weorbitant Deploying to Production
Overview
Deploy a merged PR to production by finding its Docker image tag from the CI build and triggering the manual deployment workflow via GitHub CLI.
Configuration
Read project-specific values from the project's CLAUDE.md (## Infrastructure section):
- Build workflow name, deploy workflow name, image tag input field name
The image tag is always the GitHub Actions run ID (numeric databaseId) of the build workflow run.
Process
digraph deploy {
rankdir=TB;
node [shape=box];
verify_merged [label="1. Verify PR is merged"];
find_image [label="2. Find image tag (build run ID)"];
check_deployed [label="3. Check what's currently deployed"];
confirm [label="4. Confirm with user"];
deploy [label="5. Trigger deployment workflow"];
verify_deploy [label="6. Verify deployment succeeded"];
has_migrations [label="PR has migrations?" shape=diamond];
run_migrations [label="Follow infra-migrations skill"];
done [label="Done"];
verify_merged -> find_image -> check_deployed -> confirm -> deploy -> verify_deploy;
verify_deploy -> has_migrations;
has_migrations -> run_migrations [label="yes"];
has_migrations -> done [label="no"];
run_migrations -> done;
}
See references/commands.md for all shell commands per step.
Common Mistakes
| Mistake | Prevention |
|---|---|
| Not reading CLAUDE.md | Workflow names and input field vary per project. |
| Passing full registry URL as image tag | Only pass the numeric run ID. |
| Deploying a PR that isn't merged | Always verify PR state first. |
| Deploying without confirmation | Always show what will be deployed and ask. |
| Confusing build run ID with deploy run ID | Build = CI on merge. Deploy = manual workflow. |
| Not checking what's currently in prod | Always show current state before deploying. |