Deployment Ops — Edho Ferdian Mode

Boundary — read before doing anything else

This skill owns the interval between "the image builds" and "the release is observably healthy". Everything on either side belongs elsewhere:

Question	Skill that owns it
How do I structure the Dockerfile / compose file?	container-ops-edho-ferdian
Should this container run as root? Secrets in layers?	security-review-edho-ferdian
Is this page fast enough? What is the LCP budget?	performance-audit-edho-ferdian
Does this user journey still work?	e2e-testing-edho-ferdian
Is this deployed build broadly OK right now, once?	e2e-testing-edho-ferdian → qa-sweep.md
Should this be microservices? Which datastore?	system-design-edho-ferdian
Did I, the agent, actually finish the task I claimed?	safe-execution-edho-ferdian (delivery gate)

That last row matters: safe-execution's delivery gate is about agent claim discipline. references/production-readiness.md here is about application risk. They ask different questions and must not restate each other.

Workflow

Step 1 Pick the release strategy → references/release-strategies.md Step 2 Wire the pipeline gates → references/cicd-pipeline.md Step 3 (K8s only) write the manifests → references/kubernetes.md Step 4 Production-readiness verdict → references/production-readiness.md (fans out to 4 risk-lens agents in parallel on a harness that supports delegation — see that file's "Delegation, not duplication" section) Step 5 Deploy, then watch → references/post-deploy-watch.md Step 6 Make it operable → references/observability.md

references/cicd-pipeline.md (Step 2) covers gate ordering (lint → typecheck → unit test → security scan → build → deploy) with rationale, GitHub branch-protection required-status-checks setup, a full concrete GitHub Actions workflow with parallel gates and a branch/event-guarded deploy job, lockfile-hash-keyed caching, and the fail-fast-vs-fail-informatively distinction.

Language routing (fixed — see skill-authoring-edho-ferdian's canonical contract)

Communication to the user in Bahasa Indonesia; deploy/rollback commands, config, and status output in English — fixed, never ask. Full contract: skill-authoring-edho-ferdian §7.

Global rules

• Never claim a deploy is healthy from the deploy tool's exit code. The pipeline says "deployed"; only Step 5 says "healthy".
• A release without a written rollback path is not ready — this is a hard cap in Step 4, not a suggestion.
• Deploying to production is a destructive-class action: run safe-execution-edho-ferdian Gate 2 before executing one.