AgentSkillsCN

Ralph Plan

拉尔夫计划

SKILL.md

/ralph-plan — Generate or Refine the PRD

Usage

/ralph-plan [description]

Description

Analyzes the current codebase and generates or refines PRD.md tasks. Use this to break down a feature request into right-sized tasks for the ralph loop.

Instructions

  1. Read the current PRD.md, CLAUDE.md, and AGENTS.md.

  2. If a description argument is provided, generate new PRD tasks for that feature. If no argument, analyze the existing PRD for task sizing issues.

  3. For each task, ensure:

    • It can be completed in one context window (~one ralph iteration)
    • It has clear acceptance criteria
    • It lists expected files to touch
    • It follows the dependency order (foundational tasks first)
    • It touches no more than 5 files
  4. If tasks are too large, split them. If tasks are too small, consolidate them.

  5. Present the proposed PRD changes to the user for approval before writing.

  6. After approval, update PRD.md with the new/refined tasks.

Task Sizing Guide

Too smallRight-sizedToo large
Add one importImplement one function with testsBuild entire feature end-to-end
Fix a typoAdd one API endpointRefactor whole module
Rename one variableCreate one componentMulti-file integration

Model

sonnet