AgentMachinist

AgentMachinist is a local-first, issue-to-reviewed-PR build pipeline for solo developers. It connects GitHub issues to Claude Code, OpenCode, Pi, or Codex, with a human-approved specification between planning and implementation.

issue + trigger label → spec commit → draft PR → SHA-bound approval
                    → implementation → test gate → ready PR → human merge

The controller—not the harness—owns commits, pushes, PR transitions, and task records. AgentMachinist never merges.

Current release: AgentMachinist 0.8.3 on PyPI.

Install

uv tool install agentmachinist

You also need git, an authenticated gh, and one supported harness executable (claude, opencode, pi, or codex).

machinist update-check compares the installed release against PyPI and prints the upgrade command for how this copy was installed (uv tool, pipx, pip, or a source checkout). machinist doctor reports the same result as a diagnostic row. Set MACHINIST_NO_UPDATE_CHECK=1 to suppress both probes on offline or CI machines.

Upgrading the package is not always the whole upgrade. Managed workflows are projected files, so a workflow change only takes effect once you run machinist sync-workflows. machinist watch reports that drift at startup and machinist update-check reports it alongside the release comparison, so you do not have to run doctor to find out. The advisory never blocks a command and never appears in update-check --json.

Start

cd your-repository
machinist init
# Answer the setup questions, review the generated files, then:
machinist doctor
machinist sync-workflows --check
git status --short
git add machinist.yaml .machinist/specs/.gitkeep .gitignore
git add -p .github/workflows
git diff --cached
git commit -m "chore: configure AgentMachinist"
git push
machinist watch

In a terminal, machinist init asks a short set of setup questions — dispatch mode, managed workflows, harness, test gate, and notifications — each with a one-line explanation and a safe default. Flags such as --harness, --test-cmd, --spec-source, and --notifications pre-answer their questions; --no-input (or a non-interactive shell) skips the questions and uses the defaults plus test-command auto-detection.

Review the staged diff before committing. Managed workflows must be pushed before GitHub comment or label approval can record SHA-bound evidence. machinist init also adds /.machinist/runs/ to .gitignore. If you manage workflows yourself, machinist init --no-workflows records github.manage_workflows: false; doctor then reports that its drift check was intentionally skipped.

The default github.spec_source: local makes watch own spec generation. Choose github-actions and run machinist sync-workflows if CI should own that phase instead. Exactly one source is active, preventing duplicate spec runs.

Approval is bound to the exact PR head commit. Use either:

machinist approve --issue 57
# or: machinist approve --pr 18
# or post the SHA-bound comment shown in the Spec PR body:
# /machinist-execute <full-spec-commit-sha>

The positional target remains available when that number identifies only one Task; use --issue or --pr when GitHub numbers overlap.

Editing the spec after approval makes that approval stale and blocks execution until the new head is approved.

Revise or explicitly abandon a successful Spec by issue number:

machinist spec 42 --dry-run
machinist spec 42 --revise
machinist spec 42 --abandon --reason "requirements changed"

The dry run prints the proposed Spec without commits, pushes, or a PR. Revision regenerates the Spec on its existing branch and draft PR. Abandonment records the reason, removes the trigger and approval labels, and closes the open draft PR.

Commands

Documentation

• Getting started
• Visual first-run field guide
• Architecture and lifecycle
• Operator runbook
• Trust model
• Harness support
• Contributing
• Changelog

The trust model is deliberately narrower than “the agent cannot use git.” Harness flags, credential reduction, repository postconditions, and push leases reduce risk, but local harnesses still execute with the operating-system access of the user who launched them. Read the trust model before unattended use.

Releasing

Releases use PyPI Trusted Publishing. Bump pyproject.toml, update the changelog, and publish a GitHub Release tagged v<version>. The release job first checks tag/version equality, runs tests and workflow checks, builds both distributions, smoke-tests the wheel, and records SHA-256 hashes. A minimal job then publishes those verified artifacts. Only after publication do separate jobs attach the distributions and checksum file to the GitHub Release and verify that the exact version is visible and installable from PyPI.

License

MIT — see LICENSE.