Introduction

contractor is a command-line workflow tool that takes AI coding agents from a rough prompt all the way to a merged branch. It manages each change as a blueprint — a small graph of artifacts (proposal, requirements, design, tasks) — and drives it through a repeatable pipeline of implement, review, and close steps.

It's designed for Claude Code and invoked via slash commands like /contractor:propose, /blueprint:implement, and /blueprint:review.

Who it's for

contractor is built for engineers using AI coding agents on real codebases. It's most useful when:

• You want the agent to produce a proposal and requirements before touching code, not after.
• Your changes are too big to hand to an agent as a single prompt, and you want them broken into reviewable task-groups with a commit per group.
• You want the agent to iterate against a test runner, not its own self-review, before marking work done.
• You want a durable record of why a change exists — not just the code diff.

What you get

• A four-artifact graph per change (proposal → requirements + design → tasks) with dependency-aware status tracking.
• A pipeline runner that chains phases (implement, review, close) into an automated sequence, with human-approved gates between them.
• Three nested loops — inner (write → verify → fix inside a task-group), outer (the full blueprint lifecycle), and middle (observations carried between sessions).
• A git worktree per blueprint so multiple changes can be in flight without stepping on each other.
• A TUI dashboard (and a companion web dashboard) for watching runs and approving gates.

Where to go next

• Get started — install contractor, initialize a repo, and drive your first blueprint end-to-end.
• Concepts — the three loops, the artifact graph, and how pipelines advance work.
• CLI reference — every command, grouped by setup, blueprints, and execution.

You can also open the dashboard to watch runs from the browser.