rlm — Recursive Language Model loop

A faithful instantiation of Recursive Language Models (Zhang, Kraska, Khattab; arXiv:2512.24601), Algorithm 1, on Claude Code's primitives. The paper's insight: an arbitrarily long prompt should not be fed into a model's context window at all. It should live in an environment the model interacts with programmatically, recursively calling a model over slices of it. That's what this skill does.

Mental model

You (the main Claude Code conversation) are the root model. You do not read the big context into this conversation. Instead: