Hi everyone, I’m testing a small pre-inference controller for long-context LLM / agent sessions, and I’d love feedback from people building LangChain-style agent workflows. The idea is to run a lightweight controller before the model call and route each input into one of: FAST: answer with compact reference context HOLD: wait / ask for clarification BLOCK: prevent contradictory or unsafe execution DEEP: run deeper reasoning RECOVER: restore long-context consistency Why I built it: In long-...