Cookbook: Release Flow

This recipe models a safe, profile-aware release with retries and explicit unsafe controls.

Scenario

You want one release command that is strict in production but still testable locally.

qp.yaml Recipe

profiles:
  _default: "{{env.QP_PROFILE}}"
  staging:
    vars:
      deploy_env: staging
  prod:
    vars:
      deploy_env: prod
    tasks:
      deploy:
        when: branch() == "main"
        timeout: 30m

vars:
  deploy_env: local

tasks:
  build:
    desc: Build release artifacts
    cmd: make dist
    safety: idempotent

  publish:
    desc: Publish artifacts
    cmd: ./scripts/publish.sh --env {{vars.deploy_env}}
    safety: external
    retry: 3
    retry_delay: 5s
    retry_backoff: exponential

  deploy:
    desc: Deploy release
    cmd: ./scripts/deploy.sh --env {{vars.deploy_env}}
    safety: external
    retry: 2
    retry_delay: 10s

  release:
    desc: Release pipeline
    run: build -> publish -> deploy

Run It

Staging rehearsal:

QP_PROFILE=staging qp release --allow-unsafe --events

Production:

QP_PROFILE=prod qp release --allow-unsafe --events --json

deploy only runs on main in prod profile due to task-level profile override.

Guarding Releases

Add a pre-release validation guard:

guards:
  release-ready:
    steps: [check, build]

Then run:

qp guard release-ready --json && QP_PROFILE=prod qp release --allow-unsafe

Why This Works Well

1. Local/staging/prod share one graph with profile overlays.
2. Publish/deploy retries absorb transient failures.
3. Unsafe side effects require explicit opt-in.
4. Event stream gives auditable release timeline.