Skip to content
Li_Js edited this page Jul 21, 2026 · 2 revisions

skill-manager Wiki

English | 简体中文

A practical field guide for keeping Claude Code / Codex skills and MCP servers visible, searchable, and safe to manage.

skm demo

Why This Exists

A local AIDE setup starts tidy. Then one day there are dozens of skills, a few MCP servers, duplicated folders, old experiments, shared symlinks, and commands you half-remember using three weeks ago.

skm gives that toolbox a dashboard.

You wonder Run You get
What is installed? skm scan Skills, MCP servers, categories, sources, context estimate
Is this setup healthy? skm Health score, duplicates, zombie skills, session size
Which skill fits this task? skm ask "task" Best match, reasons, alternatives
Can I see the whole map? skm graph --format html A draggable single-file knowledge graph
Can I share one overview? skm report --format html A local HTML report with health, risks, usage, sessions

The 30-Second Path

git clone https://github.com/GrubbyLee/skill-manager.git
cd skill-manager
node scripts/install.mjs

skm scan
skm
skm ask "convert a web page to Markdown"
skm report --format html --output skm-report.html
skm graph --format html --output skill-graph.html

No npm install path is recommended yet. The package name is reserved, but the current installation path is git clone.

Visual Story

Tool chaos Scan and label
Too many tools Scan and label
Knowledge map Safe cleanup
Knowledge graph Safe cleanup

Wiki Map

Page What it covers
Quick Start Install, first scan, language options, first report
Command Map Every command, grouped by workflow
Skill Recommendation ask / recommend, local ranking, advisor mode
Knowledge Graph Relationship types, HTML controls, reading dense graphs
Overview Report One-page HTML report for health, risks, usage, sessions
Safety and Data Read-only boundary, write operations, backups, MCP privacy
Community and Roadmap How to share, contribute, and help shape the project
FAQ Common installation, usage, and safety questions

Design Principles

  • Local first: no external model is called unless you explicitly use --advisor.
  • Read-only by default: cleanup and disable actions require explicit commands and safeguards.
  • Zero third-party npm dependencies: Node.js built-ins only.
  • Bilingual by design: English and Simplified Chinese output are both supported.
  • Script-friendly: JSON field names stay stable.

Current Release

  • Latest release: v0.1.3
  • CI: macOS / Windows on GitHub Actions, Linux validated locally
  • Runtime: Node.js >= 18
  • License: MIT

Clone this wiki locally