wtf is claude-fable-5-spanking-claude-opus-4.8?

This project is a skill file for Claude Code, the command line coding assistant, that changes how work gets split between two different AI models to save money. It is a set of instructions rather than a running program. The idea comes from a difference in how AI usage gets billed. A frontier model running as the main assistant, in this example called Fable 5, charges for every token it produces against a metered account, while smaller subagent models it spawns to do work are billed instead against a flat subscription plan. Since writing new code, documentation, and tests produces a lot of output tokens, and that is exactly the expensive kind of usage, the skill's rule is that the expensive main model should read, plan, and judge work, but never write it itself. Under this skill, the main assistant keeps four jobs for itself: breaking a task into pieces with clear instructions for each one, checking and approving the work that comes back, making decisions and talking to the user, and making small quick edits where writing out full instructions would take longer than just doing it. Everything else, including first drafts of code or user facing text, goes to subagents running a different, plan covered model, in this example Opus 4.8. Checking the returned work follows a specific order: free checks like linters and tests first, then a close read of only the most sensitive parts, then a full review only for the most important or customer facing code. If a subagent's work does not meet the bar after one round of revisions, the main assistant redoes that piece itself rather than accepting weaker work to save money. Installing it means copying a single file into a folder inside your Claude settings, either for all your projects or just one, then calling it by name at the start of a session. It is released under the MIT license.