Build tool for AI coding skills — compiles skill definitions into tool-specific output for Claude Code, Cursor, and other AI coding assistants.

Think dbt for AI skills: define your team's coding skills once in YAML, build them into the format each tool expects.

pip install aictrl

If your repo already has an .aictrl/ directory (set up by aictrl.dev or manually):

aictrl build

This reads skill definitions from .aictrl/data/skills/ and renders them into .claude/ and .cursor/ folders.

To start from scratch:

aictrl init --org-id my-org
# Add skill YAML files to .aictrl/data/skills/
aictrl build

your-repo/
├── .aictrl/                    # Committed to git
│   ├── config.yaml             # Org connection config
│   ├── skills.lock             # Locked versions + checksums
│   ├── data/
│   │   ├── org.yaml            # Organization metadata
│   │   └── skills/             # Skill definitions (YAML)
│   │       ├── code-review.yaml
│   │       └── testing-guide.yaml
│   └── overrides/
│       └── skills/             # Team customizations (partial YAML)
│           └── code-review.yaml
├── .claude/                    # Gitignored — built output
│   ├── skills/
│   │   └── code-review/
│   │       └── code-review.md
│   ├── settings.json
│   └── hooks/
│       └── skill-telemetry.sh
└── .cursor/                    # Gitignored — built output
    ├── hooks.json
    └── hooks/
        └── skill-telemetry.sh

.aictrl/data/skills/code-review.yaml   (base skill from aictrl)
  + .aictrl/overrides/skills/code-review.yaml  (team customizations)
  → rendered through Jinja2 templates
  → .claude/skills/code-review/code-review.md
  → .cursor/ hook configs

Each skill is a self-contained YAML file:

slug: code-review
name: code-review
description: "Guides thorough code reviews"
version: "1.2.3"

instructions: |
  You are a code review assistant. Focus on:
  1. Security vulnerabilities
  2. Performance implications
  3. Code readability

sections:
  checklist: |
    ## Review Checklist
    - [ ] No hardcoded secrets
    - [ ] Error handling is complete

tags: [review, quality]
allowed_tools: [Bash, Read, Grep]

Customize skills for your team without forking. Create partial YAML files in overrides/skills/:

# .aictrl/overrides/skills/code-review.yaml
# Only include fields you want to change.

allowed_tools:
  - Bash
  - Read
  - Grep
  - Edit

sections:
  team_standards: |
    ## Our Standards
    - Always check test coverage
    - Review error handling patterns

Merge rules:

• Scalars: override replaces base
• Lists: override replaces base entirely
• Dicts: deep merge (override keys win)
• _delete: [key1, key2]: removes keys from base

Overrides survive skill updates — when aictrl pushes new skill versions, your customizations are merged on top automatically.

Install a git hook so skills rebuild after git pull, git checkout, or git merge:

aictrl install-hook

Use aictrl check in CI to ensure builds aren't stale:

# .github/workflows/ci.yml
- name: Check skills are built
  run: |
    pip install aictrl
    aictrl check

aictrl.dev manages skills centrally and pushes updates to your repos via PR:

1. aictrl creates a PR updating .aictrl/data/skills/ and skills.lock
2. Your team reviews and merges
3. Developers run aictrl build (or it auto-builds via post-checkout hook)
4. .claude/ and .cursor/ are regenerated with latest skills + your overrides

MIT