gepetto_skill

Overview

This skill orchestrates a structured, multi-step implementation planning workflow that produces sectionized, self-contained deliverables for engineering and autonomous execution. It combines research, stakeholder interview, spec synthesis, external multi-LLM review, and parallel section writing to produce a complete plan and execution artifacts. Use it to turn a markdown spec into a reproducible implementation roadmap and autonomous-run prompts.

How this skill works

On invocation the skill validates a supplied .md spec file and initializes a planning directory. It walks through staged steps: decide and run research, conduct a detailed interview, synthesize a formal spec, generate a self-contained implementation plan, run parallel external LLM reviews, integrate feedback, create a SECTION_MANIFEST and parallel section files, and finally produce two execution outputs for autonomous tools. Each step writes clear files into the planning directory so engineers or agents can continue independently.

When to use it

• Planning a feature that needs thorough pre-implementation analysis and traceability.
• Turning a vague product idea or short spec into a full engineering-ready plan.
• Coordinating multi-stakeholder input and external LLM reviews before development.
• Preparing deliverables for autonomous execution tools or CI-driven implementation.
• Resuming or iterating on an existing plan stored as a markdown spec file.

Best practices

• Always invoke with an explicit .md spec file path; the tool stops if none is provided.
• Keep the initial spec focused on intent and constraints; gepetto handles synthesis and gaps.
• Allow research and interview steps to run fully—findings significantly improve plan quality.
• Review claude-plan.md after integration steps before authorizing autonomous execution.
• Use the SECTION_MANIFEST and per-section files to assign work and verify acceptance criteria.

Example use cases

• Converting a product brief into a developer-ready PRD with sectioned implementation tasks.
• Onboarding a new team to a feature by providing self-contained section files with acceptance checks.
• Running multi-LLM external reviews to catch architectural or security blind spots before coding.
• Generating ralph-loop and Ralphy-ready prompts to automate stepwise implementation.
• Resuming paused planning work by re-invoking the skill on the same spec to continue from the last step.