aiee-skills

The AI Enhanced Engineer's Team

🌐 Website · 📚 Field Guides · 📧 Subscribe

The skill library behind production-grade AI engineering — 11 specialist agents and 102 skills, the open starter set from AIEE's Agent Skills series. Clone it, read the code, reproduce the system.

---

Table of Contents

• What This Is
• Skills as a Living System
• Quick Start
• Install by Technology
• The Team
• The Skills
• Commands
• How It Fits
• Repository Structure
• Contributing
• License

---

What This Is

Ask Claude to build an API and it writes 500 lines of FastAPI in 30 seconds: no tests, no migration rollback, no SQL injection guard. The model can code; it just defaults to the average of everything it was trained on. Production needs the opposite: the specific decisions a senior engineer makes without thinking.

A skill encodes those decisions. It is a directory with one required file, SKILL.md, carrying a mental model and its decision rules. An agent loads the skills a task needs, when it needs them — the same RAG mechanic that keeps generated code at production standards instead of internet-average defaults. Loading skills on demand rather than inlining every standard into every agent saves roughly 70% of the context an always-on approach would burn.

This repository is that library in the open: the actual skills and the agents that consume them, so you can read the craft, not just the claims. It is the curated, generalizable subset — 102 skills and 11 agents — distilled from AIEE's production engineering practice.

Skills as a Living System

A skill library isn't written once. It's a living system — designed to stay legible, maintained to stay honest. Every skill here follows the same five disciplines the Agent Skills series lays out:

• Shape — partition by urgency. Act-now rules go in SKILL.md; depth moves to optional reference.md and examples.md.
• Findability — the description field is the routing contract. The agent reads it and nothing else to decide whether to load the skill, so it's written in the caller's vocabulary.
• Voice — describe a consequence, not a command. The reader is a reasoning engine, not a recruit.
• Portability — no project-specific paths in SKILL.md. A skill is written to leave home.
• Evolution — if you can't write a one-sentence description that separates a candidate from every skill already here, it's an update, not a new skill.

The two failure modes these guard against: bloat (one skill grows into an essay nobody loads) and sprawl (hundreds of thin skills nobody can find).

Quick Start

Plugin marketplace (recommended for Claude Code):

/plugin marketplace add ai-enhanced-engineer/aiee-skills
/plugin install aiee-skills@aiee-skills
/agents   # verify the specialists are listed

npx (no marketplace — copies into a .claude directory):

npx aiee-skills install              # into ./.claude for this project
npx aiee-skills install --global     # into ~/.claude for every project

Runs straight from GitHub before any npm publish: npx github:ai-enhanced-engineer/aiee-skills install. Other commands: --dry-run, --force, uninstall, --help.

Local development: git clone https://github.com/ai-enhanced-engineer/aiee-skills.git then claude --plugin-dir ./. JSON manifests validate with just validate.

Install by Technology

The pack is grouped by technology, and each group brings its specialist agent(s) plus skills. dev-practices is shared by every agent, so it's always included. A full install (no --groups) gets all 11 agents and 102 skills.

npx aiee-skills --list-groups                              # groups, skill + agent counts
npx aiee-skills install --groups=frontend-web,backend-api  # install just those