每日 Harness 开源 · Source
全部刊期 · All issues

每日 Harness

2026-08-18 · Tuesday, August 18, 2026

智能体走向系统化工程

视图 · View

今日重点 · Today's Highlights

Solo - 用自建 ELF 加载器和 glibc ABI 桥，让 musl 静态程序直接调用宿主 GPU 驱动，CI 已覆盖两种架构上的 2,100 多个共享对象。

全文 ↓

论文 · Papers

15 项 · 论文

开源 / 项目 · Projects

15 项 · 开源 / 项目

本期重点Solo16github.com原文 ↗

github.com

Solo 让 musl 静态可执行文件直接加载宿主上依赖 glibc 的 GPU 驱动，无需捆绑第二套 libc、容器或 AppImage。

–

Openleetcode17github.com原文 ↗

github.com

Openleetcode 是 Haskell 编写的本地算法题运行器，将普通解答经 manifest、语言 harness 和可插拔 Piston 后端送入本地 judge。

–

Visimer21github.com原文 ↗

github.com

Visimer 为 Mermaid 增加双向所见即所得编辑，通过 SVG 元素与具体语法树 span 的关联，把画布动作还原为最小文本变更。

–

Augur22github.com原文 ↗

github.com

Augur 用 Go 扫描文本、图片、PDF 与 Office 文件中的隐形字符、水印、元数据和隐写载荷。

–

Zenfmt25insanai.github.io原文 ↗

insanai.github.io

Zenfmt 是 Zig 编写的文档到 Markdown 转换器，以内容签名识别 19 种输入格式，并用 `.zenfmt.json` 清单记录转换。

–

KubeSentry27kubesentry.io原文 ↗

kubesentry.io

KubeSentry 是面向 Kubernetes 的可自托管运行时威胁检测项目，关注集群已经运行后的异常活动，而非只做部署前静态扫描。

–

Winuse29github.com原文 ↗

github.com

Winuse 以 Rust 核心统一 macOS Accessibility 和 Windows UI Automation，并向上提供 Python API 与 JSON CLI。

–

行业动态 · Industry News

12 项 · 行业动态

A Preview of DuckDB v2.031duckdb.org原文 ↗

duckdb.org

DuckDB 官方提前列出 v2.0 的重点能力和兼容性变化，为这一嵌入式分析数据库的主版本迁移设定预期。

–

Incident with GitHub.com32githubstatus.com原文 ↗

githubstatus.com

GitHub 状态页记录了一次同时影响 GitHub.com 网站和 Pull Request 访问的服务事件，并按调查、缓解、恢复更新进度。

–

The Defender’s Window39openai.com原文 ↗

openai.com

OpenAI 主张，AI 降低攻击旧漏洞门槛的同时，也能通过安全代码、持续告警分诊、攻击路径探测和基础配置修复扩大防守方窗口。

–

博客文章 · Blog Posts

13 项 · 博客文章

AI;DR (AI; Didn't Read)44rickmanelius.com原文 ↗

rickmanelius.com

Rick Manelius 讨论一种闭环：作者让 AI 生成文本，读者再让 AI 摘要，双方都绕过了原本用写作和阅读形成判断的过程。

–

Teaching Everyone to Fish for Tokens45interconnects.ai原文 ↗

interconnects.ai

Nathan Lambert 把开放模型生态的稀缺资源扩展到数据、训练 token、算力、评测和持续维护，而不只讨论是否发布权重。

–

How to Ship a Database Every Day48turbopuffer.com原文 ↗

turbopuffer.com

turbopuffer 介绍其数据库控制面如何把配置、编排、升级和回滚从数据路径中分离，以支持每日发布。

–

How I Over-Engineered My Book51ben.balter.com原文 ↗

ben.balter.com

Ben Balter 把写书和发布过程按软件项目组织，用版本控制、自动构建、检查与可重复输出管理正文。

–

引用来源 · References

61 条 · 引用
  1. 1 GPU Offload in Rust: Portable, Safe, and Fast. arXiv:2608.13759https://arxiv.org/abs/2608.13759 ↩ 回到正文 · back to text
  2. 2 Inducing Reward-Free Judging Rubrics that Reduce Over-Crediting in Agent Evaluation. arXiv:2608.13564https://arxiv.org/abs/2608.13564 ↩ 回到正文 · back to text
  3. 3 A Year in LLM Serving: Workload Evolution, Caching and Load-Balancing. arXiv:2608.13573https://arxiv.org/abs/2608.13573 ↩ 回到正文 · back to text
  4. 4 No Universal Signal Predicts Sample-Level LLM Regression under Version Updates. arXiv:2608.13607https://arxiv.org/abs/2608.13607 ↩ 回到正文 · back to text
  5. 5 Second Thought: Reasoning in Parallel as LLM Agents Act and Observe. arXiv:2608.13667https://arxiv.org/abs/2608.13667 ↩ 回到正文 · back to text
  6. 6 When Personal Memory Has No Single Answer: Evaluating LLM Agents under Irreducible Conflict. arXiv:2608.13921https://arxiv.org/abs/2608.13921 ↩ 回到正文 · back to text
  7. 7 Demystifying Agent Skills: Why They Work - Until They Don't. arXiv:2608.14036https://arxiv.org/abs/2608.14036 ↩ 回到正文 · back to text
  8. 8 Scaling Domain Data Repetition in LLM Pretraining. arXiv:2608.14071https://arxiv.org/abs/2608.14071 ↩ 回到正文 · back to text
  9. 9 Mandato: Protocol-Level Enforcement of Digitally Signed Mandates on AI Agent Actions. arXiv:2608.14074https://arxiv.org/abs/2608.14074 ↩ 回到正文 · back to text
  10. 10 AgentRewind: Recoverable Execution for Long-Horizon LLM Agents. arXiv:2608.14380https://arxiv.org/abs/2608.14380 ↩ 回到正文 · back to text
  11. 11 Knowing When to Stop: Bayesian Optimal Stopping for LLM Evaluations. arXiv:2608.14425https://arxiv.org/abs/2608.14425 ↩ 回到正文 · back to text
  12. 12 Does a Language Server Save Tokens for Coding Agents? arXiv:2608.13568https://arxiv.org/abs/2608.13568 ↩ 回到正文 · back to text
  13. 13 Engineering Reliable Coding Agents: Evaluating and Operating the System Around the Model. arXiv:2608.13867https://arxiv.org/abs/2608.13867 ↩ 回到正文 · back to text
  14. 14 Agentic Transaction: Towards ACID-Compliant Agent Systems. arXiv:2608.13900https://arxiv.org/abs/2608.13900 ↩ 回到正文 · back to text
  15. 15 PhoneWorld: Scaling Phone-Use Agent Environments. arXiv:2605.29486https://arxiv.org/abs/2605.29486 ↩ 回到正文 · back to text
  16. 16 Solo. GitHub repositoryhttps://github.com/pg83/solo ↩ 回到正文 · back to text
  17. 17 Openleetcode. GitHub repositoryhttps://github.com/therepanic/openleetcode ↩ 回到正文 · back to text
  18. 18 Suki. GitHub repositoryhttps://github.com/grandimam/suki ↩ 回到正文 · back to text
  19. 19 Engelbart. GitHub repositoryhttps://github.com/divadbaroon/claude-plugins ↩ 回到正文 · back to text
  20. 20 HarnessRouter. GitHub repositoryhttps://github.com/harnessrouter/harnessrouter ↩ 回到正文 · back to text
  21. 21 Visimer. GitHub repositoryhttps://github.com/inkeep/visimer ↩ 回到正文 · back to text
  22. 22 Augur. GitHub repositoryhttps://github.com/dejo1307/augur ↩ 回到正文 · back to text
  23. 23 OpenWebSearchhttps://openwebsearch.ai ↩ 回到正文 · back to text
  24. 24 Agent Mesh. GitHub repositoryhttps://github.com/cbalgeman/agent-mesh ↩ 回到正文 · back to text
  25. 25 Zenfmthttps://insanai.github.io/zenfmt/ ↩ 回到正文 · back to text
  26. 26 Eve Software Factory. GitHub repositoryhttps://github.com/vercel-labs/eve-software-factory-template/tree/main ↩ 回到正文 · back to text
  27. 27 KubeSentryhttps://kubesentry.io/ ↩ 回到正文 · back to text
  28. 28 AgenticSchema. GitHub repositoryhttps://github.com/searchstefano/agenticschema ↩ 回到正文 · back to text
  29. 29 Winuse. GitHub repositoryhttps://github.com/lgxz/winuse ↩ 回到正文 · back to text
  30. 30 Doberman. GitHub repositoryhttps://github.com/fu351/Doberman-Core ↩ 回到正文 · back to text
  31. 31 A Preview of DuckDB v2.0https://duckdb.org/2026/08/17/duckdb-20-highlights ↩ 回到正文 · back to text
  32. 32 Incident with GitHub.comhttps://www.githubstatus.com/incidents/zkxwbgr0cnmx ↩ 回到正文 · back to text
  33. 33 AI-Generated GitHub Copilot “Autofix” Allowed Compromise of Snowflake's Jirahttps://www.wiz.io/blog/red-agent-snowflake-copilot-cicd-bug ↩ 回到正文 · back to text
  34. 34 Apple's App Tracking Transparency Treated Its Own Apps Better Than Rivalshttps://www.bundeskartellamt.de/SharedDocs/Meldung/EN/Pressemitteilungen/2026/08_17_2026_Apple_ATTF.html ↩ 回到正文 · back to text
  35. 35 Stripe Will Reportedly Acquire OpenRouter for $7B+https://techcrunch.com/2026/08/16/stripe-will-reportedly-acquire-ai-gateway-startup-openrouter-for-7b/ ↩ 回到正文 · back to text
  36. 36 Nvidia Scales Back OpenAI Data-Center Guaranteehttps://www.reuters.com/business/nvidia-scales-back-250-billion-openai-data-center-guarantee-wsj-reports-2026-08-14/ ↩ 回到正文 · back to text
  37. 37 Anthropic IPO Valuation Hinges on 2028 Revenue Forecasthttps://www.reuters.com/business/anthropic-ipo-valuation-hinges-190-200-billion-2028-revenue-forecast-sources-say-2026-08-15/ ↩ 回到正文 · back to text
  38. 38 Rare Books Shipment Ended at an Amazon AI Training Facilityhttps://www.404media.co/we-tracked-a-shipment-of-rare-books-it-ended-at-an-amazon-ai-training-facility/ ↩ 回到正文 · back to text
  39. 39 The Defender’s Windowhttps://openai.com/index/the-defenders-window ↩ 回到正文 · back to text
  40. 40 New Policy Ideas for the Intelligence Agehttps://openai.com/index/new-policy-ideas-for-the-intelligence-age ↩ 回到正文 · back to text
  41. 41 Meta Faces ‘Astronomical’ Consequences as Legal Fight Reaches Critical Momenthttps://www.cnbc.com/2026/08/17/meta-attorneys-general-california-federal-trial-astronomical-consequences.html ↩ 回到正文 · back to text
  42. 42 India Paves the Way for Merchant Fees on UPI Transactionshttps://www.bbc.com/news/articles/c8xnwqe00v1o ↩ 回到正文 · back to text
  43. 43 Qwen 3.8 27B Is Excellent, but It Defaults to Overthinking Thingshttps://simonwillison.net/2026/Aug/16/qwen-38-27b/ ↩ 回到正文 · back to text
  44. 44 AI;DR (AI; Didn't Read)https://www.rickmanelius.com/p/aidr-ai-didnt-read ↩ 回到正文 · back to text
  45. 45 Teaching Everyone to Fish for Tokenshttps://www.interconnects.ai/p/teaching-everyone-to-fish-for-tokens ↩ 回到正文 · back to text
  46. 46 Anthropic's “Watermark” Text Adulteration in Claude Is a Perversion of Writinghttps://daringfireball.net/2026/08/anthropics_watermark_text_adulteration_in_claude_is_a_perversion_of_writing ↩ 回到正文 · back to text
  47. 47 GitHub Has Alternatives, but No Replacementhttps://lalitm.com/post/github-alternatives/ ↩ 回到正文 · back to text
  48. 48 How to Ship a Database Every Dayhttps://turbopuffer.com/blog/control-plane ↩ 回到正文 · back to text
  49. 49 Vetted AI Code Is Hard to Justifyhttps://amoffat.github.io/blog/vetting-burnout.html ↩ 回到正文 · back to text
  50. 50 GPT-5.6 Sol Is the Best “Vision” Model OpenAI Ever Releasedhttps://blog.roboflow.com/openai-gpt-5-6/ ↩ 回到正文 · back to text
  51. 51 How I Over-Engineered My Bookhttps://ben.balter.com/2026/08/17/how-i-over-engineered-my-book/ ↩ 回到正文 · back to text
  52. 52 Self-Hosted Email Continues to Steeply Declinehttps://labs.ripe.net/author/artem-berezin/two-providers-a-stubborn-plateau-and-a-very-long-tail-email-in-the-tranco-top-1m/ ↩ 回到正文 · back to text
  53. 53 Beyond WASI: Running Any Rust Application in the Browser with BrowserPod 3.0https://labs.leaningtech.com/blog/browserpod-rust.html ↩ 回到正文 · back to text
  54. 54 Thinking About Tests: Assertions and Matchershttps://zverok.space/blog/2026-08-16-assertions-and-matchers.html ↩ 回到正文 · back to text
  55. 55 Starting a Decompilation Project from Zerohttps://gambiconf.substack.com/p/starting-a-decompilation-project ↩ 回到正文 · back to text
  56. 56 xai-org/grok-1. GitHub repositoryhttps://github.com/xai-org/grok-1 ↩ 回到正文 · back to text
  57. 57 0xSero/ai-data-extraction. GitHub repositoryhttps://github.com/0xSero/ai-data-extraction ↩ 回到正文 · back to text
  58. 58 basecamp/omarchy. GitHub repositoryhttps://github.com/basecamp/omarchy ↩ 回到正文 · back to text
  59. 59 unslothai/unsloth. GitHub repositoryhttps://github.com/unslothai/unsloth ↩ 回到正文 · back to text
  60. 60 Lightricks/LTX-2. GitHub repositoryhttps://github.com/Lightricks/LTX-2 ↩ 回到正文 · back to text
  61. 61 chaitanyagiri/munder-difflin. GitHub repositoryhttps://github.com/chaitanyagiri/munder-difflin ↩ 回到正文 · back to text