What problem does it solve? A PR is squash-merged into main with all CI checks green, yet the change never appears in production. The hidden cause is that the deploy workflow's if: gate (a required label like auto-deploy, or a path filter) silently skipped the run — and GitHub Actions renders conclusion=skipped with the same green checkmark as success, so nothing on the PR page reveals the deploy never happened. ## Core Features & Use Cases - Three-step diagnosis: Confirm the deploy run was skipped via gh run list, read the workflow's on:/if: gate to identify the label or path filter, and compare against a sibling PR that deployed successfully. - Recovery paths: Run the manual deploy script with a git-state preflight (including a verified detached-HEAD + CI=1 variant for worktrees and Cloud Run Job bakers), re-fire via workflow_dispatch if available, or add the label post-merge for audit trail. - Prevention layers: PR template checklists, subagent briefing text, required-status-check enforcement, and an actions/labeler@v6 auto-labeling workflow — plus the inverse trap where the labeler over-labels config/docs-only PRs and triggers unwanted redeploys. - Use Case: You merge a dashboard code PR, CI is green, but the live service is unchanged an hour later. The skill walks you through confirming the deploy workflow skipped for missing the auto-deploy label, then safely deploying the exact merge commit from a detached HEAD. ## Quick Start Ask the agent to diagnose why your merged PR with green CI is not live in production and to check whether the deploy workflow was skipped by a missing label gate.
