Skip to content

gitbulk

gitbulk is a nightly fleet-maintenance tool for a developer who works across many GitHub repositories.

Give it a list of repos you contribute to, and gitbulk reports the state of your open pull requests, flags the ones that need your attention, and — when you ask it to — launches Claude Code agents to fix common problems, rebases PRs onto their default branches, auto-merges PRs that meet a policy you configure, and closes stale PRs. It also treats your local clones as first-class: discovering and cleaning up post-merge cruft (orphaned worktrees, undeleted branches) and surfacing repos that need work no PR yet exists for.

Safe to run from cron

gitbulk is designed to run unattended alongside your ongoing development work on the same clones. It never touches your working tree, index, or current branch — every operation that needs a checkout uses a disposable worktree. See the local-git safety contract.

Get started

  • Install — grab the single binary (or install from source) and put it on your PATH.
  • Configure — point gitbulk at your repos and set your merge/triage policy.
  • Commands — the full subcommand reference: report, merge, dispatch, prune-branches, and the rest.
  • Running unattended — wire it into cron and surface attention in your shell prompt.

How a typical day looks

  1. Overnight, cron runs gitbulk report against your fleet and writes a structured triage report. If anything needs you, it refreshes an ATTENTION sentinel.
  2. In the morning, your shell prompt shows a quiet ⚠ gitbulk marker. You run gitbulk show to read the dashboard.
  3. On a schedule you choose, gitbulk can merge PRs that satisfy your policy, rebase-pr the ones that fell behind, dispatch AI agents at common fixes, and prune-branches / prune-worktrees to clean up after merged work.

Every run is recorded under ~/.cache/gitbulk/runs/, so you can always go back and read exactly what happened — see Inspecting runs.

Requirements

gitbulk shells out to the GitHub CLI (gh) and git for everything, so both must be installed and gh must be authenticated for your account. The binary needs only Python 3.10+ on the machine.

Looking to contribute? The developer README covers cloning, running the tests, and the project's intent-first methodology.