Deploy Pipeline
Deploy orchestration is intentionally staged. Hub should never hand a worker a vague “deploy everything” prompt when the next safe stage can be named.
Pipeline
Section titled “Pipeline”- Confirm Hub review passed.
- Confirm human acceptance when required.
- Commit accepted source work in the owning worker worktree.
- Push and prepare the merge path.
- Merge into the production branch.
- Update the canonical main/deploy worktree.
- Deploy from the validated main/deploy worktree.
- Verify production.
- Sync clean sibling worktrees.
Hub pauses for:
- destructive cleanup
- dirty unrelated work
- branch mismatch
- missing deploy target
- credentials or secret changes
- infrastructure creation
- production data writes
- ambiguous project ownership
Worktree Rule
Section titled “Worktree Rule”Implementation may happen in a worker-owned worktree. Production deploy should happen from the canonical main/deploy worktree unless a human explicitly approves a different path.
Evidence
Section titled “Evidence”Final deploy reports should name:
- repo and worktree
- branch
- commit
- command or workflow used
- target service
- live URL
- validation result
- follow-up risks