Read original ↗
repoGitHubTrust 82 · PrimaryPublished yesterdayLive · 20h ago

xianyu-sheng/Xenon

开源终端 AI 编程 Agent:以在线 Evidence Runtime 验证链(LLM 输出是 Claim、工具结果才是 Evidence)为核心。SWE-bench_Lite 官方评测 40.0% 实例级通过率(同模型 A/B +6.7pp,可复现报告)。7 种推理范式、MCP 工具、落盘补救与验证闭环。Python。

Lineage graph

Paper → model → repo connections mined from source citations (Tier-1 exact match).

Topics