AutoSkillit is a Claude Code plugin that runs YAML recipes through a two-tier orchestrator. Bundled recipes turn GitHub issues into merged PRs by chaining plan, dry-walkthrough, worktree, test, and PR-review skills against 42 MCP tools and 109 bundled skills.

autoskillit-demo.mp4

Requirements

• Python 3.11+
• uv — package manager
• Claude Code CLI — npm install -g @anthropic-ai/claude-code
• gh CLI — required for PR creation, issue management, CI status

Install

curl -fsSL https://raw.githubusercontent.com/TalonT-Org/AutoSkillit/stable/install.sh | sh

Quick try

cd your-project
autoskillit init
autoskillit order implementation

What it does

Each bundled recipe is a sequenced graph of skill invocations. The orchestrator holds a kitchen of 40 kitchen-tagged MCP tools plus 2 free range tools (open_kitchen, close_kitchen), launches headless Claude sessions for the heavy work, and routes verdicts through retry, merge, and review gates. The 5 bundled recipes are implementation, implementation-groups, merge-prs, remediation, and research.

Documentation

Full docs are under docs/. Topic entry points:

• docs/getting-started.md
• docs/installation.md
• docs/configuration.md
• docs/recipes/overview.md
• docs/execution/architecture.md
• docs/safety/hooks.md
• docs/developer/contributing.md

Examples

Real end-to-end runs against TalonT-Org/spectral-init are in docs/examples/research-pipeline.md.

License

MIT