fivetaku/fablize — reverse-engineered prompt

Build me a Claude Code plugin called fablize that makes Opus, or really any Claude model, work more like Fable in the ways that actually transfer.

I want it to push the model to see tasks through to the end instead of stopping early or saying it did something when it has not. If the task has multiple stories or steps, it should break the work down, keep track of what still needs proof, and refuse to say done without evidence. If it is debugging, it should follow a real investigation flow, reproduce the problem, compare possible causes, and trace the actual chain of failure. If it creates something runnable or visual, like HTML, SVG, charts, or a small game, it should run it and check the result before claiming success.

Make it trigger with a fablize command and also support an always on setup with smart routing by task type. Keep the limits honest, if the model is at its ceiling, say so and suggest escalating instead of bluffing. Look up current Claude Code plugin docs if needed.