Greenkeeper
Keep GitHub pull requests moving when checks fail or reviews need attention.
Greenkeeper runs in the background on your computer, watching the GitHub pull requests you choose. It sends failed automated checks, merge conflicts, and unresolved review comments to your AI coding tool so the branch can move forward.
Each repair runs in a separate checkout, so your current work stays undisturbed. The tool is written in Go and works with your existing GitHub and coding-agent setup. You review the changes and decide when to merge.
Highlights
- Watch failed checks, conflicts, and reviews
- Run repairs in separate checkouts
- Review changes and decide when to merge
The work after the feature
Finishing a feature does not finish its pull request. Checks can fail after a push, the base branch can move, and a review can arrive while you are working on something else. Reopening tabs and reconstructing the same context is a surprisingly large part of keeping several branches moving.
Greenkeeper moves that monitoring into a local Go process. You choose the repositories and pull requests to watch, then configure which situations should start a repair. It uses your existing GitHub and coding-agent setup, rather than requiring another hosted service to sit between you and the repository.
Turn a signal into work
On each polling cycle, Greenkeeper asks GitHub about merge conflicts, failed automated checks, and unresolved review threads. It compares that snapshot with your rules before starting an agent. Watching a branch does not mean inviting changes on every cycle; there needs to be a qualifying problem.
The resulting prompt includes the failure logs or review text, repository context, and naming preferences. That gives the coding tool a specific piece of work instead of a vague instruction to improve the project. After the process exits, Greenkeeper checks GitHub again to see whether the branch still needs attention.
Another checkout, the same repository
Repairs run in Git worktrees: separate checkouts of the same repository, with their own working files. Your current checkout can stay on the task you are doing while an agent handles the watched branch elsewhere. That separation is especially useful when a rebase leaves files halfway through a conflict resolution.
Claude Code, Codex, and configurable commands run as subprocesses. Greenkeeper does not embed a particular model provider’s SDK, so the integration follows the coding tools you already use. Global and per-repository settings choose providers and concurrency, with limits on simultaneous work and session runtime.
Keep the merge decision
Greenkeeper can repair and push a branch, but it never merges the pull request. Rebase pushes use Git’s lease checks so a changed remote branch is not blindly overwritten. Prompts aim at the reported problem, keeping the work tied to a failure or review rather than encouraging unrelated refactors.
A clean check is useful evidence, not proof that every change is correct. The diff still deserves a person’s attention. The tool handles the repetitive maintenance loop so you can review what changed and decide when the branch is ready.
Related writing
Greenkeeper: A Local Daemon That Babysits Your PRs
Greenkeeper is a local Go daemon that watches your GitHub PRs and runs your AI CLI in a git worktree when CI fails, conflicts appear, or reviews stall.
Interface Driven Development in the Age of AI
Ship AI-generated code without regret by hiding it behind interfaces: contracts, dependency injection, and why the old Java patterns were right all along.