Turn: A Language for Agentic Computation

This feed contains research on agentic programming languages and tooling and on prompt/system-prompt security for LLM-based agents. Key items: (1) Turn — a compiled, actor-based language and Rust bytecode VM for building autonomous LLM agents that enforces "cognitive type safety" (compiler-generated JSON Schema validation of model outputs), a confidence operator for flow control, an Erlang-like actor model with isolated context windows and persistent mailboxes, and a capability-based identity system to prevent raw credential exposure; (2) Arbiter — a framework that uses formal rules and multi‑