deploy

Overview

Post-merge deployment and monitoring skill. Picks up where /ship leaves off: merging the PR, verifying the deployment succeeds, and monitoring production health afterward.

When to Use

Run land once /ship pr produced a green, approved PR, and canary right after a deploy lands, whether this session merged it or not.

Subcommand Routing

Invocation	Action
/deploy or /deploy land	Merge PR and verify deployment (default)
/deploy canary	Post-deploy monitoring for errors and regressions

If no subcommand is given, default to land.

Process

1. Route the invocation per Subcommand Routing.
2. For land, run its Steps in order: identify, pre-merge checks, merge, wait for the deploy, health check, switch to local main.
3. For canary, run its Steps: baseline, monitor loop, per-interval report, final report, rollback suggestion on issues.
4. End with the Output block for land or the Canary Report for canary.

---

land

Merge a PR, wait for the deployment pipeline, and verify production health.

When to use

• After /ship pr created a PR and CI is green.
• When a PR is approved and ready to merge.

Arguments

• No arguments: detect the PR for the current branch.
• PR number or URL: target a specific PR.
• --squash: squash merge (default).
• --merge: create a merge commit instead.
• --no-delete: keep the branch after merge.

Steps

1. Identify the PR. If no argument, find the PR for the current branch:

   GH_TOKEN=$(gh auth token --user <account>) gh pr view --json number,state,reviewDecision,statusCheckRollup
   

   If no gh is available, try glab. Prefix with the appropriate token per standards/github-accounts.md or standards/gitlab-accounts.md.

2. Pre-merge checks. Verify all conditions are met:

   Condition	How to check	Blocker?
   CI is green	statusCheckRollup all passing	Yes
   PR is approved	reviewDecision is APPROVED	Yes
   No merge conflicts	mergeable is true	Yes
   Branch is up to date	Compare with base branch HEAD	Warning only

   If any blocker condition fails, stop and report the issue. Do not merge.

3. Merge the PR. Default to squash merge:

   GH_TOKEN=$(gh auth token --user <account>) gh pr merge <number> --squash --delete-branch
   

   If --no-delete is passed, omit --delete-branch. If this is a dual-base PR, same change targeting two branches, omit --delete-branch per ../../rules/git-workflow.md.

4. Wait for deployment. Detect the deployment mechanism:

   Signal	Platform	How to monitor
   .github/workflows/ with deploy job	GitHub Actions	GH_TOKEN=$(gh auth token --user <account>) gh run list --branch main --limit 1 --json databaseId,status, then poll a one-shot GH_TOKEN=$(gh auth token --user <account>) gh run view <id> --json status,conclusion every 60 seconds, at most 20 times, and every 3 minutes while the run is queued; GitLab: glab ci status -F json on the same schedule
   Vercel project	Vercel	Check Vercel MCP or vercel ls --limit 1
   Railway, Render, Fly	PaaS	Suggest user check the dashboard
   Manual deploy	Any	Ask user to trigger and confirm

   If no deployment mechanism is detected, ask the user how deployments work.

5. Verify production health. After deployment completes:

   • If a health endpoint is known, from project docs or user input, curl it and verify 200.
   • If the project has a public URL, verify it loads.
   • Check for new error reports if monitoring is configured, such as through the Sentry MCP.
   • Report the deployment result.

6. Switch to local main. After successful deployment:

   git checkout main && git pull origin main
   

Output

## Deployment Summary

**PR:** #<number> - <title>
**Merged:** <timestamp GMT>
**Method:** squash / merge commit
**Branch deleted:** yes / no

### Deployment
**Platform:** <detected platform>
**Status:** success / pending / failed
**URL:** <production URL if known>

### Health Check
**Endpoint:** <URL checked>
**Status:** <HTTP status code>
**Errors:** none / <error details>

---

canary

Monitor a recently deployed change for errors, performance regressions, and unexpected behavior.

When to use

• Immediately after /deploy land succeeds.
• When a deploy happened outside this session and the user wants a health check.
• When monitoring a risky change in the first minutes after deployment.

Arguments

• No arguments: monitor for 5 minutes with checks every 60 seconds.
• --duration <minutes>: monitor for a custom duration.
• --url <url>: target a specific URL for health checks.
• --baseline: capture current metrics as the baseline before a deploy, run this first.

Steps

1. Establish baseline. If --baseline was passed or no prior baseline exists:

   • Record current error rate from monitoring tools, if accessible.
   • Record response time for key endpoints.
   • Save as the comparison point.

2. Monitor loop. For each check interval:

   • Verify the health endpoint returns 200.
   • Check for new errors in monitoring tools, Sentry MCP if available.
   • Compare current response time against baseline.
   • Check browser console for JavaScript errors, if Playwright MCP is available and a frontend URL is known.

3. Report per interval:

   Check	Status
   Health endpoint	200 OK / <error>
   New errors	0 / <count and summary>
   Response time	<current> vs <baseline>
   Console errors	0 / <count>

4. Final report after monitoring completes:

   ## Canary Report
   
   **Duration:** <N> minutes
   **Intervals:** <count>
   **URL:** <monitored URL>
   
   ### Health
   **Uptime:** <percentage>
   **Errors detected:** <count>
   
   ### Performance
   **Response time (baseline):** <value>
   **Response time (current):** <value>
   **Delta:** <percentage change>
   
   ### Verdict
   <CLEAN: no issues detected / WARNING: <issues found>>
   

5. If issues detected: suggest rollback steps:

   • git revert <merge-commit> && git push
   • Or revert via the platform's rollback mechanism.

Rules

• Never merge a PR with failing CI checks. No exceptions.
• Never merge without approval unless the user explicitly requests it.
• Prefix every gh or glab command with the appropriate token per standards/github-accounts.md or standards/gitlab-accounts.md.
• All timestamps in GMT.
• Canary monitoring is non-blocking: report findings, do not auto-rollback without user confirmation.
• If rate limits are a concern for monitoring, use one-shot checks instead of continuous polling per ../../rules/git-workflow.md.
• Never access production databases or infrastructure directly. Use health endpoints and monitoring tool APIs only.

Common Rationalizations

• "statusCheckRollup is green, so CI passed": the rollup reads success when only an unrelated check ran. Confirm the CI workflows ran on the head commit with GH_TOKEN=$(gh auth token --user <account>) gh run list --commit <full-sha>.
• "The reviewer is away and the change is small": merging without approval needs the user's explicit request. Ask.
• "The canary showed errors, revert now to be safe": canary is non-blocking. Report the findings and the rollback steps, then wait for confirmation.
• "The dashboard is faster than the health endpoint": production databases and infrastructure stay off limits. Use health endpoints and monitoring APIs.
• "Delete the branch, it merged": a dual-base PR still needs its branch for the second merge. Omit --delete-branch.

Red Flags

• Stop and report when any pre-merge blocker row fails, and leave the PR unmerged.
• Stop and ask how deployments work when no row of the deployment detection table matches.
• Stop and present the rollback steps when the health endpoint returns anything other than 200 after the deploy.
• Stop and switch to one-shot GH_TOKEN=$(gh auth token --user <account>) gh run view checks when GH_TOKEN=$(gh auth token --user <account>) gh api rate_limit shows fewer than 500 remaining.

Verification

• GH_TOKEN=$(gh auth token --user <account>) gh pr view <number> --json state returns MERGED after land.
• The deploy run for the merge commit shows completed with conclusion success, or the user confirmed a manual deploy.
• The health endpoint returned 200, with the URL and status in the Deployment Summary.
• git branch --show-current prints main and git status reports the branch up to date with origin/main.
• After canary, the Canary Report carries a CLEAN or WARNING verdict with the interval count.

Related skills

• /ship pr -- Create the PR that this skill merges.
• /ship checks -- Diagnose CI failures before merging.
• /incident -- Document the incident if the deploy causes problems.
• /investigate -- Debug issues found during canary monitoring.