Babysit Pr

---
description: Monitor a PR through bot reviews, triage threads, fix issues, and repeat until clean
argument-hint: "[optional PR number]"
version: 1.0.0
origin: agentive-starter-kit
last-updated: 2026-03-25
created-by: "@movito with planner2"
---

# Babysit PR

Monitor the current PR (or PR `$ARGUMENTS`) through bot review cycles.
Wait for bots, triage findings, fix issues, push, and repeat until the PR
is clean or the cycle limit is reached.

> **Shell rule**: Never use `$()` subshells in Bash calls — they trigger
> permission prompts. Run each command as a **separate Bash call** and capture
> values yourself.

## Configuration

| Setting | Value | Notes |
|---------|-------|-------|
| Max cycles | 10 | Prevents infinite fix loops |
| Bot wait timeout | 15 min | Per cycle, via `wait-for-bots.sh` |
| Bots tracked | CodeRabbit, BugBot | Both must be CURRENT before triage |

## Procedure

### Cycle Start

Track which cycle you are on (1 through 10). Report at the start of each cycle:

Babysit Cycle [N]/10 — PR #[number]

### Phase 1: Wait for Bots

Run the wait-for-bots script. This polls every 30 seconds until both bots
have reviewed the HEAD commit (or timeout after 15 minutes).

```bash
./scripts/core/wait-for-bots.sh $ARGUMENTS

• Exit 0: Both bots reviewed HEAD. Proceed to Phase 2.
• Exit 1: Timeout. Report which bot is still missing/stale. Stop babysitting — ask the user to investigate manually with /check-bots.

Phase 2: Triage Threads

Gather thread data (run these as separate Bash calls):

gh pr view --json number,url,headRefOid --jq '"PR #\(.number) | HEAD: \(.headRefOid)"'

./scripts/core/gh-review-helper.sh threads PR_NUMBER

./scripts/core/gh-review-helper.sh summary PR_NUMBER

Present the triage table for unresolved threads:

#	Bot	File:Line	Issue	Severity	Verdict
1	...	...	...	...	Fix / Resolve

Use the severity triage from the bot-triage skill:

• Fix: Real bugs, security, compatibility, reasonable improvements
• Resolve without fixing: False positives, cosmetic, platform-irrelevant

Phase 3: Decision Gate

If no unresolved threads (or all are resolve-without-fixing):

• Resolve any remaining trivial threads via GraphQL
• Report: "PR is clean after [N] cycle(s). Ready for human review."
• Stop babysitting.

If fixable threads exist AND this is cycle 10:

• Report the remaining issues
• Say: "Cycle limit reached (10/10). [N] threads still need attention."
• Stop babysitting — let the user decide next steps.

If fixable threads exist AND cycles remain:

• Proceed to Phase 4.

Phase 4: Fix and Push

1. Implement all fixes in a batch (read the affected files, make changes)
2. Run local checks: ./scripts/core/ci-check.sh
3. Commit with a descriptive message referencing the bot findings
4. Push to the feature branch
5. Reply to each fixed thread:

./scripts/core/gh-review-helper.sh thread-reply PRRT_thread_id 'Fixed in COMMIT_SHA: description.'

6. Resolve all addressed threads:

./scripts/core/gh-review-helper.sh resolve PRRT_node_id

7. Go back to Cycle Start (increment cycle counter).

Exit Conditions

The babysit loop ends when ANY of these are true:

1. All clean: No unresolved threads after bot review — report success
2. Cycle limit: 10 cycles completed with issues remaining — report what's left
3. Bot timeout: Bots didn't review within 15 minutes — ask user to investigate
4. CI failure: Local ci-check.sh fails after a fix — report the failure, stop

Final Report

Always end with a summary:

## Babysit Complete — PR #[number]

| Metric | Value |
|--------|-------|
| Cycles run | [N]/10 |
| Threads fixed | [N] |
| Threads resolved (no fix) | [N] |
| Threads remaining | [N] |
| Status | Clean / Needs attention / Bot timeout / CI failure |

[Next step recommendation]

If clean: suggest /preflight to run the full completion checklist before human review. If not clean: describe what remains and suggest next action. ```