promote-to-prod

name	promote-to-prod
description	After a PR is merged, monitor CI/CD, approve PROD deploy, and validate
user-invocable	true
allowed-tools	["Bash","Glob","Grep","Read","Agent"]

Promote merged changes to PROD by monitoring CI/CD, approving the production deployment, and validating the live bot.

Prerequisites

• A PR has been merged to master
• The user should provide context on what was changed (ticket number, feature description)

Steps

Execute these steps sequentially. Stop and report if any step fails.

Step 1: Identify the Terraform Workflow Run

After a PR merge, the "Run Terraform" workflow triggers automatically. Find the latest run:

gh run list --workflow terraform.yml --branch master --limit 3 --json databaseId,status,conclusion,createdAt

Identify the most recent run that is in_progress or queued. Save the databaseId.

Step 2: Monitor Plan Phase

Poll the workflow run every 30 seconds until the plan jobs complete:

gh run view <RUN_ID> --json jobs --jq '.jobs[] | {name, status, conclusion}'

Wait for the plan jobs (names containing "plan") to reach completed status.

Step 3: Approve Production Deployment

The PROD apply job pauses for GitHub Environment approval. Poll for the pending deployment:

gh api repos/{owner}/{repo}/actions/runs/<RUN_ID>/pending_deployments

When a pending deployment appears for the production environment, approve it:

gh api repos/{owner}/{repo}/actions/runs/<RUN_ID>/pending_deployments \
  -X POST \
  -f environment_ids[]='<ENV_ID>' \
  -f state=approved \
  -f comment='Approved by Claude Code'

If no pending deployment appears after 10 minutes, check if the workflow auto-approved or if there's an error.

Step 4: Wait for Deploy to Complete

Poll the workflow run every 60 seconds until it completes. This typically takes 15-30 minutes:

STATUS=$(gh run view <RUN_ID> --json status,conclusion --jq '.status')

• If completed with success — proceed to validation
• If completed with failure — stop and report the failure. Use gh run view <RUN_ID> --log-failed to get error details.

Step 5: Validate PROD Bot

Message the PROD bot in Slack to validate the deployed changes work:

1. Send a test message relevant to the feature that was deployed
2. Wait 60-90 seconds for the bot to respond
3. Read the thread to verify the response is correct
4. Send a regression test (simple question like "What year is it?")
5. Verify the regression response

Use the Slack MCP tools (slack_send_message, slack_read_thread) for all interactions.

Step 6: Report

Report to the user:

• Workflow run ID and URL
• Deploy duration
• Validation results (what was tested, pass/fail)
• Any issues found

Notes

• The Terraform workflow deploys infrastructure AND triggers the worker container update
• If only Python files changed (no .tf changes), the Terraform plan will show "no changes" but the workflow still runs
• The worker container is deployed separately via the /deploy-worker-dev and /deploy-worker-prod skills