🤖 AI Agent Skills
Production-grade AI agent skills for real-world applications — and everyday use
Inspired by addyosmani/agent-skills and forrestchang/andrej-karpathy-skills
🎯 What is this?
AI agents are only as good as the instructions they follow. AI Agent Skills is a curated, battle-tested library of skill files — structured markdown prompts — that transform AI coding agents (Claude, Gemini, Cursor, Copilot, etc.) into disciplined, production-ready engineers.
Unlike generic prompts, these skills encode actual workflows with steps, checkpoints, verification gates, and anti-pattern guards — the same discipline senior engineers bring to every production deployment.
THINK → PLAN → BUILD → TEST → REVIEW → HARDEN → SHIP
↓ ↓ ↓ ↓ ↓ ↓ ↓
Clarify Break Write Verify Quality Security Deploy
Goals Down Code Output Gates Gates Live
🏆 Why This Repo Goes Beyond the Rest
| Feature | This Repo | agent-skills | karpathy-skills |
|---|---|---|---|
| Production + Everyday use | ✅ | Production only | General |
| Multi-agent orchestration | ✅ | ❌ | ❌ |
| AI safety & hallucination guards | ✅ | ❌ | Partial |
| RAG / Memory skills | ✅ | ❌ | ❌ |
| Prompt injection defense | ✅ | ❌ | ❌ |
| GitHub Pages search UI | ✅ | ❌ | ❌ |
| 30+ curated skills | ✅ | 20 skills | 4 principles |
| Everyday productivity skills | ✅ | ❌ | ❌ |
⚡ Quick Start
Step 1 — Clone the repo
git clone https://github.com/DevelopersGlobal/ai-agent-skills.git
Step 2 — Pick a skill and copy it into your agent
Claude / Cursor / Copilot / any agent:
Copy the contents of any SKILL.md file into your agent's system prompt, instructions file, or context window.
For Claude Code — add to your project's CLAUDE.md:
cat skills/think-before-coding/SKILL.md >> CLAUDE.mdFor Cursor — copy into .cursor/rules/:
cp skills/security-hardening/SKILL.md .cursor/rules/security-hardening.mdc
For Gemini CLI — add to your project's GEMINI.md:
cat skills/goal-driven-execution/SKILL.md >> GEMINI.mdOne-liner (any skill, any agent):
# View raw skill content, then paste into your agent
curl https://raw.githubusercontent.com/DevelopersGlobal/ai-agent-skills/main/skills/production-deployment/SKILL.mdBrowse and copy skills visually → developersglobal.github.io/ai-agent-skills
📚 All 30 Skills
🧠 Think — Before You Code
| Skill | Description |
|---|---|
| think-before-coding | Surface assumptions, manage confusion, prevent hallucination |
| goal-driven-execution | Transform tasks into verifiable goals with success criteria |
| idea-to-spec | Convert vague ideas into concrete, testable specifications |
📐 Plan — Break It Down
| Skill | Description |
|---|---|
| task-decomposition | Break features into atomic, independently verifiable tasks |
| context-loading | Load minimum necessary context; avoid token bloat |
| multi-agent-orchestration | Design and coordinate multi-agent pipelines |
🏗️ Build — Write Production Code
| Skill | Description |
|---|---|
| incremental-coding | Build in verifiable increments; never big-bang rewrites |
| simplicity-first | Minimum code that solves the problem — nothing speculative |
| surgical-changes | Touch only what you must; leave the rest untouched |
| api-design | Design stable, versioned, self-documenting APIs |
| frontend-engineering | Accessible, performant, responsive UI patterns |
| rag-and-memory | Retrieval-Augmented Generation and agent memory patterns |
✅ Test — Prove It Works
| Skill | Description |
|---|---|
| test-driven-development | Red-green-refactor with meaningful coverage |
| debugging-methodology | Systematic root cause analysis; never guess-and-check |
| integration-testing | Test real system boundaries, not mocks of mocks |
| ai-output-validation | Validate, parse, and sanitize AI-generated outputs |
🔍 Review — Quality Gates
| Skill | Description |
|---|---|
| code-review | Structured review checklist; correctness over style |
| performance-optimization | Measure first, optimize second; no premature optimization |
| refactoring | Safe, behavior-preserving transformation with tests |
| documentation | Document decisions, not just implementations |
🔒 Harden — Security & Reliability
| Skill | Description |
|---|---|
| security-hardening | OWASP Top 10, secrets management, least privilege |
| prompt-injection-defense | Guard AI agents against prompt injection attacks |
| hallucination-prevention | Detect and mitigate LLM hallucinations in pipelines |
| error-handling | Graceful degradation and meaningful error messages |
| observability | Structured logging, tracing, and alerting for AI systems |
🚀 Ship — Deploy with Confidence
| Skill | Description |
|---|---|
| production-deployment | Zero-downtime deploys with rollback plans |
| ci-cd-pipelines | Automated quality gates from commit to production |
| git-workflow | Trunk-based development, atomic commits, clean history |
🌟 Everyday — Productivity Skills
| Skill | Description |
|---|---|
| code-explanation | Get clear, layered explanations of unfamiliar code |
| meeting-notes-to-tasks | Convert meeting notes into structured action items |
| research-and-summarize | Distill complex topics into actionable summaries |
🏛️ Skill Anatomy
Every skill follows the same battle-tested structure:
┌─────────────────────────────────────────────────────┐
│ SKILL.md │
│ │
│ ┌─ Frontmatter ────────────────────────────────┐ │
│ │ name: kebab-case-name │ │
│ │ description: One-line trigger description │ │
│ │ applies-to: [claude, gemini, cursor, ...] │ │
│ │ category: think|plan|build|test|... │ │
│ └──────────────────────────────────────────────┘ │
│ │
│ Overview → What this skill accomplishes │
│ When to Use → Exact triggering conditions │
│ Process → Numbered, verifiable steps │
│ Rationalizations → Common excuses + rebuttals │
│ Red Flags → Warning signs the skill is needed │
│ Verification → Non-negotiable evidence required │
│ References → Supporting checklists/docs │
└─────────────────────────────────────────────────────┘
🤝 Contributing
We welcome contributions! The best skills are:
- Specific — Actionable steps, not vague principles
- Verifiable — Clear exit criteria with evidence
- Battle-tested — Based on real production experience
- Minimal — Only what the agent needs; no bloat
See CONTRIBUTING.md and docs/skill-anatomy.md.
🌐 GitHub Pages
The full skills library is browsable at developersglobal.github.io/ai-agent-skills with:
- 🔍 Full-text search across all skills
- 🏷️ Filter by category and agent type
- 📋 One-click copy for any skill
- 📱 Mobile-friendly responsive design
📊 Inspiration & Credits
This project builds on and extends the ideas from:
- addyosmani/agent-skills — Production engineering skills for AI agents
- forrestchang/andrej-karpathy-skills — Karpathy's observations on LLM coding pitfalls
- DevelopersGlobal — Community-driven developer tools
📄 License
MIT — Use these skills in your projects, teams, and tools. Attribution appreciated but not required.
⭐ Star this repo to help developers everywhere build better AI agents ⭐
Made with ❤️ by DevelopersGlobal