wtf is fables?

Fables is a pair of plugins for Claude Code, Anthropic's coding assistant tool, built around a single idea: the most capable AI model, called Fable 5, is also the most expensive one, so it should only be used at the moments when its judgment actually matters. The two plugins point in opposite directions but share that same underlying logic. The first plugin, fable-advisor, is meant for sessions running on a cheaper or mid tier model. It teaches that main model to occasionally consult Fable 5, but only at specific turning points where a decision would be costly to reverse later, such as choosing a database schema, an API contract, or a system architecture. Rather than dumping the whole conversation into Fable, the main model prepares a short, focused briefing and asks for a terse verdict, capping how many times it can consult Fable per task. The idea is that a disciplined consultation costs a small amount of money, while a wrong architectural decision made without one can cost many hours of rework. The second plugin, fable-orchestrator, is meant for sessions where Fable 5 itself is doing the main planning. Instead of having the expensive model read through large amounts of text directly, that bulk reading and searching work gets pushed down to cheaper worker models running in parallel, and only their distilled findings come back up to Fable. This way, Fable spends its more expensive tokens on planning and decisions, not on grinding through raw information. Both plugins are installed through Claude Code's plugin marketplace system with a couple of short commands. The project also includes benchmark results showing how reliably each plugin's built in triggers fire in practice, along with example prompts and evaluation scenarios that were used to test and refine the behavior during development.