skills-library
August 20, 2026 · View on GitHub
A Claude Code skill that runs a guided interview and recommends the right AI skills for your repo or team.
If it's useful, a star helps others find it.
The skill runs entirely inside Claude — no CLI, no server, no external dependencies. The interview, recommendations, onboarding, and feedback flows are all driven by SKILL.md and Claude Code's native tooling.
Install
Paste this into any Claude Code session:
Install the skills-library skill from https://github.com/lindblomstefan/skills-library and set it up so I can type /skills-library to start it.
Claude will clone the repo, copy .claude/skills/skills-library/ into ~/.claude/skills/, and add the trigger lines to ~/.claude/CLAUDE.md. Nothing else runs.
Use
Type /skills-library in any Claude Code session to start the guided interview. Claude will:
- Check if you're in a git repo and offer to inspect it
- Ask how you want to share context (describe freely, problem only, or chat)
- Ask narrowing questions — work area, stack, constraints, sensitivity
- Recommend up to 5 skills with license state, status, and blockers
Claude will also suggest /skills-library automatically when you ask what AI skills, tools, or AI-powered workflows to use for a project or team.
After using a recommended skill, say "I have feedback on [skill name]." Claude appends a dated entry to a local feedback file at ~/.claude/skills/skills-library/feedback/<skill-id>.md. Once a skill reaches 3 entries it is marked validated — confirmed working in practice, not universally recommended.
To read the feedback for any skill, ask Claude: "Show me the feedback for graphify" — or open the file directly.
Catalog
Over 100 skills tracked across all stages. The skill recommends from the full catalog — status is always shown so you know the confidence level.
See CATALOG.md for the full list. Skills move from candidate → evaluating → approved through real usage and community feedback.
Contributing
This library is still young. The catalog has over 100 skills but very few have real-world feedback yet. That is the gap. If you have feedback of a skill that is not in the library, add it first.
The most valuable thing you can do is use a skill and leave feedback. No GitHub account needed:
- Give feedback on a skill — use a skill, then say "I have feedback on [skill name]" in any Claude Code session. Claude appends a dated entry to a local file. Once a skill reaches 3 real-session entries it is marked validated. This is how the library earns trust.
- Add a skill — type
/skills-libraryand say "I have a skill to add." Only when there is something genuinely missing. - Improve the library — fork, edit, open a PR. Run
npm testbefore pushing.
See CONTRIBUTING.md for the validation model and what makes a good skill addition.
This repository
The installable artifact is the skill folder:
.claude/skills/skills-library/
├── SKILL.md # Interview, recommendation, onboarding, and feedback instructions
├── VERSION # Used for update detection
├── feedback/ # Per-skill feedback files; read before recommending
│ ├── _template.md # Template for new skills
│ ├── gstack.md
│ └── graphify.md
├── agents/ # Claude agent descriptor
└── references/ # Catalog overview for offline use
The rest of the repo is catalog governance tooling — YAML manifests, evaluation runs, and validation scripts. It is not part of the installed skill.
catalog/ # Library skill manifests and taxonomies
evaluations/ # Evaluation runs and evidence
tools/ # Catalog validation and guardrail checks
docs/ # Engineering decisions and guardrails