Reef Open-Sourced for Continual Agent Learning

Ao Qu announced the open-sourcing of Reef by the Human-Agent-Society team. The infrastructure turns user requests, agent trajectories, execution results, and feedback generated during deployment into experience data that updates models, prompts, memory, and skills. It includes recipes for SAO, OpenClaw-RL, TTT-Discover, SkillClaw, and Meta-Harness, and supplies the first fully open-source TTT-Discover implementation. Paul Liang and Zhaofeng Wu shared the post, describing the system as converting inference into a learning loop for general-purpose agents.