What is fable-mode?

Fable Mode is a skill for Claude Code that tries to make a smaller or different AI model behave with the same working discipline as Claude Fable 5, a specific frontier model from Anthropic. Instead of changing the underlying model, it loads a set of rules taken from Fable 5's own instructions and applies them to whatever model you are actually running. Those rules cover seven areas: judgment, planning, delegation, checking work, reasoning, how it communicates, and how carefully it writes code. Once you turn it on with the command slash fable-mode, or by simply typing use fable mode, the model starts stating its assumptions out loud before it writes code, breaks multi-step work into pairs of a step followed by a check, and verifies changes by actually running the new behavior rather than just trusting that the code built without errors. It also reports failures honestly instead of glossing over them, runs a self critical review before saying a task is finished, hands off bulky repetitive work to cheaper helper models to keep its own attention focused, and puts the outcome first when it answers. You can install it as a Claude Code plugin through the plugin marketplace command, copy it manually into your personal skills folder, or use the packaged skill file included in the repository for tools that accept uploaded skills, such as claude.ai's settings page. The README suggests using it for models weaker than Fable 5 on work where a wrong answer is costly, for long multi-step tasks like refactors and migrations, for debugging sessions, and for anything touching configuration, infrastructure, or git history you care about. It suggests skipping it if you are already running Fable 5, for trivial one-line questions, or notes that very small models can follow the process but will not fully manage the harder parts like the self critical review. The rules are adapted from a copy of the Fable 5 system prompt maintained by another GitHub user, credited in the README. The project states it is not affiliated with or endorsed by Anthropic, and it is released under the MIT license.