What is rocket-fuel-skill?

Rocket Fuel is a Claude Code skill that runs two different AI models together as if they were co-founders working on the same project. Claude Fable 5 plays the role of the Visionary: it interviews the user, writes the plan, sets the standards for the work, and reviews the finished result. OpenAI Codex plays the Integrator: it filters that plan down, pushes back on anything that looks like unnecessary scope, builds the actual code inside its own sandboxed environment, and reports back honestly on what it did. The idea is based on the Visionary and Integrator operating system described in the book Rocket Fuel by Gino Wickman and Mark C. Winters. Before any code gets written, the two models argue with each other in a structured, bounded meeting, and only once they reach a stated verdict of being on the same page does building actually begin. The user only makes three real decisions during a run: answering the interview questions, breaking rare ties if the two models get stuck disagreeing, and approving the final commit. The skill supports a few different modes depending on plain-English instructions. A full kickoff runs the whole interview, plan review, build, and verification cycle for a new feature. Telling it a repository is messy triggers a codebase audit that produces a ranked list of issues for the user to choose from before fixing. Pointing it at an existing plan file has Codex attack that plan without making changes until it approves it. A narrower request, like adding one specific feature, gets built by Codex and then checked by Fable. Every argument between the two models during a run is saved to a log file in the repository, alongside the plan file, so the reasoning stays visible after the fact. Installing it requires the Codex command line tool, Claude Code, and git. It is released under the MIT license.