> workflow / production

Production

Operate the live release with go/no-go discipline, progressive exposure, verified rollback, and customer communication readiness.

What should I do next?

Complete the steps and evidence below, clear the human gates, then continue to Continuous Improvement.

When to use

  • Stage validation and monitoring are ready
  • You are executing a production deploy or GA ramp
  • Hotfix path may be needed under incident pressure
  • Compliance needs a release evidence bundle

Step-by-step

  1. Run go/no-go against checklist: tests, security, perf, monitoring, rollback drill.
  2. Deploy via the release pattern (canary/flags); watch abort metrics.
  3. Keep hotfix-pattern ready for containment without large rewrites.
  4. Publish release notes and customer/internal comms as planned.
  5. Capture evidence bundle (what shipped, gates, approvers).
  6. If abort criteria hit, roll back immediately — do not "push through".
  7. Hand off to continuous-improvement with open actions from the release.

Decisions to make

  • Go / no-go / abort
  • Ramp percentage and hold duration
  • Hotfix vs rollback vs forward fix
  • Customer notification timing and channel

Evidence to collect

  • Deployed release with artifact digest
  • Go/no-go record and approvers
  • Release notes and comms sent
  • Evidence bundle for compliance/audit

Role-specific next actions

  • PM: Own customer comms and go/no-go business call.
  • Architect: Advise abort when boundary or data risk appears mid-ramp.
  • Engineer: Execute deploy steps; stay on hotline through ramp.
  • DevOps: Drive pipeline, canary, and rollback mechanics.

Inputs

  • Release candidate and pipeline
  • Runbooks and abort metrics
  • Comms plan and on-call coverage

Outputs

  • Production release
  • Release notes
  • Evidence bundle and open follow-ups

Required prompts

  • Release

    Go/no-go structure, notes, and rollback criteria.

  • Documentation

    Finalize operator and customer-facing docs for what shipped.

  • Incident

    If the release degrades — structured response.

Required skills

  • DevOps

    Production delivery and rollback discipline.

  • Documentation

    Accurate release and operator docs.

Required MCP servers

  • GitHub MCP

    Release tags, deploy PRs, and status checks.

  • Kubernetes MCP

    Live cluster inspection during ramp when applicable.

  • Sentry MCP

    Watch error burn during canary.

  • Slack MCP

    Optional incident/release channel coordination.

Tooling & handbook

Recommended agents & patterns

Human approval gates

  • Go/no-go with rollback verified before ramp
  • Abort authority named and exercised if metrics trip

Automation opportunities

  • Canary deploy automation with metric abort
  • Feature flags for progressive exposure
  • Evidence bundle generation from CI + approval records

Enterprise best practices

  • Evidence retained for compliance audits
  • Customer comms templates approved by legal/support where required
  • Change records linked to SoA when applicable

Common mistakes

  • Big-bang deploys without canary or flags
  • Skipping rollback drill
  • Pushing through abort criteria to "make the date"