Hat	Stage	Upstream gate	Output	Unblocks
Orchestrator	Execution plan	PRD + SPEC + TDD + TSD all approved	repos/*/ai/fix_plan.md per repo + ralph/workspace-plan.md rollup	/ralph-dispatch

When to Use

All upstream artifacts are approved and user wants per-repo fix_plans so ralph can execute.

Hard Gate — Refuses Unless All True

For every PRD being planned:

1. PRD approved AND approved.json has PRD-{NNN}.
2. For every target repo with role=service or shared-lib: SPEC approved; if a TDD exists, TDD approved.
3. For every target repo with role=automation-tests: TSD approved and test-case set approved.

Any failure → print Gate failure. Missing approvals: {list}. Run wb.publish then wb.approve after review. and stop.

What This Skill Adds Over ralph-plan Alone

Feature	Source
Plan generation	ai-ralph native
Approval gate	this skill
Context routing via sync-context.sh	this skill
Workspace rollup ralph/workspace-plan.md	this skill

Protocol

1. Scope — ask PRDs (default: approved PRDs with empty fix_plan repos column in EPIC-PIPELINE.md), target repos (default: all), engine (default: DEVIN_DEFAULT).
2. Run adapter: ./scripts/ralph-plan.sh [--agent {engine}] [{repo}].
3. Aggregate per-repo fix_plans into ralph/workspace-plan.md rollup — PRDs planned, tasks per repo table, risks/blockers.
4. Update EPIC-PIPELINE.md — populate fix_plan repos, set Exec to ~.
5. Suggest next: /ralph-dispatch.

Do Not

• Bypass approval gate. “Urgent” is not a reason.
• Switch engines mid-run. Pick one; rerun if needed.

Source

skills/ralph-workspace-plan/SKILL.md