The harness provides a means for the LLM to write file output and run tools (commands) to test output. Alternative harnesses including things like OpenCode.