Drives a GitHub PR through Copilot review rounds until merge-ready with scoped triage.
npx skills add https://github.com/kpiteira/devops-ai --skill kbabysit-kpiteira
Please help me install this Agent Skill. Skill: kbabysit Source: https://github.com/kpiteira/devops-ai/tree/main/skills/kbabysit Command: npx skills add https://github.com/kpiteira/devops-ai --skill kbabysit-kpiteira
## Review scope section, with dispositions of IMPLEMENT, PUSH BACK, DISCUSS, or OUT OF SCOPE, and out-of-scope items filed as issues. - Auditable reporting: Posts a Babysit report comment on the PR with per-round provenance columns, effort levels, push-back reasoning, and a merge-ready or needs-decision verdict. - Use Case: After opening a milestone PR, run /kbabysit 42 from the PR's worktree; the loop requests Copilot review, addresses real findings, pushes fixes, and stops after the round where all findings sit on prior fix commits, ending with a TL;DR report for the human. ## Quick Start Run /kbabysit followed by the PR number from a session checked out on that PR's branch to drive it to merge-ready.High-intent search queries and answers about installing and using this skill.
Run /kbabysit with the PR number from a session checked out on the PR's branch. The loop requests the Copilot review, waits for it, triages findings via kreview, pushes fixes, and repeats until a stop condition fires, then posts a report.
Use scope-bounded stop rules: end the loop when all line-anchored findings sit on previous fix commits (second-order round), when no in-scope implementable items remain, or when a round budget is reached. Unreviewed fix commits get a kselfreview pass instead of another paid round.
No. The review effort level (Lite or Balanced) is a repository or organization setting under Copilot code review settings, not a request-time parameter. The REST and GraphQL review-request APIs accept only reviewers, so the loop records the level per round but never changes it.
The Review scope section lists the outcomes the PR delivers, and every review finding is judged against it. Without it, anything in the diff is in scope by construction and the loop cannot distinguish real findings from out-of-scope noise.
The loop stops and reports a needs-human-decision verdict, since non-convergence within the budget means the disagreement is real. The default budget is 3 rounds, raised with the max-rounds argument, and all scope and provenance rules still apply.
basicmachines-coManages the full GitHub pull request lifecycle from branch hygiene through merge gating.
dbc-oduffyVerify tests and present three branch completion options.
dcossonIntegrate external review findings into project plans with disposition tables.
biz-StreamRetrieves GitHub PR review comments and sorts them by priority categories.