Speculative Macro-Commit

A runtime acceleration mechanism for tool-using LLM agents that employs a dual-layer architecture consisting of an authoritative model and a draft model. It pre-executes multi-step action chains in an isolated environment and reuses high-frequency action skeletons via a macro library, breaking the bottleneck of sequential action-observation loops and reducing agent latency.