Just shipped an update that allows Pi prompt templates (slash commands) to run any command or script *before* the LLM. The model is only brought in when you actually need reasoning: `handoff: never` → run, done. No LLM latency. `handoff: on-failure` → only ask the model if the c…↗