メニュー
Two-step deploy: merge approved PR to main, then deploy through promotion order. Enforces environment discipline.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Behavioral floor management. Propose changes, review proposals, apply approved changes, view status. Routes to COO as behavioral floor guardian.
Compliance program management. Propose changes, review proposals, apply approved changes, audit conformance, view change log.
Generic floor management. Routes to the declared guardian for any governance floor. Use /floor propose <floor-name> <change> to propose changes to any active floor.
Structured agent-to-agent handoff. Validates artifact, logs metrics event, dispatches receiving agent.
Run a retrospective for the last completed work item. Pulls DORA metrics, summarizes flow, captures keep/stop/start.
Interactive project setup: copy templates, configure fleet, define compliance floor, add first specialist agent.
| name | deploy |
| description | Two-step deploy: merge approved PR to main, then deploy through promotion order. Enforces environment discipline. |
| argument-hint | <env> [component] [--type planned|hotfix] [--merge <pr-number>] |
Two-step deployment as defined in .claude/COLLABORATION.md § Work Item Lifecycle (Phase 6) and § Branching and Pull Requests. The PO owns the merge decision; platform-ops owns the deployment execution.
/deploy dev -- Deploy merged changes to dev environment/deploy prod api --type planned -- Deploy API to production (planned)/deploy prod api --type hotfix -- Hotfix deployment/deploy --merge 42 -- Merge approved PR #42 to main, then deployParse arguments. Extract environment, component (optional), type (default: planned), and PR number (optional via --merge).
Validate source. If --merge specified, verify the PR has all reviewer approvals via GitHub MCP pull_request_read. If any reviews are pending or changes requested, halt with: "PR #X has unresolved reviews. Fix before merging."
Merge PR. If --merge specified, merge via GitHub MCP merge_pull_request. Log: ops/metrics-log.sh pr-merged <item> --pr <pr-number>. Delete the remote branch: git push origin --delete <branch-name>.
Switch to main. Run: git checkout main && git pull.
Environment discipline check. Verify the target environment is the next in the promotion_order from fleet-config.json. You cannot skip environments. If the previous environment hasn't been deployed to, halt with: "Deploy to first."
Pre-deploy checks. Dispatch the compliance-auditor agent to verify the compliance floor is satisfied for the work being deployed. If any blocking violations exist, halt and report.
Confirm with user. Show what will be deployed, to which environment, and the deploy type. Wait for user confirmation before proceeding. At Fly pace, confirm only for production deployments.
Execute deployment. Run:
ops/deploy.sh <env> <component> --type <type>
The deploy script's exit code determines success (0) or failure (non-zero).
Log and report. On success, run:
ops/metrics-log.sh ext-deployed <item> --env <env> --type <type>
Report the deployment result. If the deploy script output includes a URL or identifier, surface it. Recommend running regression tests for production deployments. On failure: report the error output, do not retry automatically, log a finding if the failure indicates a systemic issue.
| Subcommand | Model | Rationale |
|---|---|---|
/deploy | Sonnet | Orchestration with structured checks |
Implementers replace ops/deploy.sh with their deployment logic (contract: exit 0 = success, exit 1 = failure). The skill workflow stays the same. Override the full skill by creating .claude/skills/deploy/SKILL.md in your project to add environment-specific checks (e.g., k8s health probes, Vercel status) or custom merge validation.