A 338-skill army for AI coding agents

What it does

claude-skills is a massive library of modular instruction packages — 338 skills across 16 domains — that you plug into AI coding agents so they stop winging specialized tasks. Each skill bundles a SKILL.md playbook, Python CLI tools (all stdlib-only, 533 total), and reference docs. The repo converts and ships native formats for 13 tools including Claude Code, OpenAI Codex, Gemini CLI, Cursor, Aider, and Windsurf.

The interesting bit

The project treats agent expertise as infrastructure: versioned, swappable, and multi-platform. A “skill” tells the agent how to execute; an “agent” defines what to do; a “persona” sets who is thinking. You can stack them — so your “Startup CTO” persona can pull the “chaos-engineering” skill and the “slo-architect” skill mid-conversation. The conversion layer is the quiet workhorse: one source tree fans out to .mdc rules for Cursor, CONVENTIONS.md for Aider, .windsurf/skills/, and so on.

Key highlights

• 338 skills covering engineering (129 total across core + advanced tiers), marketing (46 with AEO/LLM-citation tracking), regulatory (18 for ISO/FDA/GDPR/SOC 2), and full C-suite advisory (66 personas from CFO to CAIO)
• 533 Python CLI tools with zero pip dependencies — runs anywhere Python does
• Native support for 13 agents via format conversion; includes pre-generated trees for Hermes Agent and Mistral Vibe with one-shot sync scripts
• Research Operations stack (v2.9.0) spans clinical research, R&D finance, market sizing, and product research with opt-in autoresearch bridges
• Security features include PreToolUse hooks and a dedicated security-auditor skill

Caveats

• Some tools (Hermes, Mistral Vibe) are “BYO-sync tier”: you run local scripts to copy pre-generated trees into agent directories — not fully automated
• The README’s star count badge (5,200+) lags the actual 17,749 stars, suggesting the project is growing faster than its own marketing
• C-level advisory skills are elaborate (66 personas, 21 slash commands) — whether this level of roleplay granularity improves output quality over simpler prompting is untested in the sources

Verdict

Worth a look if you’re running AI agents for real work and tired of re-explaining your stack, compliance regime, or marketing playbook in every session. Probably overkill if you just want Copilot to write slightly better Python.