mr-babysit

Drives GitLab merge requests to mergeable by rebasing, fixing CI, and triaging review threads.

15|1|Updated Jan 3, 2026
One-click install
npx skills add https://github.com/fprochazka/claude-code-plugins --skill mr-babysit-fprochazka
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mr-babysit
Source: https://github.com/fprochazka/claude-code-plugins/tree/main/plugins/sdlc/skills/mr-babysit
Command: npx skills add https://github.com/fprochazka/claude-code-plugins --skill mr-babysit-fprochazka

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires glab, jq, glab-discussion, glab-pipeline, and includes references (resource) components.

What problem does it solve? After opening a merge request, authors spend hours polling pipelines, rebasing behind master, and replying to review bot comments. This Skill automates that entire unattended loop: a background watcher reports every change on the MR set, an actor subagent triages and fixes, and you only get involved for genuine product or design decisions. ## Core Features & Use Cases - Continuous MR watching: A background glab:mr-watch agent reports pipeline finishes, pushes, new review notes, draft flips, approvals, and merges across a set of one or more MRs. - Autonomous triage and fixing: An actor subagent rebases branches, retries flaky CI jobs, fixes real failures, and posts reasoned fix or dismissal replies on review threads, all critically evaluated rather than rubber-stamped. - Ledger-based state and handoff: A babysit ledger tracks every thread, job, and rebase row by id, and the run hands the MR set back to reviewers only when pipelines are green, threads are resolved, and nothing is unfinished. - Use Case: You open MRs in a service repo and a pipelines repo for one ticket, invoke the command, and it keeps both rebased and green overnight, answering bot review comments and handing over only when everything is settled. ## Quick Start Ask the AI to babysit the merge requests for the current ticket, for example by saying "babysit MR !123 and !456 until they are green and handed over".

Frequently Asked Questions about mr-babysit

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automatically fix failing CI on a GitLab merge request?▼

Invoke the mr-babysit command with your MRs; a watcher reports each failed pipeline and an actor subagent triages the failure signature, fixes real bugs, retries flaky jobs, and pushes with --force-with-lease after verifying the remote SHA.

How to handle AI review bot comments on merge requests automatically?▼

The actor triages each unresolved review thread into fix, dismiss, ask, or skip dispositions after re-reading the cited code. Approved fixes get a reply naming the SHA and a resolve; wrong findings get a reasoned dismissal, and severity tags alone never escalate a finding.

What tools does GitLab MR automation require?▼

It requires the glab CLI, jq, glab-discussion, and glab-pipeline installed locally. Without the last two, the state script cannot fetch threads or pipeline details and every change reports a missing dependency.

Can it babysit multiple merge requests across different repositories?▼

Yes, the run babysits a set of MRs, each with its source branch checked out in its own worktree. The set comes from MRs you name explicitly or MRs the session created, and the watcher covers all of them.

When does MR babysitting stop instead of handing over?▼

It stops and hands back on genuine product or design decisions, unresolvable rebase conflicts, failure signatures still red after two attempts, could-not-fix rows, or an idle timeout of three hours. A halt on one MR does not stop the others.

Why does the babysit run keep asking for permission on git operations?▼

The run is pre-authorized for rebase, force-push, CI retry, and thread replies, so pausing to ask is a failure. If your harness blocks these commands, allowlist them once per the plugin README rather than approving each operation.