Reproduce “Agent-Reflex: Teaching Language Agents to Act Reflectively” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Guanting Dong, Xinyu Yang, Jiajie Jin, Yuyang Hu, Xiaoshuai Song, Yifei Chen, Xiaoxi Li, Jiejun Tan and 4 more
Agent-Reflex trains tool-using agents to recover from tool failures and user corrections by mining environment-specific skills from successful and failed attempts, then using verified repair examples and skill-guided reinforcement learning without skill prompts at deployment. On five long-horizon benchmarks, its Qwen3-8B version scores 58.6 versus 50.9 for EnvScaler-8B; on out-of-distribution AgentNoiseBench, it scores 29.1 overall.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.