Hugging Face published a glossary by Sergio Paniego and Aritra Roy Gosthipaty clarifying AI agent terminology that has blurred amid rapid field evolution. The article distinguishes between scaffolding, the behavior-defining layer around a model, and harness, the execution layer that runs the agent. It also defines related terms including agent, context engineering, policy, tool use, skills, sub-agents, and training concepts like RL environment, trainer, rollout, and reward.
No score is assigned. Sources and their independence are shown in the citation chain below.