[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-baai-disco-repo-to-skill-library":3,"news-related-089195fb-7fe5-4ba9-a4bc-8e356fe5e923":38},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"089195fb-7fe5-4ba9-a4bc-8e356fe5e923","BAAI把1000个GitHub仓库蒸馏成5000个技能,科研agent奖牌率31%冲到73%","BAAI论文提出DisCo:会自建并使用技能的科研agent,把1000个常用ML仓库蒸馏成5000+验证过的可执行技能(AREX-Skill库,Apache 2.0)。固定底座与预算的对比中,带技能的agent在MLE-bench任意奖牌率从31.11%升到72.89%,PaperBench等三项基准也同步提升。","自主科研 agent 的能力上限,通常被归结为两件事:模型底座够不够强,harness(规划、执行、记忆、验证的外围工程)写得好不好。BAAI 团队 9 月 2 日挂在 arXiv 上的论文 Repo-To-Skill 提出了第三块拼图:操作知识(operational knowledge)——「知道一个方法」和「把一个方法跑通」之间那段 know-how。这些知识并不缺,它们躺在仓库和论文里,只是写给人读,而且大到没法在任务执行中现场加载。论文的解法,是把它们蒸馏成紧凑、经过验证、可复用的技能。\n\n## DisCo:一个会自己造技能的科研 agent\n\n论文提出的 DisCo 是 skill-powered 的研究 agent,既造技能也用技能。蒸馏分两种互补形态:task-agnostic 把领域里最常用的仓库整体浓缩成可复用技能;task-oriented 针对具体任务现场生成所需技能。前者跑遍开源生态的产出,就是 AREX-Skill 库:5000+ 个经过验证的可执行技能,来自 1000 个常用 ML 仓库,组织成 20 个领域、178 个能力族。这个库 8 月 3 日首发时只覆盖 170+ 仓库,8 月 27 日刚扩到 1000 仓库与 5000+ 技能;仓库以 Apache 2.0 开源,但每个技能另有独立许可,使用前需单独检查。\n\n## 固定变量后的四项基准\n\n论文的关键实验把 GPT-5.5 底座、研究 harness 和下游执行预算全部固定,只切换 agent 有没有技能。GitHub README 给出的对比表(标注为 Codex 与 Codex + AREX-Skill):\n\n- MLE-bench(75 个 Kaggle 竞赛,任意奖牌率):31.11% → 72.89%,相对提升 134.3%\n- PaperBench(20 篇论文复现):29.45 → 39.59,+34.4%\n- FrontierCS(188 个 Agent Track 任务):70.63 → 77.14,+9.2%\n- PassNet(200 个编译 pass 优化样本,AS 分):1.343 → 1.531,+14.0%\n\n论文对增益的解释:技能提供可复用的流程、检查点和恢复路径,让 agent 少走昂贵的无引导试错,把更多预算花在实验和验证上;且任务越难,优势越明显。\n\n## 技能长什么样\n\n每个技能以 SKILL.md 为核心,配 references\u002F 与 scripts\u002F,写清适用时机、跑什么、怎么验证、失败如何恢复。运行时路由器先把请求收窄到领域\u002F能力族\u002F仓库\u002F工作流,agent 只加载命中的分支——渐进式披露,避免 5000 个技能撑爆上下文。DisCo CLI 以 npm 包发布(v0.2.1,需 Node.js 22.19+),技能可直接导入 Codex、Claude Code、Pi 等编码 agent。\n\n## 我的看法\n\n134.3% 是相对提升,基数是 31.11%,读数字时要冷静——但在 MLE-bench 这个成熟基准上,奖牌率从 31% 跨到 73% 依然是实打实的变化。更值得注意的是路径:不改底座、不改工作流,只往固定配置里补「蒸馏过的操作知识」,知识本身成了新的性能杠杆。在算力焦虑的当下,这是一条便宜得多的路线。技能库的蒸馏流水线(圈定能力、证据落地、构建技能图、验证精修)若能复制到其他领域,GitHub 上二十多年积累的仓库就成了 agent 可开采的「肌肉记忆」矿藏。需要提醒的是,四项对比数字均为团队自报,期待独立复现。论文:https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.02749","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.02749","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"83673225-05b6-4b4e-babd-e59b2f9232f8","en","BAAI's DisCo turns 1,000 GitHub repos into 5,000 verified skills","BAAI's DisCo distills 1,000 ML repos into 5,000+ verified skills; with backbone fixed, MLE-bench medals rise 31.11% to 72.89%.","Autonomous ML research agents are usually judged on two things: the model backbone, and the harness — the surrounding engineering for planning, execution, memory, and verification. A paper posted to arXiv on Sep 2 by a BAAI team, Repo-To-Skill, argues there is a missing third layer: operational knowledge, the know-how that separates knowing a method from making it work. That knowledge is not absent from the field; it sits in repositories and papers, written for human readers and far too large to load mid-task. The paper's answer is to distill it into compact, verified, reusable skills.\n\n## DisCo: an agent that builds and uses its own skills\n\nDisCo, the agent introduced in the paper, both creates skills and uses them during research. Distillation runs in two complementary forms: task-agnostic, condensing the field's widely used repositories into reusable skills, and task-oriented, producing the skills a concrete task calls for. The task-agnostic sweep across the open ecosystem produced the AREX-Skill Library: 5,000+ verified, executable skills distilled from 1,000 widely used ML repositories, organized into 20 research areas and 178 capability families. The library launched on Aug 3 covering 170+ repositories, and was extended on Aug 27 to 1,000 repositories and 5,000+ skills with a rebuilt router. The repository is Apache 2.0 licensed, though every skill carries its own license that must be checked individually.\n\n## The fixed-variable comparison\n\nThe key experiment holds the GPT-5.5 backbone, research harness, and downstream execution budget fixed, and toggles only whether the agent has the distilled skills. The GitHub README table (labeled Codex vs. Codex + AREX-Skill):\n\n- MLE-bench (75 Kaggle competitions, any-medal rate): 31.11% → 72.89%, a relative gain of 134.3%\n- PaperBench (20 paper replications): 29.45 → 39.59, +34.4%\n- FrontierCS (188 Agent Track tasks): 70.63 → 77.14, +9.2%\n- PassNet (200 compiler-pass samples, AS score): 1.343 → 1.531, +14.0%\n\nThe paper's explanation: skills supply reusable procedures, checks, and recovery paths that let the agent avoid expensive unguided trial-and-error and spend more of its budget on experiments and validation, with the advantage strongest on difficult tasks.\n\n## What a skill looks like\n\nEach skill is organized around a SKILL.md with optional references\u002F and scripts\u002F, capturing when a capability applies, what to run, how to validate it, and how to recover when an experiment fails. At runtime a router narrows a request to an area, family, repository, and workflow; the agent then loads only the branch it needs — progressive disclosure, so 5,000 skills do not blow up the context. The DisCo CLI ships as an npm package (v0.2.1, Node.js 22.19+), and skills can be imported into Codex, Claude Code, Pi, and other coding agents.\n\n## My take\n\nThe 134.3% figure is a relative gain from a 31.11% base, so read it calmly — but moving from a 31% to a 73% medal rate on a mature benchmark like MLE-bench is a real shift. The more interesting part is the path: no backbone change, no workflow change, just distilled operating context added under a fixed setup. Knowledge itself becomes a performance lever, which is a cheap route in a compute-anxious era. If the four-stage distillation pipeline — scope capabilities, ground evidence, construct the skill graph, verify and refine — replicates to other domains, two decades of GitHub repositories become mineable muscle memory for agents. One caveat: all four comparison numbers are self-reported by the team, and independent replication is welcome. Paper: https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.02749","baai-disco-repo-to-skill-library","2026-09-03T17:07:35Z","2026-09-03T17:07:48.724734Z","2026-09-03T17:07:48.724748Z",true,"agent",62,{"items":39},[40,45,50,55,60,65],{"id":41,"title":42,"news_slug":43,"published_at":44},"c94766df-827e-4e4e-a006-b6639ec76722","DeepSeek V4-Flash-0731 转正观察:权重不动,后训练把 Agent 分数打到 V4-Pro 之上","deepseek-v4-flash-0731-agent-benchmark-official-aug2026","2026-08-01T02:00:00+00:00",{"id":46,"title":47,"news_slug":48,"published_at":49},"90af7ff5-b985-42d5-97c7-63a9579b7527","VitaBench 2.0：给 LLM Agent 出「长期用户建模」考卷，SOTA 也不及格","vitabench-2-0-long-term-user-modeling","2026-06-25T14:01:00+00:00",{"id":51,"title":52,"news_slug":53,"published_at":54},"ec2c558c-502d-43a5-9494-c766dfd515e9","EurekAgent：把科学发现的瓶颈从「工作流」拽到「环境」，11 美元跑出 26 圆 packing 新 SOTA","eurekagent-environment-engineering-11-usd","2026-06-11T17:56:35+00:00",{"id":56,"title":57,"news_slug":58,"published_at":59},"63c30bcd-3ffc-47c5-bd74-c2a9ed8f7c94","DeepSeek Harness 预览版开源:Agent 被拆成可插拔的插件栈,模型只负责想、Harness 负责做事","deepseek-harness-plugin-stack","2026-09-05T06:00:00+00:00",{"id":61,"title":62,"news_slug":63,"published_at":64},"00346b75-f071-42fd-ae16-db4c5569f01a","EarlyEval 提前叫停注定失败的 Agent:近半 token 省下,分辨率只动一两个点","earlyeval-early-stop-agent-eval","2026-09-03T21:04:52+00:00",{"id":66,"title":67,"news_slug":68,"published_at":69},"14a7f5ab-e270-461c-b862-4bde139e463f","HarnessDev 基准:让 LLM 自建 Agent Harness,代码领域仍输人类工程师","harnessdev-llm-selfbuilt-agent-harness","2026-09-03T19:10:00+00:00"]