---
description: Babysit — orchestration loop to monitor PRs, CI, and test evidence generation
type: orchestration
execution_mode: immediate
---

# /babysit

Alias to invoke the babysit orchestration and monitoring loop.

**Usage**: `/babysit [target PR, branch, or test suite]`

## Action

Execute these steps in order:

Read `.claude/skills/babysit/SKILL.md` for the full protocol, 7-gate criteria, and anti-patterns.

**Testing Gap Integration**: If at any point the target tests fail to produce compliant `/es` evidence bundles, immediately load and execute `.claude/commands/testing-gap-close.md` to identify and resolve the failure.

### Step 2: Discover All Open PRs

```bash
gh pr list --state open --json number,title,headRefName,headRefOid,mergeable \
  --jq '.[] | "\(.number)|\(.title)|\(.headRefName)|\(.headRefOid[:12])|\(.mergeable)"'

Step 3: Audit 7-Gate Status for Each PR

For each PR, check all 7 gates (see SKILL.md for exact commands). Categorize:

RED (3+ gates failing): needs code fixes → dispatch copilot-fixpr subagents
YELLOW (1-2 gates failing): needs bot triggers or comment resolution → fix inline
GREEN (6/7 gates): only Skeptic missing → trigger /smoke and /skeptic

Step 4: Dispatch Subagents to Fix RED PRs

Use Agent tool with subagent_type: copilot-fixpr, one per PR (or group related PRs). Run in parallel using run_in_background: true.

Each agent must:

Read the Green Gate log to find the exact failing gate
Fix the root cause (code, design doc, evidence, rebase)
Push the fix
Trigger bot re-reviews (@coderabbitai all good?)

Step 5: Fix YELLOW PRs Inline

Gate 3: Comment @coderabbitai all good? on PR
Gate 5: Resolve review threads via GitHub API
Gate 6: Add evidence or N/A justification to PR body
Gate 4: Bugbot errors → fix code, push

Step 6: Trigger Skeptic + Smoke on 6/7 GREEN PRs

# Trigger smoke
gh pr comment <NUM> --body "/smoke"

# Green Gate re-run (if no recent run)
gh workflow run green-gate.yml --ref <branch> -f pr_number=<NUM>

Step 7: Report Final Status

Print a table for ALL open PRs:

PR #<N> — <title> — status: <RED|YELLOW|GREEN|7-GREEN>
  Gates: 1=✓ 2=✓ 3=✗ 4=✓ 5=✓ 6=✓ 7=✗
  Action: <what was done or what's pending>

Step 8: Hold Loop — Stay Alive Until All PRs Merge

After Step 7, if any PRs are still open:

In /loop mode (user ran /loop /babysit): call ScheduleWakeup so the next iteration fires while the REPL is idle — never during an active conversation.

ScheduleWakeup(
    delaySeconds=180,  # 3 min — within cache window, catches Design Doc bot commits quickly
    prompt="<<autonomous-loop-dynamic>>",
    reason="babysit hold: re-checking stale Skeptic verdicts"
)

In one-shot mode (user ran just /babysit): set up a CronCreate job for the same re-check and tell the user it's running. Delete it with CronDelete once all PRs merge.

Each wakeup iteration runs only a lightweight stale-verdict sweep (see Phase 8 in SKILL.md) — not a full re-audit. Re-dispatch Skeptic only when both are true:

PR HEAD doesn't match the latest VERDICT: PASS SHA, and
CI is fully settled (ci_failures == 0 and ci_pending == 0 from statusCheckRollup).

Exit: Stop scheduling when no open PRs remain.

KEY RULES

PR Green Loop Protocol: Never push during active bot review. Batch fixes, push once, freeze.
Never dismiss CR CHANGES_REQUESTED — wait for re-review APPROVED
Never merge — only skeptic-cron merges when all 7 gates pass
7-green verification: Must read Green Gate gate-by-gate log, not just gh pr checks
Evidence required for non-docs PRs touching $PROJECT_ROOT/**