canary

Standalone post-deploy monitoring skill. Polls a target URL at regular intervals, checks for regressions against a baseline, and produces a report with evidence.

Invocation

Invocation	Action
/canary	Monitor the project's production URL for 10 minutes
/canary <duration>	Monitor for a custom duration, e.g., 5m, 30m, 1h
/canary <url>	Monitor a specific URL for 10 minutes
/canary <url> <duration>	Monitor a specific URL for a custom duration
/canary baseline	Capture current metrics as the pre-deploy baseline

---

Steps

1. Determine the target. Identify what to monitor:

   • If a URL argument is provided, use it directly.
   • If no URL: check the project for a production URL in package.json homepage, vercel.json, fly.toml, render.yaml, CLAUDE.md, or README.md.
   • If no URL is found, ask the user.

2. Determine the duration. Parse the duration argument:

   • Default: 10 minutes.
   • Accepted formats: 5m, 30m, 1h, 90s.
   • Check interval: every 30 seconds.
   • Calculate total check count: duration / 30 seconds.

3. Load baseline. If a baseline file exists at .canary-baseline.json in the project root:

   • Read the baseline: HTTP status, average response time, response body hash or key fields.
   • Report that a baseline is loaded and when it was captured.
   • If no baseline exists, the first check becomes the baseline for this run.

4. Baseline capture mode. If /canary baseline is invoked:

   • Run 5 checks at 10-second intervals.
   • Record: HTTP status, average response time, response body hash, content-type, response size.
   • Save to .canary-baseline.json in the project root.
   • Report the baseline and exit.

5. Monitoring loop. For each check interval:

   • Send an HTTP GET request: curl -s -o /dev/null -w "%{http_code} %{time_total} %{size_download}" <url>

   • Record: timestamp, HTTP status code, response time in milliseconds, response size.

   • If the baseline exists, compare:

     Metric	Regression threshold
     HTTP status	Any non-2xx when baseline is 2xx
     Response time	More than 2x the baseline average
     Response size	More than 50% change from baseline

   • If a regression is detected, log it immediately with the timestamp and evidence.

6. On regression detected. When a check fails the threshold:

   • Log the finding with timestamp, expected value, actual value.
   • If 3 consecutive checks show a regression, escalate: suggest a rollback command.
   • Rollback suggestion: git revert <last-merge-commit> && git push or the platform's native rollback mechanism.
   • If a webhook URL is configured in .canary-baseline.json under webhook, send a POST with the regression payload.

7. Final report. After the monitoring duration completes:

Output

## Canary Report

**Target:** <URL>
**Duration:** <actual duration>
**Checks:** <count completed>/<count planned>
**Baseline:** <loaded from file / first-run baseline / none>

### Health Summary
| Metric | Baseline | Current avg | Status |
|--------|----------|-------------|--------|
| HTTP status | <code> | <code> | OK / REGRESSION |
| Response time | <ms> | <ms> | OK / REGRESSION |
| Response size | <bytes> | <bytes> | OK / REGRESSION |

### Check Log
| # | Timestamp (GMT) | Status | Response time | Size | Notes |
|---|----------------|--------|--------------|------|-------|
| 1 | <HH:MM:SS> | <code> | <ms> | <bytes> | |
| 2 | <HH:MM:SS> | <code> | <ms> | <bytes> | REGRESSION: response time 2.5x baseline |

### Regressions
<List of all regressions detected with timestamps and evidence,
or "No regressions detected.">

### Verdict
<CLEAN DEPLOY: all checks passed within thresholds>
<REGRESSION DETECTED: N checks exceeded thresholds, rollback recommended>

### Suggested Rollback
(Only if regressions detected)
git revert <commit> && git push

Configuration

The .canary-baseline.json file stores baseline metrics and optional configuration:

{
  "url": "https://example.com",
  "captured": "2026-04-04T12:00:00Z",
  "status": 200,
  "avgResponseTime": 245,
  "avgResponseSize": 15234,
  "bodyHash": "sha256:abc123...",
  "webhook": "https://hooks.slack.com/services/..."
}

Rules

• All timestamps in GMT.
• Never auto-rollback without user confirmation. Report the regression and suggest the command.
• Rate limit awareness: 30-second intervals are the minimum. Never poll more frequently.
• If curl is not available, ask the user to install it. Do not fall back to other tools without confirmation.
• The monitoring loop is blocking. Report progress inline as each check completes.
• If the target URL requires authentication, ask the user for headers or tokens. Never guess credentials.
• Do not store authentication tokens in .canary-baseline.json. Store them in environment variables only.
• Prefix every gh or glab command with the appropriate token per ../../rules/github-accounts.md or ../../rules/gitlab-accounts.md.

Related skills

• /deploy land -- Merge a PR and trigger deployment before monitoring.
• /deploy canary -- Alternative canary implementation integrated with the deploy flow.
• /incident -- Document an incident if regressions are confirmed.
• /investigate -- Debug the root cause of a regression.