Skip to content

Back to the catalog

alirezarezvani/claude-skills
github.comrepo

alirezarezvani/claude-skills

claude-skills is an open-source MIT library of about 398 Claude Code skills, agents and plugins that also installs into Codex, Gemini CLI, Cursor and 9 other coding agents. The security-relevant parts are skill-security-auditor, a stdlib-only Python scanner that checks skills for command injection, data exfiltration, prompt injection and dependency supply-chain risks before install, plus a security-guidance PreToolUse hook, a security suite and compliance skills (ISO 27001, SOC 2, GDPR).

TL;DR Large multi-agent skills library whose security pieces include a pre-install skill scanner and a PreToolUse security hook.
Open original ↗
Also seen from another angle🧠 AI →
Document

What it is

claude-skills is an MIT-licensed library of Claude Code skills, agents and plugins by Alireza Rezvani. The page reports 5,200+ GitHub stars and v2.9.0 for the research-ops bundle. Its main security item is skill-security-auditor (added in v2.0.0), a stdlib-only Python script that scans a skill folder before install. The page also lists a security-guidance PreToolUse hook, a security suite, SecOps skills, env-secrets-manager (leak detection), dependency-auditor and compliance skills (ISO 27001, GDPR, SOC 2), but gives no detail on how any of these work.

How it works

The auditor looks for command injection, code execution, data exfiltration, prompt injection, dependency supply chain risks and privilege escalation. It returns PASS, WARN or FAIL with remediation guidance. The page does not describe its rules, how it detects prompt injection, or how it was tested.

What to do

Scan a skill before copying it into ~/.claude/skills/:

python3 engineering/skill-security-auditor/scripts/skill_security_auditor.py /path/to/skill/

To install only the scanner as a plugin:

/plugin marketplace add alirezarezvani/claude-skills
/plugin install skill-security-auditor@claude-code-skills

The marketplace can also install whole bundles, such as engineering-skills@claude-code-skills. Run the auditor on those too, since each skill ships SKILL.md instructions plus Python scripts.

On Windows, clone with git clone -c core.symlinks=true (Developer Mode on). Otherwise the .gemini, .codex, .vibe and .hermes mirror trees check out as one-line pointer files. Set PYTHONUTF8=1 so Unicode output does not crash the tools.

Caveats

  • The page's own counts disagree: 380, 398, 345 and 346 skills, and 580, 706 and 766 Python tools. Treat any single figure as approximate.
  • The README says every tool is stdlib-only. The book-to-skill document extractors can optionally use third-party parsers, and nothing is installed implicitly.
  • One install path pipes a remote script into bash: bash <(curl -s https://raw.githubusercontent.com/alirezarezvani/claude-skills/main/scripts/openclaw-install.sh). The auditor is meant for skills, and the page does not say it covers the repo's own install scripts.
  • The page offers no false-positive or false-negative data for the auditor, and no description of what the PreToolUse hook blocks.

Organisations named here

  • agentskills.io
  • skill-security-auditor

Technologies named here

  • security-guidance PreToolUse hook
  • incident-commander
  • env-secrets-manager
  • dependency-auditor
  • pr-review-expert
  • Claude Code
  • Codex
  • Gemini CLI
  • Cursor
  • Python
  • OpenClaw
  • PreToolUse hook

Techniques named here

  • Skill security audit
  • Secret leak detection and rotation workflows
  • Prompt Injection
  • Command injection
  • Data exfiltration

Related in AI Security