What problem does it solve? Turning a feature spec or ticket into an executable implementation plan is error-prone: tasks end up too large, verification steps are vague, and the implementer (often a smaller AI model with no conversation history) gets stuck on ambiguity. This Skill produces a structured plan.md where every task has a Files manifest, literal verification commands, and exact expected outputs. ## Core Features & Use Cases - Task decomposition by layer: Splits work into Types, Data, Domain, UI, and Route tasks so storage, domain logic, and components never share a task, matching a Next.js App Router repo structure. - Contract-based steps with red-green verification: Each step carries an exact command (npx tsc --noEmit, npm run lint, npx playwright test) and its literal expected output, using compiler errors as the red signal since the repo has no unit-test runner. - Multi-pass review: Runs self-review, an independent coverage check, an adversarial critique for non-trivial plans, and a plan-document review via parallel subagents before saving. - Plan amendment and maintenance: Supports extending existing plans while preserving completed tasks byte-for-byte, with log.txt as append-only resumption state. - Use Case: Given features/005.record-expense.md, produce doc/features/005-record-expense/plan.md with 7 ordered tasks, each independently executable by a fresh implementer agent. ## Quick Start Ask the AI to write an implementation plan for features/005.record-expense.md and save it as plan.md in the feature's doc folder.
