claude-skills-rj0217

SOP-driven Claude Code skills collection. Each skill delegates deterministic work to scripts and keeps AI focused on judgment. Skills are guides for AI, not command lists.

📄 Available in: 繁體中文

Quick Install · Skills · Philosophy · Backward Compatibility · Support the project

A curated collection of Claude Code skills under one consistent design philosophy. Each skill ships with:

• SKILL.md — flow framework, judgment guidance, common AI failure modes
• scripts/ — deterministic, repetitive work delegated to bash
• references/ — checklists, severity matrices, report templates
• docs/ — fact source + iteration history

The AI reads the SOP and applies judgment; repetitive work is handed off to scripts. This saves tokens, ensures reproducibility, and keeps AI focused on value-add decisions.

Scope: this repo is a collection, not a single-topic toolbox. Current inventory is security-focused (v1 is pre-launch-audit), but future additions may span documentation workflow, memory management, code review, and other domains.

/plugin install https://github.com/p62003/claude-skills-rj0217

Or via marketplace (if the repo evolves into one later):

/plugin marketplace add https://github.com/p62003/claude-skills-rj0217
/plugin install claude-skills-rj0217

git clone https://github.com/p62003/claude-skills-rj0217.git
cp -r claude-skills-rj0217/skills/* ~/.claude/skills/

Restart Claude Code / Desktop to load.

Audit any website from an AI crawler's perspective. Works with Docusaurus / Next.js / Hugo / WordPress / static sites.

• 4-phase flow: infrastructure check (robots.txt, sitemap, llms.txt) → content discoverability (homepage, entry pages, sample articles) → known pitfalls (SPA, search hijack, tag dilution) → structured report
• No scripts — this skill is pure AI judgment; all data fetching uses Claude's built-in WebFetch
• Framework-specific pitfall database: Docusaurus, Next.js, Hugo, WordPress
• Includes llms.txt creation guide (llmstxt.org spec)

Differentiator: most SEO audit tools check human-facing signals (meta tags, Core Web Vitals). This skill checks AI-facing signals — can ChatGPT/Perplexity/Gemini actually discover and understand your content?

Designed for Next.js / React / Vue / static sites before going public. Covers:

• 5-phase flow: context loading → passive recon → white-box audit → active testing (authorization-gated) → fix + verify + report
• 4 scripts: passive_recon.sh, bundle_scan.sh, http_method_fuzz.sh, dep_cve_scan.sh
• 13 attack vectors mapped with expected/anomalous responses (XSS, SQLi, path traversal, CSRF, rate-limit probing, bundle secret leaks, dependency CVEs, etc.)
• Severity matrix (HIGH/MEDIUM/LOW/GOOD) with explicit anti-patterns
• Report template including meta-learning section and competitive positioning block

Differentiator from Claude's built-in /security-review: /security-review audits pending changes on the current branch; pre-launch-audit audits the entire site including live behavior and can execute authorized active tests.

All skills in this repo follow one principle:

Skill ≠ command list. A skill is an SOP for the AI to read and a toolbox for the AI to use. Judgment is the AI's job; repetitive execution is delegated to scripts.

Why this matters:

• Token efficiency: AI doesn't regenerate the same curl chains every session
• Reproducibility: scripts give deterministic output, AI behavior can vary
• Focus: AI value is in judgment, not in executing boilerplate
• Cross-session stability: scripts aren't affected by context window or model version
• No false sense of completion: framework-style SKILL.md forces per-step thinking rather than checklist checkmarks

claude-skills-rj0217/
├── .claude-plugin/
│   └── plugin.json                # Claude Plugin manifest
├── README.md / README_zh.md
├── .gitignore
└── skills/
    └── <skill-name>/
        ├── SKILL.md               # SOP (main)
        ├── scripts/               # Deterministic tools
        ├── references/            # Attack vectors / severity / templates
        └── docs/                  # Fact source + iteration history
            ├── index.md
            └── work_progress.md

This repo commits to:

• Adding a new skill = only add a directory under skills/; no structural change
• Skill version upgrades = logged in each skill's docs/work_progress.md; no breaking changes to the SKILL.md trigger conditions
• Future marketplace upgrade (if the collection grows large enough) = this repo will remain a valid plugin install path; existing users will not need to migrate

Dev:     ~/.claude/skills/<name>/            # WSL ext4, Claude Code reads here directly
Publish: /mnt/d/SKILL/skills/<name>/         # This repo, git-tracked

New skill:

1. Develop + test in ~/.claude/skills/<new>/
2. When stable, cp -r to /mnt/d/SKILL/skills/<new>/
3. Write docs/index.md (fact source) + docs/work_progress.md (first iteration log)
4. Update the "Included Skills" table in both READMEs
5. Commit + push

Skill iteration:

1. Edit + test in dev
2. cp -r to publish (skip docs/ — those only live on publish side)
3. Prepend an entry to <skill>/docs/work_progress.md
4. Commit + push

If these skills save you time or money (e.g. skipping a paid pentest, or faster shipping), consider supporting further development:

0x55c439b27807415e80452f59ba00fee3441a802d

Network: BEP20 (Binance Smart Chain). Accepted: USDT, USDC. ⚠️ Other networks (ERC20 / TRC20) are NOT monitored on this address — do not send.

Costs nothing but helps discoverability. If this repo is useful, a star tells others it's worth their time.

• Discord: https://discord.gg/x52CBg4rcE
• Email: dont.stop.ha@gmail.com

rj0217 · created 2026-04-18