skill-admin

Original：🇺🇸 English

Translated

3 scripts

Manage skills across 20+ AI platforms (Claude Code, Cursor, Copilot, Gemini, OpenClaw, Hermes, etc.). Use `list` as the unified entrypoint. Default behavior is listing skills only; only guide/recommend when the user explicitly asks what skill to use.

1installs

Sourcesogud/ninjutsu

Added on2026-04-18

NPX Install

npx skill4agent add sogud/ninjutsu skill-admin

SKILL.md Content

View Translation Comparison →

Skill Admin

Manage skills across 20+ AI platforms with single source of truth architecture.

Supported Platforms

claude

cursor

copilot

vscode

cline

roo

goose

codex

windsurf

trae

kiro

opencode

openclaw

hermes

qoderwork

gemini

antigravity

agents

CLI Commands

bash

# Unified list / scan
python3 skill-admin.py list [platform]

# Backward-compatible alias
python3 skill-admin.py scan

# Sync all to .agents + symlink others
python3 skill-admin.py sync-all

# Delete skill from all platforms
python3 skill-admin.py delete <name>

# Generate category graph
python3 skill-admin.py graph [platform]

# Install from GitHub
python3 skill-admin.py install <url> [platform]

# Compare two platforms
python3 skill-admin.py compare <p1> <p2>

# Show symlink topology
python3 skill-admin.py topology

# Security scan (11 dimensions)
python3 skill-admin.py security [path]

# Backup all skills
python3 skill-admin.py backup

AI Output Templates

CRITICAL: After running CLI, format output using templates below. Do NOT dump raw script output.

Guide Flow

Default:

Run
```
python3 skill-admin.py list [platform]
```
List installed skills

Only when the user explicitly asks what skill to use:

Run
```
python3 skill-admin.py list [platform]
```
Use only each skill's
```
name
```
and
```
description
```
to identify relevant skills
Explain what the relevant skills do and why they fit the current task
Do not read full skill bodies unless the user explicitly asks

List Output Contract

list

outputs stable structured text for AI consumption:

```
SKILL_LIST_BEGIN
```
```
SUMMARY ...
```
```
PLATFORM ...
```
```
SKILL ...
```
```
EMPTY_DIR ...
```
if any
```
SKILL_LIST_END
```

Parsing rules:

Fields are tab-separated
Each field is
```
key=value
```
Use
```
SKILL.name
```
and
```
SKILL.description
```
for guide/recommendation
Do not depend on colors or box-drawing characters

Template 1: Summary (for "how many skills")

统计项	数量
总技能数	XXX
已安装平台	XXX
去重后唯一技能	XXX

Template 2: Full Skill List (for "list all skills")

分类	数量	技能列表
💻 开发工具	XX	`skill-a` `skill-b` ...
🎨 设计工具	XX	`skill-a` `skill-b` ...
📚 文档工具	XX	`skill-a` `skill-b` ...
🌐 网络交互	XX	`skill-a` `skill-b` ...
🤖 AI 能力	XX	`skill-a` `skill-b` ...
⚙️ 系统自动化	XX	`skill-a` `skill-b` ...
🧪 测试质量	XX	`skill-a` `skill-b` ...
📊 数据分析	XX	`skill-a` `skill-b` ...
📦 其他	XX	`skill-a` `skill-b` ...
总计	XXX

List ALL skills, NO truncation.

Classification Rules

类别	关键词
💻 开发工具	code, dev, debug, test, tdd, review, git, typescript, react, rax, fliggy, ship, ultron, clamjs
🎨 设计工具	design, ui, ux, figma, mobile, ios, web-design, generative, canvas, infographic, drafter
📚 文档工具	doc, pdf, docx, pptx, xlsx, obsidian, notebook, writing, guide, release
🌐 网络交互	web, browse, browser, gstack, opencli, fetch, search, screenshot, websearch, webfetch
🤖 AI 能力	ai, llm, claude, gemini, agent, pua, superpowers, brainstorm, plan, office-hours
⚙️ 系统自动化	freeze, guard, sync, backup, file, organizer, loop, scheduler, health
🧪 测试质量	qa, test, quality, verify, doctor, investigate, security
📊 数据分析	data, odps, query, rum, analytics, yfinance, stock, finance, correlation

Security Scan

11 dimensions: Prompt Injection, Permission Abuse, Hook Behavior, Data Exfiltration, Social Engineering, Scope Creep, Obfuscation, Destructive Ops, Supply Chain, Identity Impersonation, Sensitive Data.

Score	Label	Action
85-100	TRUSTED	APPROVE
70-84	LIKELY_SAFE	APPROVE_WITH_CAUTION
50-69	SUSPICIOUS	APPROVE_WITH_CAUTION
25-49	HIGH_RISK	REJECT
0-24	DANGEROUS	REJECT

Edit

scripts/security-rules.yaml

to customize rules.

Requirements

Python 3.8+, stdlib only (no dependencies).