Render Unto Caesar

RUC lets Claude write and execute Python code snippets that call back to the LLM mid-execution, splitting work between deterministic loops and semantic judgment. You give it a task like triaging thousands of support tickets or classifying survey responses, and it generates code to handle iteration, state, and validation while delegating classification and summarization back to the model. Useful when you need LLM reasoning applied systematically across large datasets without the usual context limits, hallucinated completion, or retry hell. Runs in Docker via stdio transport. The pitch is opinionated but the architecture makes sense: code for control flow, LLM for the fuzzy parts, with clean handoffs between them.