with a known expected behaviour.