Inventory the roots
Scan shared, Codex, and Claude global skill folders. Capture each path, package source, file size, and instruction content.
Local-first · MIT · $0 forever
Your agent skills keep receipts.
See where every Codex and Claude skill came from, when local session evidence last referenced it, and how much discovery context its name and description may occupy. Archive with a ledger. Restore to the original path.
- Local processing
- Account-free
- Zero telemetry
Start here
One command. A complete local inventory.
The audit reads supported skill folders and local session evidence. Your files stay on your machine.
$ npx github:Ryan-yang125/skill-manager audit$ brew install Ryan-yang125/skill-manager/skill-manager$ npx skills add Ryan-yang125/skill-manager --skill audit-agent-skills -g -yMatch local evidence
Search Codex and Claude session logs for traceable references. Every count points back to evidence on disk.
Keep, review, archive
Use age, overlap, context size, and evidence coverage to make a human decision. Archive and restore stay reversible.
The inventory layer
Know what loads before the prompt.
Agent skills multiply across folders, plugins, and package sources. Skill Manager turns that filesystem sprawl into an inspectable library.
Origin and path
See the exact folder, agent root, package metadata, source URL, and every detected location for a skill.
Usage evidence
Review usage count, last reference time, session kind, confidence, and the local evidence path behind each match.
Catalog context estimate
Estimate the discovery metadata formed by each skill name and description, then compare that catalog footprint across the inventory.
Reversible cleanup
Write a durable ledger before moving a skill. Restore returns it to the original path and protects existing folders.
3
default global skill roots covered in v0.6.0
Evidence over guesses
Every signal stays inspectable.
Skill Manager records the session path, evidence kind, timestamp, and confidence behind every usage match. Cleanup reports export as Markdown and JSON for review.
Interpretation rule: zero matched evidence means “review this skill.” Session coverage can be incomplete, and rare skills may still be essential.
Read the evidence guide →Field guides
Answers for a growing skills folder.
Practical references for finding skill locations, reading usage signals, reducing duplicates, and keeping every cleanup recoverable.
Health check
Audit agent skills from the terminal
Build a trustworthy inventory across Codex, Claude, and shared roots.
Usage evidenceFind unused agent skills safely
Combine session evidence, recency, overlap, and human review.
CodexCodex skill locations explained
Understand shared, user, system, and project skill scope.
ContextEstimate Claude skill catalog context
Measure name-and-description metadata and interpret the estimate within its defined scope.
ConflictsResolve duplicate agent skills
Compare paths, sources, content, and agent precedence.
RecoveryArchive and restore skills safely
Use dry runs, ledgers, path guards, and post-action scans.
Free and open source
Run your first skill audit.
Use the CLI for a fast report, or download the desktop app for visual inspection, file previews, archive history, and restore.