🔥厉害了!仅需一行代码让 Agent 自己进化!FlashRSI 开源,RSI 研究终于不再“东一榔头西一棒”!


This is a dev post classified by AGK Intelligence as AI dev tools (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
🔥厉害了!仅需一行代码让 Agent 自己进化!FlashRSI 开源,RSI 研究终于不再“东一榔头西一棒”! 人民大学 NLPIR 实验室刚放了个大招:FlashRSI 首个真正统一的递归自进化(RSI)研究框架。 以前复现一个算法、换个 Benchmark、对个 Harness 要翻好几个仓库,现在直接Algorithm × Benchmark × Harness,一键跑通。 核心亮点: 🔹12 种 RSI 算法一键复现(SkillOpt / OPRO / GEPA / ACE / RRSI / Meta-Harness 等全覆盖) 🔹30+ Benchmark + 多类 Datasets,统一实验流 🔹原生支持 MiniAgent、Claude Code、Codex 三大 Harness,完整 Snapshot + 轨迹全程记录,可复现性拉满 🔹初步评估已做:in-domain / near-domain 收益明显,OOD 仍有挑战,限制演化空间有时反而更稳 3 步就能开跑: 1️⃣git clone https://github.com/RUC-NLPIR/FlashRSI.git && pip install -e ".[all]" 2️⃣设置 TARGET_API_KEY 和 WORKER_API_KEY 3️⃣flashrsi run --algo
Posted by SuSu_酥酥👅 (25.1k followers) 1 h ago · 11 likes · 1.2k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: flashrsi.git, flashrsi, Quickstart, harness2.
More dev work like this
- 𝗛𝗲𝗿𝗺𝗲𝘀 Agent "telegram not responding"? Telegram hides group messages from bots by… — @BkashJosi
- Someone found a rather embarrassing problem with traditional RAG. — @agenticgirl
- Alibaba is inviting selected developers to test qwen 4 models — @Mr_Salio
- DeepSeek-V4.1 Flash 1017 update on two DGX Sparks. — @jayleaton
- I open-sourced 43 agent skills for shipping real products. No coding. — @PawelHuryn
- Before my agents wrote any code, the plan review caught 5 bugs: broken saves, coins… — @btsouth
- 每次打开一篇 arXiv 论文,公式一多就看不下去了,丢给 AI 总结又只剩几句结论,概念还是似懂非懂。 — @GitHub_Daily
- 划重点:GLM-5.5将是一个RSI的产物! — @NFT_Chen
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.2k posts from 5k X accounts over the last 21 days, 8.9k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-10-31 14:10 UTC. Full method.