Claude-Code-Workflow-Guide

Babysit PRs — Automated PR Queue Management

Boris Cherny uses /loop 5m /babysit-prs to handle PR management without manual attention. This workflow automates the tedious parts of maintaining a PR queue.

---

What “Babysitting PRs” Means

A PR that’s open but stale has a cost:

• Merge conflicts accumulate
• CI results expire
• Reviewers forget context
• Features sit unreleased

“Babysitting” means continuously monitoring and advancing PRs through their lifecycle — resolving CI failures, addressing review comments, rebasing, and merging when ready.

---

The /babysit-prs Command

<!-- .claude/commands/babysit-prs.md -->
# Babysit PRs — Monitor and Advance Open PRs

Check all open PRs and take appropriate action for each.

## Fetch Open PRs
```bash
gh pr list --state open --json number,title,statusCheckRollup,reviewDecision,mergeable,updatedAt

For Each PR, Take Action:

If CI is failing:

1. Fetch the failure details: gh pr checks [number]
2. Checkout the branch locally
3. Investigate the failure
4. Fix if the issue is clear and safe to fix
5. Push the fix
6. Comment on the PR: “Fixed CI failure: [brief description]”

If there are review changes requested:

1. Read the review comments: gh pr review [number] --comments
2. If the changes are clear and unambiguous: make them
3. If the changes require discussion: add a comment asking for clarification
4. Re-request review when done: gh pr review [number] --request-review

If approved and CI passes:

1. Check for conflicts: gh pr view [number] --json mergeable
2. If no conflicts: gh pr merge [number] --squash --auto
3. If conflicts: rebase and resolve

If stale (no activity in 5+ days):

1. Post a comment: “This PR has been inactive for [N] days. Is it still in progress?”

If draft:

• Skip — drafts are not ready for action

Report

After processing all PRs, output a summary:

• PRs merged: [list]
• PRs fixed: [list]
• PRs needing human attention: [list with reason] ```

---

The /loop Setup

# Run babysit-prs every 5 minutes
/loop 5m /babysit-prs

# Run every 30 minutes during business hours
/loop 30m /babysit-prs

# Run once (manual)
/babysit-prs

---

Safety Guards

Before setting up automated PR merging, add safeguards:

<!-- Add to babysit-prs.md -->

## Safety Rules (NEVER violate these)

- NEVER merge to main without at least 1 human approval
- NEVER merge if there are unresolved review threads
- NEVER force-push to shared branches
- NEVER merge PRs tagged with "do-not-merge" or "WIP"
- NEVER merge if the PR author explicitly said "don't merge yet"
- ALWAYS check that CI passes completely (not just required checks)

If any safety rule would be violated: skip the PR and add it to "needs human attention" report.

---

PR Health Dashboard

Add this to your session start or as a scheduled check:

# .claude/commands/pr-dashboard.md
# Show PR queue health

Generate a PR health dashboard:

```bash
gh pr list --state open --json number,title,author,createdAt,reviewDecision,statusCheckRollup,mergeable

Format as a table: | PR | Title | Age | Status | CI | Action Needed | |—|—|—|—|—|—| [populate from data]

Color coding (text):

• READY: approved + CI passing + no conflicts
• BLOCKED: has unresolved review threads
• FAILING: CI failing
• STALE: no activity in 3+ days
• CONFLICT: has merge conflicts ```

---

Integration with Notifications

Add a hook to notify you when babysit-prs takes significant action:

#!/bin/bash
# hooks/pr-notification.sh

# Called after babysit-prs completes
RESULT=$(cat)

MERGED=$(echo "$RESULT" | grep "Merged:" | wc -l)
NEEDS_HUMAN=$(echo "$RESULT" | grep "needs human attention" | wc -l)

if [ $MERGED -gt 0 ] || [ $NEEDS_HUMAN -gt 0 ]; then
  # macOS notification
  osascript -e "display notification \"$MERGED merged, $NEEDS_HUMAN need attention\" with title \"PR Babysitter\"" 2>/dev/null

  # Slack notification
  [ -n "$SLACK_WEBHOOK" ] && curl -s -X POST "$SLACK_WEBHOOK" \
    -d "{\"text\":\"PR Update: $MERGED merged, $NEEDS_HUMAN need your attention\"}" 2>/dev/null
fi