skillkit
Powered by skills.sh

Know which skills
actually matter

Local-first analytics for your AI agent skills. See what you use, what wastes context, and what to drop.

$npx @crafter/skillkit
View on GitHub
skillkitnormal
-- NORMAL --skillkit v0.1.4utf-8

Works with the tools you already use

Claude CodeCursorCodexWindsurfGemini CLIClineRoo CodeGitHub CopilotOpenHandsGoose+30 more

How it works

01

Scan

Already have skills installed? One command picks them all up.

$skillkit scan
Scanning ~/.claude/skills/ ...
Scanning .claude/skills/ ...
Found 12 skills (8 via skills.sh, 4 manual)
Indexed 211 sessions · 1,847 invocations
Ready. Run skillkit stats to see usage.
02

Measure

See which skills you actually use. Real data from your sessions.

$skillkit stats --top 5
SKILL30dTRENDcommit42▂▃▅▇█▆▅▇█review38▁▃▅▆▇▇▆▅▃deploy27▁▁▂▃▅▇█▇▅lint3▁▁▁▁▁▁▁▁▂scaffold0░░░░░░░░░░
5 skills · 127 total invocations · 30d window
03

Optimize

Drop what you don't use. Reclaim context budget.

$skillkit health
[████████░░] 78% metadata budget (12.5K / 16.0K)
! scaffold — 0 uses in 30d (0.9K wasted)
! lint — 3 uses in 30d (2.1K, low ROI)
$skillkit prune
Removed scaffold (0.9K)
Removed lint (2.1K)
Reclaimed 3.0K · 10 skills active · 9.5K / 16.0K

The observability layer for AI skills

Usage Sparklines

30-day invocation trends for every skill. See at a glance which skills earn their context budget.

Health Checks

Flag unused skills, detect context bloat, and get actionable recommendations to reclaim tokens.

Context Budget

Track the token cost of each installed skill. Know exactly how much of your context window skills consume.

Session Scanning

Extracts real invocation data from JSONL session files. Local SQLite database, zero telemetry, your data stays on your machine.

Distribution + Observability

skills.sh installs your skills. SkillKit tells you which ones matter.

Comparison of SkillKit, skills.sh, and manual skill management
FeatureSkillKitskills.shManual Copy
Install / update skills
Registry search
Usage analytics
Context budget tracking
Health checks
Unused skill pruning
Session JSONL scanning
Local-first (no telemetry)
MIT License

Built in the open

SkillKit is free and open source. All analytics run locally on your machine. Contribute, fork, or audit the code.

View on GitHub

Frequently asked questions

How is this different from skills.sh?

skills.sh handles skill distribution — install, update, search. SkillKit is the analytics layer on top: local-first usage tracking, health checks, context budget analysis, and pruning that skills.sh doesn't provide.

Which AI coding agents are supported?

Claude Code, Cursor, Codex, VS Code (via extensions), Windsurf, and Gemini CLI. SkillKit scans session JSONL files to extract skill invocations, so any agent that logs tool use is supported.

Is SkillKit free?

Yes. SkillKit is MIT-licensed and free forever. All analytics are local-first — your data never leaves your machine.

What data does SkillKit collect?

None. All analytics are stored locally in a SQLite database at ~/.skillkit/analytics.db. SkillKit scans your local session files and never phones home.

How does the session scanning work?

Run `skillkit scan` to discover installed skills and scan ~/.claude/projects/ for JSONL session files. It extracts Skill tool_use blocks and populates a local analytics database with invocation counts, timestamps, and patterns.

Can I use SkillKit without skills.sh?

Yes. SkillKit is purely an analytics tool — scan, list, stats, and health all work independently. Use skills.sh (npx skills add) to install and manage skills, then use SkillKit to measure and optimize.

Stop guessing. Start measuring.

One command. No signup. No telemetry.