mill

mill

Think it through. Ship it right. Sharper every cycle.

mill is a skill pack for Claude Code that brings structure to AI-assisted software delivery. Markdown skills orchestrate Claude Code's native tools to manage the full workflow — from intent to verified Pull Request.

You describe intent. mill asks the right questions, writes a complete spec, assembles a team of agents, implements in verified steps, and opens a PR. Every cycle feeds learnings back into your project's knowledge base. The more you ship, the sharper it gets.

Quick Start

/plugin marketplace add mindrevolution/mill
/plugin install mill@mindrevolution

Then in your project:

/mill:init                # Set up .mill/ directory
/mill:spec                # Think through what to build → GitHub Issue
/mill:ship 42             # Assemble team, implement, verify → PR

How It Works

flowchart LR
    subgraph Knowledge
        G[Ground]
    end

    subgraph Delivery
        I[Idea] --> S[Spec] --> X[Ship]
    end

    G -.->|informs| I
    G -.->|validates| S
    G -.->|guides| X

    X -.->|learnings| G

Skill	What it does
/mill:init	Initialize .mill/ — directory structure and gitignore
/mill:ground	Build your project's knowledge base — personas, conventions, decisions
/mill:idea	Capture a rough thought — 30 days to develop into a spec or drop with learnings
/mill:spec	Turn intent into a precise, self-contained spec → published as GitHub Issue
/mill:ship	Assemble a team of agents, implement a spec, verify independently → PR
/mill:warmup	Orient Claude to your codebase — architecture, patterns, recent changes

Agent Teams

When you /mill:ship, mill doesn't just generate code in a single pass. It assembles a team:

1. Lead — reads the spec, determines team size, delegates work, manages cross-team contracts
2. Implementer(s) — 1-4 agents with explicit file ownership boundaries, working in parallel where possible
3. Verifier — a separate agent with clean context that reviews the full changeset against every spec criterion

The verifier never sees the implementer's reasoning. This structural independence catches what self-review misses.

The Learning Loop

Every skill writes observations — patterns discovered, gaps noticed, conventions found. /mill:ground curates these into permanent project knowledge: personas, rules, vocabulary, architectural decisions. The next spec is informed by what the last ship learned.

Requirements

• Claude Code
• gh CLI (GitHub CLI) — authenticated
• Git repository on GitHub

Documentation

See AGENTS.md for the full reference — architecture, project structure, skill details, and development guide.

License

MIT