skill-guide

A prompt-driven skill navigator for Claude Code -- scans your installed skills and explains them in your language.

你的 Claude Code 技能導覽員 -- 掃描已安裝的 skills，用你的語言解釋給你聽。裝了就會用，不再盲目 clone。

Status: MVP -- Tested with 10 skills + 15 commands. Core features work. See Known Limitations.

繁體中文完整說明 | English (this page)

---

TL;DR — 這是什麼？

你從 GitHub clone 了一堆 Claude Code skills，但不知道怎麼用？/guide 幫你：

/guide              → 列出所有已安裝的 skills，分類說明，告訴你從哪開始
/guide hunt         → 用白話解釋某個 skill 怎麼用
/guide 我想找漏洞    → 根據你的目標，推薦該用哪些工具、什麼順序
/guide --check      → 檢查你的環境缺少哪些工具
/guide --diff <repo> → 比較你跟某個 skill 套件差了什麼

安裝只要一行：

git clone https://github.com/VV1NN/skill-guide.git && cd skill-guide && ./install.sh

---

The Problem

The Claude Code skills ecosystem is booming -- hundreds of thousands of skills on GitHub. But after installing them, most users hit the same wall:

• Skills are documented in English (or not documented at all)
• README files explain what a skill can do, not how to actually use it
• You install 10+ skills and have no idea where to start
• There's no way to see all your installed skills at a glance

skill-guide is a Claude Code skill that acts as a navigator for all your other skills -- it scans what you have installed and explains everything in your language.

How It Works

skill-guide is not a traditional program. It consists of two markdown files that instruct Claude how to behave:

File	Type	Purpose
SKILL.md	Skill (auto-loaded)	Background knowledge -- Claude automatically understands how to navigate skills when the topic comes up
commands/guide.md	Slash Command	The /guide command -- explicitly triggered by the user, contains scanning logic and output format

This means: the quality of output depends on Claude's ability to read your skill files, understand their content, and generate helpful explanations. It works well in practice, but it's heuristic, not deterministic. See Known Limitations.

Skills vs Commands -- a distinction many users don't understand:

• Skills load automatically when Claude detects relevant context. You don't need to do anything.
• Commands require you to type /command-name to trigger them.

skill-guide itself helps explain this distinction for all your other installed skills.

What It Does

Command	What happens
/guide	Scans all installed skills & commands, shows a categorized overview with workflow recommendations
/guide <name>	Deep dive into a specific skill -- explains what it does, when to use it, how to use it, with examples
/guide <goal>	Describe what you want to do in plain language, get a step-by-step plan using your installed skills
/guide --check	Best-effort dependency detection -- extracts tools/APIs mentioned in your skills and checks if they're installed
/guide --diff <repo>	Compares your installed skills against a GitHub repo and shows what's missing. Works best with repos that follow standard skill directory layout

Quick Start

Option 1: One-line install

git clone https://github.com/VV1NN/skill-guide.git && cd skill-guide && ./install.sh

Option 2: Manual install

git clone https://github.com/VV1NN/skill-guide.git

mkdir -p ~/.claude/skills/skill-guide
cp skill-guide/SKILL.md ~/.claude/skills/skill-guide/SKILL.md
cp skill-guide/commands/guide.md ~/.claude/commands/guide.md

Verify installation

Open Claude Code and type:

/guide

You should see a categorized overview of all your installed skills.

Usage Examples

Overview -- "What do I have?"

/guide

Returns a skill-centric map -- each skill is a section with its related commands underneath:

10 Skills (auto-loaded) | 15 Commands (type /command to use)

### bug-bounty [SKILL]
Complete vulnerability hunting knowledge base.

Related commands:
  /scope target.com    -- confirm target is in scope
  /hunt target.com     -- actively hunt for vulnerabilities
  /chain               -- chain small bugs into bigger ones

Suggested flow: /scope -> /hunt -> /chain

### web2-recon [SKILL]
Recon methodology for web targets.

Related commands:
  /recon target.com    -- run full recon pipeline
  /surface target.com  -- rank attack surface

Suggested flow: /recon -> /surface

### General / Utilities
  /guide               -- skill navigator (this)
  /autopilot target.com -- fully automated workflow
  ...

Deep Dive -- "How does this work?"

/guide hunt

Returns a beginner-friendly explanation:

## /hunt -- Vulnerability Hunter

One-sentence: Actively searches for security vulnerabilities on a target.

When to use:
- You've finished recon and have a list of endpoints
- You want to test a specific vulnerability class (SSRF, IDOR, etc.)

How to use:
  /hunt target.com
  /hunt target.com --vuln-class idor

What comes before: /recon, /surface
What comes after: /validate, /report

Recommendation -- "I want to do X"

/guide I want to find IDOR vulnerabilities on a website

Returns a step-by-step plan:

Goal: Find IDOR vulnerabilities on a target website

Step 1: /scope target.com                       -- Confirm it's in scope
Step 2: /recon target.com                        -- Map all endpoints
Step 3: /hunt target.com --vuln-class idor       -- Hunt for IDORs
Step 4: /validate                                -- Verify the finding is real
Step 5: /report                                  -- Generate submission report

You do NOT need: web3-audit, meme-coin-audit, token-scan

Dependency Check -- "Am I ready?"

/guide --check

Best-effort scan of tools and APIs your skills depend on:

## Environment Health Check

| Tool       | Status        | Required by           |
|------------|---------------|-----------------------|
| subfinder  | OK (v2.6.3)   | web2-recon, /recon    |
| nuclei     | MISSING       | web2-recon, /recon    |

### How to fix
- nuclei: `go install -v github.com/projectdiscovery/nuclei/v3/cmd/nuclei@latest`

Note: Dependency detection is heuristic -- it extracts tool names mentioned in skill content. It may miss unlisted dependencies or flag tools that are optional.

Diff -- "Am I missing anything?"

/guide --diff user/awesome-bb-skills

Compares your skills against a GitHub repo:

| Skill        | In source | Installed | Status  |
|--------------|-----------|-----------|---------|
| bug-bounty   | Yes       | Yes       | OK      |
| api-testing  | Yes       | No        | MISSING |

Note: Diff works best with repos that use standard skills/*/SKILL.md and commands/*.md directory layouts. Non-standard structures may not be fully detected.

Tested With

This skill has been tested in the following environment:

• 10 skills: bb-methodology, bug-bounty, meme-coin-audit, report-writing, security-arsenal, triage-validation, web2-recon, web2-vuln-classes, web3-audit, skill-guide
• 15 slash commands: autopilot, chain, guide, hunt, intel, recon, remember, report, resume, scope, surface, token-scan, triage, validate, web3-audit
• Shell: zsh (macOS default) and bash
• All 5 modes tested: overview, deep dive, recommendation, --check, --diff

Bugs found and fixed during testing

Bug	Cause	Fix
${!var} env var check failed on macOS	zsh doesn't support bash indirect expansion	Wrapped in explicit bash -c
Glob patterns errored when no files matched	zsh treats no-match as error (unlike bash)	Wrapped all scans in explicit bash -c

Requirements

• Claude Code (CLI, desktop app, or IDE extension)
• Skills and commands must have frontmatter with name: and/or description: fields
• --diff mode requires git to clone remote repos
• --check mode requires bash (for env var detection)

Known Limitations

• Prompt-driven, not deterministic: Output quality depends on Claude's interpretation of your skill files. Results are best-effort and may vary between sessions.
• Dependency detection is heuristic: --check extracts tool names mentioned in skill content. It cannot detect dependencies that aren't explicitly named, and may flag optional tools as required.
• Diff requires standard layout: --diff looks for skills/*/SKILL.md and commands/*.md. Repos with non-standard structures may not be fully scanned.
• Frontmatter required: Skills without name: or description: in their YAML frontmatter will be listed with incomplete information.
• No caching: Every /guide invocation rescans the filesystem. This is by design (always fresh) but means no persistent state between sessions.
• Plugin marketplace skills: Scanning ~/.claude/plugins/ paths is supported but less thoroughly tested than user-level skills.

Compatibility

• Works with any Claude Code skills that use standard SKILL.md format
• Scans user-level (~/.claude/) and project-level (.claude/) skills
• Compatible with plugin marketplace skills (best-effort)
• Shell-compatible: tested on both zsh (macOS default) and bash
• No dependencies -- it's just two markdown files

Uninstall

rm -rf ~/.claude/skills/skill-guide
rm ~/.claude/commands/guide.md

Contributing

Contributions are welcome! Some ideas:

• Test with different skill sets -- report how guide works with your skills
• Edge cases -- skills without frontmatter, unusual directory structures
• Screenshots -- real output examples in different languages
• Translations -- improve the README in your language

License

MIT