What problem does it solve? Turning agreed requirements into an implementation plan that an AI agent or developer can execute without stopping to ask questions is hard: plans written from imagination name wrong files, miss dependencies, and bury actions in prose. This Skill produces a detailed, codebase-grounded backend plan where every task traces to agreed acceptance criteria and names real files, classes, and patterns. ## Core Features & Use Cases - Right-sized requirements intake: Uses an existing approved PRD or bug investigation when present, runs a lightweight scoping pass for moderate changes, or captures a 2-3 line inline Goal & Acceptance block for simple ones. - Codebase mapping via subagents: Spawns parallel exploration agents to map domain modules, command handlers, controllers, GraphQL surfaces, migrations, and exact signatures before drafting anything. - DAG-structured task plan: Produces .ai/<TICKET>/backend-plan.md with a task manifest, dependency wiring, execution waves, atomic checkbox steps, and an intended frontend-facing surface. - Approval-gated review loop: Presents the manifest first, hands the document over for developer self-review, sweeps DAG neighbours for ripple effects after each edit, and finalizes only on explicit approval. - Use Case: After approving a PRD for a new reservations feature, ask for the backend plan and receive a task DAG covering the migration, value objects, command handlers, API endpoints, and events — each task independently executable by an implementation agent. ## Quick Start Ask the assistant to plan the backend for your ticket, for example: plan the backend for PHG-418 and break it into tasks.
