andyshaman/senior-fable

Senior Fable is a plugin for Claude Code that changes how work gets split between different AI models during a coding session. The idea is to treat your most capable, most expensive model as a tech lead: it handles planning, architecture decisions, contested calls, and pulling results together, while cheaper models handle the routine work. It routes tasks based on what kind of work they are. Simple, mechanical jobs with a clear spec, like writing tests, formatting, or boilerplate edits, go to a faster and cheaper model called the fast-worker. Long, messy investigations, like digging through a large codebase or grinding through logs to find one bug, go to a separate read-only model called the deep-reasoner, so that exploration does not clutter the main conversation. If a decision is risky or contested, and an additional tool called Codex is installed, that decision can get a second opinion from a different model family. Several rules keep this delegation predictable rather than chaotic. The deep-reasoner is blocked from editing or writing files at the tool level, so it can only investigate and report back. The fast-worker only has access to six tools, with no web access and no ability to spawn its own subagents. Every task handed off follows a fixed template covering the goal, what files are in and out of scope, constraints, and how success is reported, and a result that does not match this format is rejected rather than accepted as done. Installing it is done through the Claude Code plugin marketplace with two commands, after which you switch to the top-tier model and activate the plugin with a slash command. It only makes sense to use on sessions already running the strongest available model, since delegating work to a stronger model would defeat the purpose. This is aimed at developers who use Claude Code regularly and want to reduce token usage and context clutter without giving up the quality of a top-tier model's decisions.