babysit-pr
Skill272 starsChanged 16 days agoMITView on GitHub
Monitor an existing GitHub pull request through CI and review, fix valid findings, and reply to review comments. Use when the user asks to watch, monitor, or babysit a PR, or handle its review comments. Publishing alone does not trigger this workflow; merge only when explicitly requested.
65 other projects use a version of this.
Adopted by
Most recently taken up by
- tekumara/skillstheir own version, first seen Nov 16, 2026
- zanellig/skillstheir own version, first seen Nov 16, 2026
- arraydude/agent-skillstheir own version, first seen Nov 16, 2026
- rodeyseijkens/skillstheir own version, first seen Nov 16, 2026
- tunnckoCore/skillstheir own version, first seen Nov 16, 2026
- gmunaro/claude-skillstheir own version, first seen Nov 16, 2026
Used with
Projects with it also use leaving-pr-comments, sol-loop and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
cogine-aiSkillleaving-pr-comments
Use whenever leaving comments on GitHub.
Also in 12 projects that use this
tekumaraSkillsol-loop
Use when the user says "sol loop".
Also in 10 projects that use this
tunnckoCoreSkillfull-stop
Trigger this if the user says 'full stop', even as a closing "Full stop." that reads like emphasis.
Also in 13 projects that use this
cogine-aiSkillmodernize
For getting old prs up to date. Use when asked to rebase a pr or "modernize".
Also in 10 projects that use this
tekumaraSkillfile-pr
Raise a concise pull request. Use when the user asks to raise, open, or create a PR.
Also in 17 projects that use this
Solly922Skillhtml-communication
Create self-contained HTML artifacts for plans, architecture maps, comparisons, reports, timelines, diagrams, and technical explainers when visual structure improves understanding. Use for complex or spatial information; do not use for short, linear answers that are clearer in Markdown.
Also in 3 projects that use this
- +1 more from tekumara
- +1 more from cogine-ai
Other versions65 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 65
- BuilderIO/agent-native★ 7.7ksame name
- camunda/camunda★ 4.4ksame name
- t3dotgg/tstack★ 3.8ksame name
- UniClipboard/UniClipboard★ 2ksame name
- rome-os/rome★ 904same name
- trailofbits/coop★ 816same name
- Mr-Quin/danmaku-anywhere★ 551same name
- amElnagdy/review-skills★ 170same name
- owid/etl★ 166same name
- jellydn/my-ai-tools★ 129same name
55 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- babysit-prSkill
- publish-prSkill
- release-notes-discordSkill
- release-notesSkill
- publish-prSkill
- release-notes-discordSkill
- release-notesSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. 0 are common, 12 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 63 edited versions we read keep these.
12
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Read AGENTS.md and the architecture rules relevant to any fixes.
- Distinctive: line 19
Read the live PR state, latest head SHA, checks, review submissions, top-level comments, and inline review threads, including resolved …
- Distinctive: line 32
Verify each actionable finding against the current source and repository rules.
- Distinctive: line 33
Read any cited review configuration.
- Distinctive: line 41
Run checks appropriate to each fix, commit, and push to the PR's branch.
- Distinctive: line 49
Check replies, quoted text, and attachments for private details before posting.
- Distinctive: line 50
Verify the returned reply URL before resolving a thread.
- Distinctive: line 55
Do not dismiss human reviews or treat a reply as satisfying required approval.
- Very specific: line 59
Use one supported watcher per PR.
- Distinctive: line 64
Only the thread owning the PR starts a watch.
- Very specific: line 72
Call unwatch_pull_request before a final handoff, blocker report, or question requiring user input.
- Distinctive: line 82
Before declaring readiness, fetch the base branch and reread the live PR.
- Distinctive: line 94
Stop when ready, when only human review remains, when a decision or external blocker prevents progress, or when the PR is merged or closed.
- Distinctive: line 100
Leave the PR open and unmerged by default.
“Common” is measured: kept by at least 6 in 10 of the 63 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Babysit a Pull Request
Bring the PR to a verified handoff by handling CI failures and review feedback within the user's requested scope. A babysitting request includes making, validating, committing, and pushing relevant fixes, replying on the user's behalf, and resolving addressed bot threads. If the user requests observation only, report findings without changing code, posting replies, or resolving threads.
Start or Resume
- Read
AGENTS.mdand the architecture rules relevant to any fixes. Confirm the target PR, repository, head branch, and base branch. Inspect local changes and preserve unrelated user work; never commit directly tomain. - Read the live PR state, latest head SHA, checks, review submissions, top-level comments, and inline review threads, including resolved state. Fetch all pages. Comments alone do not show whether a thread is resolved or approval is pending.
- Inspect existing unresolved feedback before starting a watcher. Reassess older comments against the current code rather than filtering by the latest push time.
- On resume, use GitHub's current state and existing replies to recover progress. Track comment/thread IDs, reply URLs, and the head assessed to avoid duplicate replies or repeated fixes. A new head requires reassessing affected findings.
- In T3 Code, link the PR with
link_pull_requestimmediately after taking it over or pushing to it, following the host's linking rules.
Handle Findings
- Verify each actionable finding against the current source and repository rules. Read any cited review configuration. Treat bot findings as claims to investigate.
- Fix real bugs and CI failures within the intended PR scope. Avoid unrelated refactors or speculative compatibility work. Explain declined suggestions with evidence; report useful out-of-scope work as follow-ups without creating issues unless requested. Escalate decisions that materially change the user's goal.
- Inspect failing-check logs and distinguish code failures from infrastructure failures. Retry only when there is evidence of a transient failure; report persistent infrastructure blockers rather than repeatedly rerunning checks.
- Run checks appropriate to each fix, commit, and push to the PR's branch. Keep validation claims accurate, including the distinction between automated, Storybook, and actual iRacing testing. Refresh the live head and checks after pushing; checks or reviews for an earlier commit do not prove the new head.
- Reply in the original thread with the fixing commit and relevant validation, or a concrete reason the finding does not apply. Lead with the conclusion, cite code or observed behavior, and distinguish verified facts from judgment. Avoid filler, repetitive status comments, and unsupported claims of success.
- Check replies, quoted text, and attachments for private details before posting. Verify the returned reply URL before resolving a thread. After a failed or ambiguous post, read existing replies before retrying to avoid duplicates.
- Resolve bot threads only after the fix is pushed and the reply succeeds, or an evidence-backed explanation has been posted for a declined finding. Leave human threads open for the reviewer unless the user explicitly asks to resolve them. Do not dismiss human reviews or treat a reply as satisfying required approval.
Wait for Events
Use one supported watcher per PR. Do not build a polling script or use sleep
loops, gh --watch, or repeated status calls to wait.
In T3 Code:
- Only the thread owning the PR starts a watch; delegated reviewers never do.
- Handle existing checks and feedback, finish needed fixes and pushes, then call
watch_pull_requestand end the turn. Only subsequent events trigger a wake. - The watch persists across wakes. On each wake, read current PR state, handle actionable changes, and assess readiness again. Do not restart an active watch or infer readiness solely from the wake message.
- While checks or automated reviews remain pending, keep the watch active and end the turn. Avoid repeated messages when an event needs no action.
- Call
unwatch_pull_requestbefore a final handoff, blocker report, or question requiring user input. A merged or closed PR ends monitoring.
Outside T3 Code, use an existing supported watcher if available and follow its resume/stop behavior. If none is available, perform a current-state review and report that ongoing monitoring could not be started. Do not claim to be watching. On rate-limit or authentication failures, report the blocker without retry loops.
Readiness and Handoff
Before declaring readiness, fetch the base branch and reread the live PR. Confirm:
- The PR is open and ready for review. Preserve an explicitly requested draft and report it as a draft rather than ready to merge.
- Required checks pass for the latest head. Configured automated reviews that apply to this PR have completed for that head; do not wait for absent bots.
- Actionable findings are addressed or declined with written evidence. Pending human approval, unresolved human threads, or requests for changes are identified.
- GitHub reports the PR as mergeable, and the branch meets the repository's base freshness requirements. Inspect new base commits that overlap the PR's files; update and validate the branch when needed before declaring it ready.
Stop when ready, when only human review remains, when a decision or external blocker prevents progress, or when the PR is merged or closed. Unwatch before handing back. Report the PR URL, final head SHA, validation performed, feedback addressed, and any remaining approvals or blockers. Distinguish ready to merge from waiting for human review or blocked.
Leave the PR open and unmerged by default. Merge, enable auto-merge, or enter a merge queue only when the user explicitly requests it. Do not close the PR or bypass branch protection as part of babysitting.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

