Harness Engineering organizes memory, skills, and protocols to make LLM-based agents more efficient, even with model limitations.