Human-Agent-Society/reef ships continual-learning infrastructure for self-improving agents under Apache 2.0

GitHub·low signalXBlueskyLinkedInCopy link

reef was created 2026-08-31 and reached 357 stars by 09-04 with daily pushes, tagged across continual-learning, self-improving-agents, llm-training, reinforcement-learning and agent-infrastructure. It lands in the same week as three arXiv papers on the same problem (HarnessDev, Environment Evolution, Repo-To-Skill), which is the notable part: the self-improvement loop is moving from papers to installable infrastructure. Single-source on capability claims since there is no accompanying writeup, so treat the description as unverified.