wtf is fable2opus_handoff_kit?

This project is a kit of prompts and document templates built for a specific situation: you have temporary access to an unusually strong AI model, and you want it to prepare a project so that a different, more commonly available model can keep running that project reliably afterward, without the strong model around to make judgment calls. The kit refers to the strong model as Fable and the successor model as Opus, since it grew out of an earlier project where that exact handoff happened once, informally, and worked well enough to generalize. The idea is to front load every decision that requires real judgment into written documents while the strong model is available, so the successor model never has to guess. Instead, the successor follows a fixed set of documents: a constitution style file laying out stages and rules, a single active execution plan with literal, checkable verification steps, a memory file that carries state between sessions, and a set of custom slash commands that turn every future work session into a repeatable, guided loop rather than an improvised one. Using the kit has three parts. First, you copy it into a new project and run nine sequenced prompts, roughly one per session, with the strong model. These prompts survey the project, interview you about risks and decisions, write the constitution and plan documents, build a verification script whose checks are proven to actually fail when something is wrong, and customize a set of generic command templates to the specific project. Second, both models overlap for a short burn in period, where the weaker model runs a few real sessions on the finished setup and the strong model audits what happened and patches any gaps. Third, once the strong model is no longer available, the weaker model operates the project alone using a single repeating command that reads the current state, decides what to do next through a fixed priority order, completes one task, and updates the memory file before ending. This is aimed at people who use multiple Claude models in Claude Code and want a structured way to hand off complex, unattended work between them.