Overview

Patchmill is a CLI for running issue-tracker-driven agent workflows in a repository.

It uses the issue tracker as the visible state record: issues, labels, comments, pull requests, and approval labels show where work is in the workflow. Patchmill uses local git state and worktrees for repository changes, then writes status back to the issue tracker.

Patchmill is configured through skills. A skill is a versioned instruction set for one phase of the workflow, such as triage, specification and plan writing, implementation, visual evidence, review, or landing decisions.

The default workflow relies heavily on the Superpowers skill pack. Patchmill provides the factory floor: orchestration, state, safety checks, and host updates. Superpowers provides much of the expertise that moves work through that floor: planning discipline, implementation habits, debugging, review, and branch finishing.

The important boundary is the contract around the black box:

• Patchmill code owns the workflow state machine, host-provider side effects, git safety checks, worktree creation, artifact validation, approval gates, and final result validation.
• Skills own the instructions given to agents for judgment-heavy work inside a phase.
• The issue tracker records what happened so humans can inspect, approve, retry, or stop work without relying on hidden agent state.

Workflow

Patchmill combines deterministic code for state, safety, and validation with skills for judgment-heavy agent work so each issue gets the best end result.

What to read next

• Start with the quickstart.
• Learn how repository behavior is configured in configuration.
• Use Triage and Run-once when you want the detailed command behavior.
• Read workflow artifacts when you want Patchmill to reuse developer-authored specs or plans.