canary

name	canary
description	Post-deploy production monitoring. Watches the live site after deploy — checks pages load, captures screenshots, detects console errors, compares against baselines. Use after /skill:land-and-deploy to verify production health.
allowed-tools	bash, read, write, grep, find
compatibility	requires pi-browse extension installed, git

Canary — Post-Deploy Monitor

Watch production after a deploy. Catch issues that pass CI but break in production — missing env vars, CDN cache serving stale assets, slow DB migrations on real data. Catch in the first 10 minutes, not 10 hours.

Brain Context Load

Before monitoring, search your brain for production context:

1. Extract keywords from the deploy URL, changed services, and recent changes.
2. Use gbrain_search to find past canary reports, production incidents, or known fragile endpoints.
3. Use gbrain_get to read the top 3 matches.
4. Use this context to focus monitoring on historically unstable areas.

If gbrain tools are not available or return no results, proceed without brain context.

---

Arguments

• /skill:canary <url> — monitor URL for 10 minutes
• /skill:canary <url> --duration 5m — custom duration (1m to 30m)
• /skill:canary <url> --baseline — capture baseline BEFORE deploying
• /skill:canary <url> --pages /,/dashboard,/settings — specific pages
• /skill:canary <url> --quick — single-pass health check

Phase 1: Setup

mkdir -p .gstack/canary-reports/baselines .gstack/canary-reports/screenshots

Parse arguments. Default: 10 min, auto-discover pages.

Phase 2: Baseline (--baseline mode)

Capture current state BEFORE deploying. Uses pi-browse tools:

For each page:

browse_goto → <page-url>
browse_snapshot(interactive: true)
browse_console(errorsOnly: true)
browse_screenshot → .gstack/canary-reports/baselines/<page-name>.png
browse_js "JSON.stringify({title: document.title, loadTime: performance.timing.loadEventEnd - performance.timing.navigationStart})"

Save baseline manifest:

cat > .gstack/canary-reports/baseline.json << EOF
{
  "url": "<url>",
  "timestamp": "$(date -u +%Y-%m-%dT%H:%M:%SZ)",
  "branch": "$(git branch --show-current)",
  "commit": "$(git rev-parse --short HEAD)",
  "pages": {
    "/": { "screenshot": "baselines/home.png", "console_errors": 0, "title": "..." }
  }
}
EOF

Then STOP: "Baseline captured. Deploy, then run /skill:canary <url> to monitor."

Phase 3: Page Discovery

If no --pages specified:

browse_goto → <url>
browse_links

Extract top 5 internal navigation links. Always include homepage.

Phase 4: Pre-Deploy Snapshot (if no baseline)

If no baseline exists (--baseline not previously run), capture one now as the post-deploy reference:

browse_goto → <url>
browse_snapshot(interactive: true)
browse_console(errorsOnly: true)

Phase 5: Continuous Monitoring

For each page, at each interval:

browse_goto → <page-url>
browse_snapshot(interactive: true)           → check page structure
browse_console(errorsOnly: true)             → any JS errors?
browse_is visible "body"                     → did page load?
browse_title                                 → correct title?

Alert triggers (immediately report as findings):

• Console error that wasn't in baseline
• Page fails to load (browse_goto timeout or error status)
• browse_is visible body returns false
• Page title changed unexpectedly
• Response status is 4xx or 5xx

Check interval:

• First 5 minutes: every 60 seconds
• Remaining duration: every 2 minutes

For --quick mode: single pass, no interval.

Phase 6: Health Report

After monitoring completes, generate report:

# Canary Report — {url}
**Deploy:** {branch} @ {commit} | **Duration:** {time} | **Pages:** {count}

## Health Score: {X}/10
- Page loads: {all clear / N failures}
- Console errors: {count} (baseline: {baseline-count})
- Performance: {avg load time}

## Findings
{list of alerts triggered during monitoring}

## Screenshots
| Page | Status | Screenshot |
|------|--------|------------|
| / | ✅ | screenshots/home.png |
| /dashboard | ✅ | screenshots/dashboard.png |

## Console

{console errors from last pass}

Save to:

REPORT_PATH=".gstack/canary-reports/$(date +%Y-%m-%d-%H%M)-canary.md"

Health Score calculation:

• Start at 10
• -2 per console error (new since baseline)
• -3 per page load failure
• -1 per unexpected title change
• Floor at 0

Phase 7: Baseline Update

If monitoring was clean (score ≥ 8), offer to update baseline:

• "All clear! Update baseline to this deploy? (recommended for next canary)"
• If yes: overwrite baseline.json with current state