mrtooher/fable-mode

Fable Mode is a Claude skill, meaning a small reusable instruction set that Claude Code or a similar Claude environment can load and apply automatically, that pushes a model toward a more disciplined, staged way of working on large tasks. Instead of jumping straight into a big task and hoping it goes well, the skill makes the model first write out a stage plan, break the work down before acting. Where the environment supports it, the skill has the model delegate independent pieces of work to separate sub-agents rather than doing everything itself in one long sequence. At each stage, it requires a verification check that can genuinely fail, rather than the model simply feeling confident that its own work is correct. Before delivering a final result, it also asks the model to critique its own output skeptically instead of assuming the first draft is good enough. The author is explicit that this skill does not make the underlying model smarter. It shapes the procedure a model follows, not its raw reasoning ability. On a model that already handles long tasks coherently, the skill reinforces good habits it might already have. On a weaker model, it imposes structure the model would otherwise skip, but it cannot raise that model's actual reasoning ceiling. The README describes it plainly as a checklist, not a way to transplant capability from one model to another. The skill is meant to be triggered on tasks that span multiple files, multiple sources, or multiple work sessions, or when a user explicitly asks for a systematic approach. It is not meant for small tasks with one obvious solution, since forcing a staged process onto a trivial request would waste effort and bury a simple answer under unnecessary steps. The project consists of two files: the SKILL.md file that defines the skill itself, and an EXAMPLE.md file showing a worked before-and-after example where the built-in verification check catches an error that a single, unstaged attempt would have shipped. Installing it means placing the fable-mode directory wherever a Claude environment reads its skills from, then invoking it by name or letting it trigger automatically on a task that qualifies.