An Agent Harness Is Not a Security Boundary: Reading AgentSmith as a Workflow Layer

An AI Agent Harness is the software infrastructure surrounding a Large Language Model (LLM), enabling it to interact with the outside world via tools, memory, and execution environments. This allows LLMs to perform multi-step workflows and execute code. AgentSmith is a new open-source agent harness addressing composition problems. Its repository includes a shared core, work-type profiles, and hooks, designed to separate stable team guidance from role-specific instructions to reduce workflow drift during coding tasks.