ai-skills

Production-ready AI skills for coding agents to prevent hallucinations in API calls.

ai-skills

License: MIT Skills Claude Code Cursor

Production-ready AI skills for Claude Code, Cursor, and other AI coding agents. Verified references for frameworks, APIs, and dev workflows.

Stop your AI agent from hallucinating API calls. Give it verified, up-to-date knowledge instead.

What is a skill?

A skill is a structured reference document that gives an AI coding agent verified, up-to-date knowledge about a specific technology. Instead of guessing API signatures or hallucinating method names, the agent reads the skill and writes correct code on the first try.

Skills are not tutorials — they are concise, machine-optimized references extracted from official documentation and verified against real installations.

Why?

AI coding agents have knowledge cutoffs. Libraries evolve. When EasyAdmin 5 drops linkToCrud() in favor of linkTo(), your agent doesn't know — unless you tell it.

Skills bridge that gap:

  • Verified against official docs and real package versions
  • Structured for fast LLM consumption (tables, signatures, examples)
  • Maintained — updated when libraries release breaking changes

Available skills

SkillVersionDescription
easyadmin55.0.2+EasyAdmin 5 for Symfony — dashboard, CRUD, fields, filters, actions, events, security, design, migration from v4
symfony-security-audit1.0.0Symfony 6/7 security audit — 10 recurring antipatterns (IDOR, injection, XSS, SSRF, auth) + 4 deep dives + grep-based scan plan + report format

How to use

Claude Code

Add skills to your project's .claude/skills/ directory, or reference them from a central location in your CLAUDE.md:

# CLAUDE.md
When working with EasyAdmin, read the skill at /path/to/ai-skills/skills/easyadmin5/SKILL.md before writing any code.

Or copy the skill directly into your project:

cp -r skills/easyadmin5 /your-project/.claude/skills/

Cursor / Other agents

Most AI coding agents support context files. Add the skill's SKILL.md as context when working with the relevant technology:

  • Cursor: Add to .cursor/rules/ or reference in project instructions
  • Windsurf: Add to .windsurfrules or workspace context
  • Generic: Include the file path in your system prompt or project docs

Manual reference

Each SKILL.md is also a perfectly readable human reference. Bookmark it, cat it, or keep it open in a tab.

Structure

skills/
└── <skill-name>/
    └── SKILL.md          # The skill reference (single file, self-contained)

Each skill is a single Markdown file — no dependencies, no build step, no config.

Contributing

Contributions are welcome! Fork the repo and open a PR. You can:

  • Create a new skill for a technology that is not covered yet
  • Improve an existing skill — fix errors, add missing methods, update for new versions, improve examples

Adding a new skill

  1. Fork this repository
  2. Create skills/<name>/SKILL.md
  3. Follow this structure:
    • Header: Name, version, source URLs
    • Breaking changes (if migrating from a previous version)
    • API reference: Method signatures, parameters, return types
    • Code examples: Minimal, correct, copy-pasteable
    • Common patterns: Real-world usage, not toy examples
  4. Verify everything against official documentation and source code — never guess
  5. Open a PR from your fork

How we verify skills

Skills go through an iterative review process before merging:

  1. Extraction — API reference drafted from official docs and source code
  2. Multi-agent review — Different AI agents cross-check every signature, parameter, and example against the real source code. Each reviewer catches errors the previous one missed
  3. Human review — Final validation, judgment calls on what to include, and corrections
  4. Iteration — Repeat until the error rate converges to zero

This process is why we trust the result: no single author (human or AI) reviews their own work.

Quality rules

  • Every method signature must be verified against official docs or source code
  • No hallucinated APIs — if you are not sure, check
  • Keep it concise — agents need facts, not prose
  • Include version numbers — skills are version-specific
  • Update when the library releases breaking changes

Author

Juanjo Payá · Blog

License

MIT

If you find this useful, a star helps other developers discover it.