Complete AI Training
Sign inGet started

Skill · Research

Gepetto

Produces a sectionized implementation plan from a markdown spec file through research, stakeholder interviews, spec synthesis, multi-LLM review, and section writing. Use when the user provides a spec file and wants a detailed implementation plan, plan review, or execution files for autonomous implementation.

Complete AI SkillsLicense: MITAdded Sep 29, 2026
HealthcareGovernment & Public SectorProduct ManagementExecutives & Strategy
Use it in my AI View on TemplatesGrokBot

How to use it

  1. Start your plan and connect your AI once
  2. Ask for the task in your own words, or say it directly:
Use the Gepetto skill to help me with this.

Without a connection: copy the SKILL.md below into your AI's project instructions.

SKILL.md

Implementation Planning from a Spec File

This skill turns a user-provided markdown spec into a thorough, sectionized implementation plan. It orchestrates research, stakeholder interviews, spec synthesis, plan generation, external review, and section writing, all within the planning directory derived from the spec file's parent folder. It is for users who need a self-contained plan an unfamiliar engineer or LLM can implement without extra context.

When to use

  • The user provides a markdown spec file path and asks for an implementation plan.
  • The user wants research and stakeholder interviews to clarify requirements before planning.
  • The user wants an existing plan reviewed by external models and the feedback integrated.
  • The user wants the approved plan split into self-contained section files.
  • The user wants execution files for autonomous implementation.
  • The user resumes a planning session and wants to continue from existing files.

Workflows

Research and interview

Inputs: The user's spec file path (must end in .md). Scan the planning directory (the spec file's parent folder) for existing files to determine whether to start fresh or resume.

  1. Read the spec file and extract potential research topics.
  2. Ask the user which codebase or web research they need.
  3. Launch parallel subagents to perform the requested research.
  4. After research completes, run a detailed interview with the user to clarify requirements.
  5. Save the interview transcript to claude-interview.md.
  6. On subsequent runs, detect existing files and resume from the appropriate step without re-asking.

    7. Check: claude-interview.md exists and covers the requirements clarified in the interview. Output: Research findings and claude-interview.md in the planning directory.

Spec synthesis and plan generation

Inputs: The initial spec, research findings, and interview answers.

  1. Combine the initial spec, research findings, and interview answers into a complete specification.
  2. Write the specification to claude-spec.md.
  3. Generate a detailed, self-contained implementation plan in claude-plan.md that an unfamiliar engineer or LLM can understand without additional context.

    4. Check: claude-spec.md and claude-plan.md exist and the plan is self-contained. Output: claude-spec.md and claude-plan.md in the planning directory.

External review and integration

Inputs: The generated plan in claude-plan.md.

  1. Launch two subagents (Gemini and Codex) in parallel to review the plan.
  2. Write their feedback to the reviews/ directory.
  3. Analyze the suggestions and decide what to integrate.
  4. Document decisions in claude-integration-notes.md.
  5. Update the plan accordingly.
  6. Present the updated plan to the user for approval before proceeding.

    7. Check: reviews/ contains both reviews, claude-integration-notes.md documents the decisions, and the plan reflects the integrated changes. Output: Review files, claude-integration-notes.md, and the updated claude-plan.md, presented to the user for approval.

Section creation and execution files

Inputs: The approved plan.

  1. Read the approved plan and identify natural section boundaries.
  2. Create a section index at sections/index.md with a SECTION_MANIFEST block.
  3. Launch parallel subagents to write each self-contained section file.
  4. Generate execution files claude-ralph-loop-prompt.md and claude-ralphy-prd.md embedding all section content for autonomous implementation.

    5. Check: sections/index.md contains the SECTION_MANIFEST block, each section file is self-contained, and both execution files embed all section content. Output: Section files under sections/, plus claude-ralph-loop-prompt.md and claude-ralphy-prd.md.

Tools and data

  • Use file system access to the planning directory when available; if the tool is not available, ask the user to provide the data or connect it.

Guardrails

  • Never execute code, make commits, or deploy anything.
  • Never send or publish any plan or document without user approval.
  • Only work with files in the planning directory derived from the spec file's parent folder.
  • Never invent research findings or plan details; base everything on actual research and user input.

Getting started

Print the intro banner, then ask the user to provide a markdown spec file path. Validate that the path ends with .md, then set up the planning session by scanning for existing files to determine whether to start fresh or resume.

Credits

Adapted from an open-source original (MIT): https://www.aitmpl.com/component/skills/ai-research/gepetto

More research skills

All research skills