skillscope

Your Claude Code skill bank, with receipts.

You write skills. Claude (hopefully) uses them. Then one day you have 90 of them spread across global folders, projects and plugins, and no idea what is alive, what is dead weight, and what Claude actually triggers.

Skillscope is a local dashboard that answers all of that in one glance.

Why this exists

Skills are quietly becoming the real currency of Claude Code: long-running agents are only as good as the skill bank you feed them. But you get zero visibility in return. You cannot see all your skills in one place, you do not know whether Claude ever fires them, and dead or duplicated skills rot silently in ~/.claude/skills.

If you are the kind of person who has forty of these (you know who you are), this is your control room.

What it does

• Every skill on your machine. Global (~/.claude/skills), per-project (.claude/skills, auto-discovered from your Claude Code history), and plugin skills.
• Real usage, not vibes. Mines your local transcripts with ripgrep: how many times each skill fired, when, in which projects, with a 12-week sparkline. Under 100 ms over gigabytes of history, computed live on every refresh.
• Cleanup advice. Duplicates, never-used trash candidates, dead file references, 8k+ token heavyweights, and skills with no description (the #1 reason Claude never triggers a skill).
• Read, edit, create. Clean markdown viewer for SKILL.md and its references, in-app editor (a .bak is kept on every save), one-click scaffold for a new skill with a template that puts the trigger description first.
• Trash, never rm. Deleting a skill moves it to the system trash. Always recoverable.
• Share on Signal. Optional: with signal-cli linked, send any skill to a contact in two clicks (zipped when it has references).

Run it

git clone https://github.com/kabylesystem/skillscope
cd skillscope
npm install
npm run dev   # http://localhost:4747

Everything runs and stays on your machine: no telemetry, no cloud, nothing leaves, except the skills you explicitly share.

Requirements: Node 20+, ripgrep. Optional: gio (system trash), signal-cli (sharing).

Scan extra folders: SKILLSCOPE_ROOTS="~/dev:~/work" npm run dev

How usage mining works

ripgrep sweeps ~/.claude/projects/**/*.jsonl for Skill tool invocations and slash commands, then aggregates per skill. Last-used dates come from session file mtimes. It is an honest approximation built from data you already have, not telemetry.

License

MIT