Skip to main content

Planning gates and waves

Feature-class delivery has three planning gates the user signs in order. Config key gates.require_planning_signoff_before_build keeps developers from starting early.

Order

  1. PM or intake writes the requirements artifact. Parent stops for @signoff:requirements.
  2. Parent applies architect-policy.md (or RUN_ARCHITECT). Large stories run Architect. Micro, minor, and jira-bug skip Architect.
  3. Architect writes mermaid architecture + implementation plan. Optional Archify HTML is extra; mermaid stays required.
  4. Parent stops for @signoff:architect (skipped when skip_architect).
  5. BA writes nested features/{slug}/{child}/specification.md, spec-order.md, test plan. BA critic, then @signoff:ba.
  6. Waves: per child, developer then developer-critic. Telemetry only if RUN_TELEMETRY.
  7. One tester at the parent slug. Devops only after qa-signoff.md has FEATURE_SIGNOFF: passed.
  8. Retro is the last step (gates.complete_after).

Clarify-first

PM, Architect, and BA read prior features/{slug}/ artifacts (decisions.md, plan, architecture) before asking. They ask remaining checklist items instead of silent defaults. Do not re-ask a decision already in decisions.md.

Do not

  • Auto-continue after PM, Architect, or BA SUCCESS
  • Spawn Architect or BA before signoff-requirements.md
  • Start waves before signoff-ba.md on feature class
  • Run tester per child, or between waves
  • Put sibling slugs at features/ instead of features/{feature}/{child}/

Machine contract for the next Task is JSON on disk, not a pasted HANDOFF. See slim handoffs.