Hugging Face publishes glossary clarifying AI agent terminology—harness, scaffold, and related concepts—to resolve confusion in a rapidly evolving field.

The post addresses terminology confusion in AI agents following ICLR 2026, where terms like 'harness' and 'scaffold' lacked consistent, shared definitions across explanations and frameworks.

Key distinction: scaffolding is the behavior-defining layer around the model (system prompt, tool descriptions, context management), while the harness is the execution layer that calls the model, handles tool calls, and decides when to stop. An agent = Model + Harness + Scaffolding together.

The glossary covers terms relevant whether building, deploying, or using agents (Claude Code, Codex, Hermes Agent), plus training-specific concepts like RL Environment, Trainer, Rollout, and Reward, acknowledging that different frameworks use the same words differently.