{
  "schema_version": 1,
  "report_date": "2026-05-31",
  "title": "AI 日报 2026-05-31",
  "summary": "今天的主线是 AI 从能力发布继续转向可审计落地：OpenAI 把第三方评测报告拆成 claim、harness 和有效性证据，GitHub 则把 Copilot 采用度推进到代码优先、agent 优先和多 agent cohort。开源侧，语音生成、coding agent workflow 和 AI 生成前端代码质检继续占据 GitHub Trending 前排。",
  "hero_highlights": [
    {
      "title": "OpenAI 第三方评测 playbook",
      "url": "https://openai.com/index/trustworthy-third-party-evaluations-foundations/",
      "reason": "OpenAI 要求第三方评测说明 harness、工具、预算和有效性证据，这会直接影响模型横评、供应商准入和安全评审的报告格式。"
    },
    {
      "title": "GitHub Copilot adoption cohorts",
      "url": "https://github.blog/changelog/2026-05-29-copilot-usage-metrics-api-adds-cohorts-for-ai-adoption/",
      "reason": "GitHub 把 Copilot 指标从活跃用户推进到使用成熟度 cohort，帮助企业识别代码补全、agent-first 和 multi-agent 工作流迁移。"
    }
  ],
  "canonical_url": "https://jasonxzwen.github.io/ai-daily-cn/reports/2026/05/2026-05-31.html",
  "html_path": "reports/2026/05/2026-05-31.html",
  "github_trending": [
    {
      "repo": "microsoft/markitdown",
      "name": "microsoft/markitdown",
      "description": "Microsoft 文件转 Markdown 工具，继续占据 Python daily 前排，适合文档摄取、RAG 预处理和评测样本清洗。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/microsoft/markitdown",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "language": "Python",
      "window": "daily",
      "rank": 1,
      "previous_rank": 2,
      "rank_delta": 1,
      "trend": "up",
      "importance": "notable"
    },
    {
      "repo": "harry0703/MoneyPrinterTurbo",
      "name": "harry0703/MoneyPrinterTurbo",
      "description": "大模型短视频生成流水线，覆盖脚本、素材、配音、字幕和成片生产。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/harry0703/MoneyPrinterTurbo",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "language": "Python",
      "window": "daily",
      "rank": 2,
      "previous_rank": 1,
      "rank_delta": -1,
      "trend": "down",
      "importance": "general"
    },
    {
      "repo": "anthropics/claude-code",
      "name": "anthropics/claude-code",
      "description": "Anthropic 官方 Claude Code 仓库，围绕命令行、编辑器和 agent 编程工作流持续升温。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/anthropics/claude-code",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "language": "Python",
      "window": "daily",
      "rank": 3,
      "previous_rank": 5,
      "rank_delta": 2,
      "trend": "up",
      "importance": "notable"
    },
    {
      "repo": "OpenBMB/VoxCPM",
      "name": "OpenBMB/VoxCPM",
      "description": "OpenBMB 的端到端 TTS 模型仓库，关注语音生成和可复现推理链路。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/OpenBMB/VoxCPM",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "language": "Python",
      "window": "daily",
      "rank": 4,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "OpenMOSS/MOSS-TTS",
      "name": "OpenMOSS/MOSS-TTS",
      "description": "OpenMOSS 的语音与声音生成模型族，进入 Python daily 前排。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/OpenMOSS/MOSS-TTS",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "language": "Python",
      "window": "daily",
      "rank": 6,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "anthropics/skills",
      "name": "anthropics/skills",
      "description": "Anthropic Agent Skills 示例与规范仓库，反映 coding agent 生态对可复用能力包的关注。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/anthropics/skills",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "language": "Python",
      "window": "daily",
      "rank": 8,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "NVlabs/Eagle",
      "name": "NVlabs/Eagle",
      "description": "NVIDIA Labs 的视觉语言模型研究仓库，强调数据中心化策略和前沿 VLM。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/NVlabs/Eagle",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "language": "Python",
      "window": "daily",
      "rank": 9,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "EveryInc/compound-engineering-plugin",
      "name": "EveryInc/compound-engineering-plugin",
      "description": "Every 的 Compound Engineering 插件仓库，面向 AI 辅助工程评审和流程约束。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/EveryInc/compound-engineering-plugin",
      "event_date": "2026-05-31",
      "source": "GitHub Trending TypeScript daily",
      "language": "TypeScript",
      "window": "daily",
      "rank": 1,
      "previous_rank": 3,
      "rank_delta": 2,
      "trend": "up",
      "importance": "notable"
    },
    {
      "repo": "iOfficeAI/AionUi",
      "name": "iOfficeAI/AionUi",
      "description": "本地 cowork 应用，面向 Claude Code、Codex、Gemini CLI 等多 agent 开发环境。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/iOfficeAI/AionUi",
      "event_date": "2026-05-31",
      "source": "GitHub Trending TypeScript daily",
      "language": "TypeScript",
      "window": "daily",
      "rank": 9,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    },
    {
      "repo": "millionco/react-doctor",
      "name": "millionco/react-doctor",
      "description": "面向 AI 生成 React 代码的检测工具，目标是捕捉组件和前端代码中的质量问题。 它把相关能力沉淀为代码、示例和集成入口，方便和同类方案做功能与工程成本比较。",
      "url": "https://github.com/millionco/react-doctor",
      "event_date": "2026-05-31",
      "source": "GitHub Trending TypeScript daily",
      "language": "TypeScript",
      "window": "daily",
      "rank": 10,
      "previous_rank": null,
      "rank_delta": null,
      "trend": "new",
      "importance": "general"
    }
  ],
  "main_items": [
    {
      "title": "OpenAI 发布第三方评测 playbook，要求把 harness 与有效性证据讲清楚",
      "event_date": "2026-05-29",
      "url": "https://openai.com/index/trustworthy-third-party-evaluations-foundations/",
      "source": "OpenAI",
      "tier": "T0",
      "entities": [
        "OpenAI",
        "third-party evaluations",
        "Codex",
        "frontier model safeguards"
      ],
      "summary": "OpenAI 发布面向前沿模型第三方评测的实践框架，核心不是给出单一榜单，而是要求评测报告明确 claim、harness、工具、预算、模型设置和有效性证据。文章特别强调 agentic 系统的表现会被调用框架显著影响，因此第三方评测不能只公布结果分数。",
      "bullets": [
        "文章把第三方评测定位为前沿模型安全与能力判断的一部分，尤其适用于 safeguards、capabilities 和 agentic system 的外部检验。",
        "OpenAI 提出评测报告应记录模型、工具、harness、预算和 validity checks，避免把一次运行结果包装成过宽的能力结论。",
        "对工程团队的直接影响是：以后做模型横评或供应商准入时，报告模板需要暴露执行环境和 elicitation 方式，不能只存最终分数。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    },
    {
      "title": "GitHub Copilot usage metrics API 增加 adoption cohorts",
      "event_date": "2026-05-29",
      "url": "https://github.blog/changelog/2026-05-29-copilot-usage-metrics-api-adds-cohorts-for-ai-adoption/",
      "source": "GitHub Changelog",
      "tier": "T0",
      "entities": [
        "GitHub",
        "Copilot",
        "usage metrics API",
        "AI adoption"
      ],
      "summary": "GitHub 在 Copilot usage metrics API 中加入 adoption cohorts，让企业管理员可以按代码优先、agent 优先、多 agent 等使用成熟度观察组织采用进展，而不是只看活跃用户数量。该能力通过 REST API 暴露，并可与 teams filter 组合。",
      "bullets": [
        "这次更新把 Copilot 采用度从 seat 和 active user 视角推进到工作流成熟度视角，更适合内部平台团队跟踪 enablement 结果。",
        "cohorts 可以帮助企业识别哪些团队停留在代码补全，哪些团队已经进入 agent-first 或 multi-agent 使用模式。",
        "对采购和治理团队来说，这类指标能把 AI 工具 ROI 讨论从总使用量转向具体工作流迁移，但仍需要结合代码质量、review 和安全事件数据。"
      ],
      "image_urls": [],
      "importance": "major",
      "editorial_category": "ai_industry"
    }
  ],
  "model_releases": [
    {
      "name": "Claude Opus 4.8 in GitHub Copilot",
      "provider": "Anthropic / GitHub Copilot",
      "availability": "closed_product",
      "release_scope": "gateway_availability",
      "event_date": "2026-05-28",
      "url": "https://github.blog/changelog/2026-05-28-claude-opus-4-8-is-generally-available-for-github-copilot/",
      "source": "GitHub Changelog",
      "summary": "GitHub 宣布 Claude Opus 4.8 在 GitHub Copilot 中 generally available，覆盖 Copilot Pro+、Business 和 Enterprise 等用户，管理员需要在策略中允许该模型。",
      "image_urls": [],
      "importance": "notable"
    }
  ],
  "hot_blogs": [
    {
      "title": "OpenAI’s Frontier Governance Framework",
      "url": "https://openai.com/index/openai-frontier-governance-framework/",
      "publisher": "OpenAI",
      "author": "OpenAI",
      "event_date": "2026-05-28",
      "topic": "前沿模型治理、风险评估与合规映射",
      "summary": "OpenAI 更新 Frontier Governance Framework，把 cyber offense、CBRN、harmful manipulation、loss of control、model reporting、security risk management、incident response 和外部专家输入放进同一治理框架。它还明确把 California Transparency in Frontier AI Act 与 EU AI Act GPAI Code of Practice 作为对齐背景。对企业来说，这类框架的价值在于提供了供应商尽调问题清单：不仅问模型能力，也要问风险分类、事件响应、外部评审和安全管理链路。",
      "content_type": "blog",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "title": "Building self-improving tax agents with Codex",
      "url": "https://openai.com/index/building-self-improving-tax-agents-with-codex/",
      "publisher": "OpenAI",
      "author": "OpenAI / Thrive / Crete",
      "event_date": "2026-05-27",
      "topic": "Codex 驱动的垂直 agent 迭代闭环",
      "summary": "OpenAI 与 Thrive、Crete 合作描述了 Tax AI 的构建方式：把税务专家反馈、生产使用信号和 Codex 驱动的实现循环结合起来，让 agent 可以围绕垂直领域任务持续改进。文章值得看的是流程形态，而不是税务场景本身：专家知识、真实任务、结构化反馈、代码修改和回归验证被放进同一个产品迭代闭环。",
      "content_type": "engineering_note",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "title": "Profiling in PyTorch (Part 1): The Basics of PyTorch Profiler",
      "url": "https://huggingface.co/blog/torch-profiler",
      "publisher": "Hugging Face Blog",
      "author": "Aritra Roy Gosthipaty, Sayak Paul, Sergio Momesso, Remi Louf, Pedro Cuenca",
      "event_date": "2026-05-29",
      "topic": "torch.profiler、LLM 推理与训练性能分析",
      "summary": "Hugging Face 的 PyTorch profiler 教程从矩阵运算、trace 读取、token/s、inference loop 和 training loop 入手，强调先知道什么可以被 profile，再看 trace 中的 CPU/GPU 时间和调度开销。对于部署 LLM 的团队，这比抽象性能建议更直接：它把 profile 对象限定到具体循环和具体指标，便于定位是数据加载、kernel、compile 还是模型调用链路造成瓶颈。",
      "content_type": "engineering_note",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "title": "Dell Enterprise Hub at Dell Tech World 2026",
      "url": "https://huggingface.co/blog/balaatdell/dell-enterprise-hub-at-dell-tech-world-2026",
      "publisher": "Hugging Face Blog",
      "author": "Bala at Dell",
      "event_date": "2026-05-29",
      "topic": "企业本地部署开放模型与 Dell Enterprise Hub",
      "summary": "这篇 Hugging Face 社区文章记录 Dell Enterprise Hub 在 Dell Tech World 2026 的展示，重点是让开放模型在 Dell 基础设施上以企业可部署方式交付，并提到 goodput 场景与 dell-ai SDK。它不是模型能力发布，更像企业 AI 基础设施信号：开放模型正在被包装进更标准化的硬件、SDK 和本地运行路径中。",
      "content_type": "blog",
      "image_urls": [],
      "importance": "notable"
    }
  ],
  "projects": [
    {
      "name": "OpenBMB/VoxCPM",
      "description": "端到端 TTS 模型仓库，在 GitHub Trending Python daily 中位居前排，语音生成方向热度明显。",
      "domains": [
        "text-to-speech",
        "speech generation",
        "open model"
      ],
      "use_case": "用于评估中文或多语音色生成、语音合成 demo 和本地推理链路。",
      "url": "https://github.com/OpenBMB/VoxCPM",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "signal": "trending",
      "importance": "notable"
    },
    {
      "name": "OpenMOSS/MOSS-TTS",
      "description": "OpenMOSS 的语音与声音生成模型族，作为新进入 Python daily 前排的语音项目值得跟踪。",
      "domains": [
        "text-to-speech",
        "voice generation",
        "model family"
      ],
      "use_case": "用于比较开放 TTS 模型族的推理形态、许可边界和中文语音能力。",
      "url": "https://github.com/OpenMOSS/MOSS-TTS",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "signal": "trending",
      "importance": "notable"
    },
    {
      "name": "NVlabs/Eagle",
      "description": "NVIDIA Labs 的视觉语言模型研究仓库，标题强调 data-centric strategies。",
      "domains": [
        "vision-language model",
        "research",
        "data-centric AI"
      ],
      "use_case": "用于跟踪 VLM 研究中数据策略、训练配方和评测复现材料。",
      "url": "https://github.com/NVlabs/Eagle",
      "event_date": "2026-05-31",
      "source": "GitHub Trending Python daily",
      "signal": "trending",
      "importance": "notable"
    },
    {
      "name": "millionco/react-doctor",
      "description": "面向 AI 生成 React 代码的质量检查工具，进入 TypeScript daily 榜单。",
      "domains": [
        "React",
        "code quality",
        "AI-generated code"
      ],
      "use_case": "用于在 AI 前端代码生成后增加静态检查和质量反馈，尤其适合 agent 生成 UI 的验收链路。",
      "url": "https://github.com/millionco/react-doctor",
      "event_date": "2026-05-31",
      "source": "GitHub Trending TypeScript daily",
      "signal": "trending",
      "importance": "notable"
    }
  ],
  "builder_observations": [
    {
      "author": "ClaudeDevs",
      "content": "Claude Code v2.1.154 的开发者更新把 Opus 4.8、dynamic workflows 和安全修复放在同一批变更中。对 coding agent 团队来说，重点是能力发布已经和路径拦截、数据外泄防护、工作流自动切换一起交付，不能只看模型名称。",
      "url": "https://x.com/ClaudeDevs/status/2060044853279617150",
      "role": "Anthropic Claude developer account",
      "event_date": "2026-05-29",
      "source": "ClaudeDevs X / DeveloperIO context",
      "image_urls": [],
      "importance": "notable"
    },
    {
      "author": "OpenAI / Thrive / Crete",
      "content": "Tax AI 案例显示垂直 agent 的构建重心正在转向生产反馈闭环：专家给出判断，真实任务产生信号，Codex 把改进转成代码与验证循环。这个模式比单次 prompt 调优更接近长期产品工程。",
      "url": "https://openai.com/index/building-self-improving-tax-agents-with-codex/",
      "role": "Applied AI product and engineering teams",
      "event_date": "2026-05-27",
      "source": "OpenAI Engineering",
      "image_urls": [],
      "importance": "notable"
    }
  ],
  "evidence_assets": [],
  "generated_at": "2026-05-30T18:48:38.339Z",
  "stories": [
    {
      "story_id": "main-openai-third-party-eval-playbook",
      "title": "OpenAI 发布第三方评测 playbook，要求把 harness 与有效性证据讲清楚",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-29",
      "primary_entity": "OpenAI",
      "event_type": "launch",
      "object": "OpenAI 发布第三方评测 playbook，要求把 harness 与有效性证据讲清楚",
      "what_happened": "OpenAI 发布面向前沿模型第三方评测的实践框架，核心不是给出单一榜单，而是要求评测报告明确 claim、harness、工具、预算、模型设置和有效性证据。文章特别强调 agentic 系统的表现会被调用框架显著影响，因此第三方评测不能只公布结果分数。",
      "why_it_matters": "OpenAI 提出评测报告应记录模型、工具、harness、预算和 validity checks，避免把一次运行结果包装成过宽的能力结论。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "OpenAI",
          "url": "https://openai.com/index/trustworthy-third-party-evaluations-foundations/",
          "type": "source"
        }
      ]
    },
    {
      "story_id": "main-github-copilot-usage-cohorts",
      "title": "GitHub Copilot usage metrics API 增加 adoption cohorts",
      "importance": "major",
      "trend": "AI industry",
      "event_date": "2026-05-29",
      "primary_entity": "GitHub Changelog",
      "event_type": "update",
      "object": "GitHub Copilot usage metrics API 增加 adoption cohorts",
      "what_happened": "GitHub 在 Copilot usage metrics API 中加入 adoption cohorts，让企业管理员可以按代码优先、agent 优先、多 agent 等使用成熟度观察组织采用进展，而不是只看活跃用户数量。该能力通过 REST API 暴露，并可与 teams filter 组合。",
      "why_it_matters": "cohorts 可以帮助企业识别哪些团队停留在代码补全，哪些团队已经进入 agent-first 或 multi-agent 使用模式。",
      "evidence_level": "primary",
      "sources": [
        {
          "label": "GitHub Changelog",
          "url": "https://github.blog/changelog/2026-05-29-copilot-usage-metrics-api-adds-cohorts-for-ai-adoption/",
          "type": "source"
        }
      ]
    }
  ],
  "huggingface_trending": [],
  "chinese_media_dynamics": [],
  "daily_tracking": [],
  "official_org_updates": [],
  "quality_status": {
    "status": "degraded",
    "public_note": "部分采集覆盖降级，公开页仅展示已核验内容。",
    "affected_sections": [
      "source_status"
    ]
  }
}