Le Chen

LLM skills behave inconsistently across different agents, but SkVM fixes this by treating skills as code and LLMs as heterogeneous processors, leading to significant improvements in task completion, token consumption, and latency.