{
  "schema_version": "1.1",
  "as_of": "2026-11-06",
  "timezone": "Asia/Shanghai",
  "scope": "2026-focused Data Agent architectures, public implementations, evaluations, papers, talks, and living documentation",
  "resources": [
    {
      "id": "aws-sagemaker-context",
      "date": "2026-01-22",
      "date_basis": "published",
      "title": "Accelerate context-aware data analysis and ML workflows with Amazon SageMaker Data Agent",
      "organization": "AWS",
      "url": "https://aws.amazon.com/blogs/big-data/accelerate-context-aware-data-analysis-and-ml-workflows-with-amazon-sagemaker-data-agent/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "catalog",
        "governance",
        "execution"
      ]
    },
    {
      "id": "openai-inhouse-data-agent",
      "date": "2026-01-29",
      "date_basis": "published",
      "title": "Inside OpenAI’s in-house data agent",
      "organization": "OpenAI",
      "url": "https://openai.com/index/inside-our-in-house-data-agent/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "memory",
        "rag",
        "evaluation"
      ]
    },
    {
      "id": "snowflake-agentic-developer-platform",
      "date": "2026-02-03",
      "date_basis": "published",
      "title": "Take Ideas to Production Faster with an Agentic AI Developer Platform",
      "organization": "Snowflake",
      "url": "https://www.snowflake.com/en/blog/agentic-ai-developer-platform/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "data-engineering",
        "governance",
        "skills"
      ]
    },
    {
      "id": "pinterest-context-intent-embeddings",
      "date": "2026-03-06",
      "date_basis": "published",
      "title": "Unified Context-Intent Embeddings for Scalable Text-to-SQL",
      "organization": "Pinterest Engineering",
      "url": "https://medium.com/pinterest-engineering/unified-context-intent-embeddings-for-scalable-text-to-sql-793635e60aac",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "text-to-sql",
        "retrieval",
        "context",
        "governance"
      ]
    },
    {
      "id": "snowflake-agent-context-layer",
      "date": "2026-03-19",
      "date_basis": "published",
      "title": "The Agent Context Layer for Trustworthy Data Agents",
      "organization": "Snowflake",
      "url": "https://www.snowflake.com/en/blog/agent-context-layer-trustworthy-data-agents/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "semantic-layer",
        "governance"
      ]
    },
    {
      "id": "meta-analytics-agent",
      "date": "2026-03-30",
      "date_basis": "published",
      "title": "Inside Meta’s Home Grown AI Analytics Agent",
      "organization": "Analytics at Meta",
      "url": "https://medium.com/@AnalyticsAtMeta/inside-metas-home-grown-ai-analytics-agent-4ea6779acfb3",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "memory",
        "sql",
        "evidence"
      ]
    },
    {
      "id": "google-event-driven-data-agents",
      "date": "2026-04-11",
      "date_basis": "published",
      "title": "Building Event-Driven Data Agents with BigQuery, Pub/Sub, and ADK",
      "organization": "Google Cloud",
      "url": "https://cloud.google.com/blog/topics/developers-practitioners/building-event-driven-data-agents-with-bigquery-pubsub-and-adk",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "event-driven",
        "bigquery",
        "observability",
        "human-in-the-loop"
      ]
    },
    {
      "id": "databricks-genie-agent-mode",
      "date": "2026-04-17",
      "date_basis": "published",
      "title": "Introducing Genie Agent Mode",
      "organization": "Databricks",
      "url": "https://www.databricks.com/blog/introducing-genie-agent-mode",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "planning",
        "analytics",
        "agent-loop"
      ]
    },
    {
      "id": "snowflake-cortex-code-expands",
      "date": "2026-04-21",
      "date_basis": "published",
      "title": "Cortex Code Expands: One Governed Agent for Your Entire Data Stack",
      "organization": "Snowflake",
      "url": "https://www.snowflake.com/en/blog/cortex-code-governed-agent-data-stack/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "data-engineering",
        "mcp",
        "sandbox",
        "governance"
      ]
    },
    {
      "id": "aws-athena-quick-agentic-analytics",
      "date": "2026-04-30",
      "date_basis": "published",
      "title": "Unleashing agentic AI analytics on Amazon SageMaker with Amazon Athena and Amazon Quick",
      "organization": "AWS",
      "url": "https://aws.amazon.com/blogs/machine-learning/unleashing-agentic-ai-analytics-on-amazon-sagemaker-with-amazon-athena-and-amazon-quick/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "structured-data",
        "unstructured-data",
        "governance"
      ]
    },
    {
      "id": "datahub-analytics-agent-post",
      "date": "2026-04-30",
      "date_basis": "published",
      "title": "Introducing the DataHub Analytics Agent",
      "organization": "DataHub",
      "url": "https://datahub.com/blog/datahub-analytics-agent/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context-graph",
        "mcp",
        "lineage",
        "semantic-layer"
      ]
    },
    {
      "id": "databricks-genie-frontier",
      "date": "2026-05-08",
      "date_basis": "published",
      "title": "Pushing the Frontier for Data Agents with Genie",
      "organization": "Databricks",
      "url": "https://www.databricks.com/blog/pushing-frontier-data-agents-genie",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "multi-agent",
        "multi-model",
        "verification",
        "retrieval"
      ]
    },
    {
      "id": "google-data-agent-kit-ide",
      "date": "2026-05-19",
      "date_basis": "published",
      "title": "Data Agent Kit brings data skills and tools to your IDE or CLI",
      "organization": "Google Cloud",
      "url": "https://cloud.google.com/blog/products/data-analytics/data-agent-kit-brings-data-skills-and-tools-to-your-ide-or-cli/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "skills",
        "plugin",
        "data-engineering"
      ]
    },
    {
      "id": "anthropic-self-service-analytics",
      "date": "2026-06-03",
      "date_basis": "published",
      "title": "How Anthropic enables self-service data analytics with Claude",
      "organization": "Anthropic",
      "url": "https://claude.com/resources/articles/how-anthropic-enables-self-service-data-analytics-with-claude",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "skills",
        "context",
        "validation",
        "analytics"
      ]
    },
    {
      "id": "google-new-data-agents",
      "date": "2026-06-15",
      "date_basis": "published",
      "title": "What’s new in data agents: Supercharging your AI workflows",
      "organization": "Google Cloud",
      "url": "https://cloud.google.com/blog/products/data-analytics/new-data-agents-across-the-agentic-data-cloud/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "data-science",
        "data-engineering",
        "observability",
        "mcp"
      ]
    },
    {
      "id": "snowflake-cortex-sense",
      "date": "2026-06-30",
      "date_basis": "published",
      "title": "Introducing Cortex Sense: Grounded Context for the Data You Never Modeled",
      "organization": "Snowflake",
      "url": "https://www.snowflake.com/en/blog/enterprise-ai-agents-grounded-context/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "semantic-layer",
        "evaluation"
      ]
    },
    {
      "id": "databricks-specialized-vs-general",
      "date": "2026-07-23",
      "date_basis": "published",
      "title": "Why a Frontier Data Agent Outperforms General Coding Agents in Quality and Cost",
      "organization": "Databricks",
      "url": "https://www.databricks.com/blog/why-frontier-data-agent-outperforms-general-coding-agents-quality-and-cost",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "memory",
        "enterprise-context"
      ]
    },
    {
      "id": "langchain-agent-first-data-stack",
      "date": "2026-07-27",
      "date_basis": "published",
      "title": "How we built LangChain’s agent-first data stack",
      "organization": "LangChain",
      "url": "https://www.langchain.com/blog/agent-data-stack",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context-as-code",
        "dbt",
        "semantic-layer",
        "adoption"
      ]
    },
    {
      "id": "snowflake-internal-context-layer",
      "date": "2026-08-17",
      "date_basis": "published",
      "title": "Building an Internal Context Layer for AI Agents",
      "organization": "Snowflake",
      "url": "https://www.snowflake.com/en/blog/snowflake-internal-context-layer-for-ai-agents/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "semantic-layer",
        "internal-tooling"
      ]
    },
    {
      "id": "google-agentic-analytics-tutorial",
      "date": "2026-09-08",
      "date_basis": "published",
      "title": "Agentic analytics with the Data Agent Kit",
      "organization": "Google Cloud",
      "url": "https://cloud.google.com/blog/products/data-analytics/agentic-analytics-with-the-data-agent-kit",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "multi-source",
        "tutorial",
        "root-cause-analysis"
      ]
    },
    {
      "id": "openai-data-agent-product",
      "date": "2026-09-10",
      "date_basis": "published",
      "title": "Now everyone can put data to work",
      "organization": "OpenAI",
      "url": "https://openai.com/index/put-data-to-work/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "analytics",
        "dashboard",
        "actions",
        "governance"
      ]
    },
    {
      "id": "devconf-rhoai-data-agents",
      "date": "2026-09-24",
      "date_basis": "event",
      "title": "Building In-House Data Agents on RHOAI",
      "organization": "DevConf.US",
      "url": "https://pretalx.devconf.info/devconf-us-2026/talk/P9W73V/",
      "section": "talk",
      "language": "en",
      "status": "published",
      "tags": [
        "spark",
        "openshift",
        "mcp",
        "human-in-the-loop"
      ]
    },
    {
      "id": "datus-semantic-layer-loop",
      "date": "2026-09-29",
      "date_basis": "published",
      "title": "The Semantic Layer in the Agent Loop",
      "organization": "Datus",
      "url": "https://datus.ai/blog/semantic-layer-in-the-agent-loop/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "semantic-layer",
        "ontology",
        "validation",
        "sql"
      ]
    },
    {
      "id": "google-dak-ga",
      "date": "2026-09-30",
      "date_basis": "released",
      "title": "Data Agent Kit is now GA",
      "organization": "Google Cloud",
      "url": "https://cloud.google.com/blog/topics/developers-practitioners/data-agent-kit-is-now-ga-bring-google-data-cloud-to-any-coding-agent",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "skills",
        "plugin",
        "iam"
      ]
    },
    {
      "id": "google-cross-cloud-analytics",
      "date": "2026-10-16",
      "date_basis": "updated",
      "title": "Implement agentic analytics workflows for distributed data",
      "organization": "Google Cloud",
      "url": "https://docs.cloud.google.com/architecture/agentic-ai-cross-cloud-analytics",
      "section": "architecture",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "reference-architecture",
        "multi-cloud",
        "multi-source"
      ]
    },
    {
      "id": "amplitude-task-evaluation",
      "date": "2026-02-17",
      "date_basis": "published",
      "title": "Can agents take on enterprise analytics? A task-based evaluation",
      "organization": "Amplitude",
      "url": "https://amplitude.com/blog/ai-analytics-agents-task-based-evaluation",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "rubric",
        "analytics",
        "benchmark"
      ]
    },
    {
      "id": "hex-eval-lab",
      "date": "2026-05-22",
      "date_basis": "published",
      "title": "How we built a lab to evaluate data agents",
      "organization": "Hex",
      "url": "https://hex.tech/blog/evaluate-data-agents/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "evaluation",
        "llm-as-judge",
        "synthetic-data"
      ]
    },
    {
      "id": "hex-evals",
      "date": "2026-08-04",
      "date_basis": "published",
      "title": "Introducing Evals",
      "organization": "Hex",
      "url": "https://hex.tech/blog/evals/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "evaluation",
        "ci-cd",
        "drift",
        "cost"
      ]
    },
    {
      "id": "clickhouse-agentic-analytics-benchmark",
      "date": "2026-09-01",
      "date_basis": "published",
      "title": "The Agentic Analytics Benchmark",
      "organization": "ClickHouse",
      "url": "https://clickhouse.com/blog/agentic-analytics-benchmark-data-agent-mnist",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "cost",
        "latency",
        "open-harness"
      ]
    },
    {
      "id": "preset-internal-data-agent",
      "date": "2026-05-21",
      "date_basis": "event",
      "title": "Anatomy of our Internal Data Agent",
      "organization": "Preset",
      "url": "https://preset.io/events/how-we-built-our-agents/",
      "section": "talk",
      "language": "en",
      "status": "published",
      "tags": [
        "video",
        "semantic-layer",
        "internal-tooling"
      ]
    },
    {
      "id": "open-data-circle-agent-data-plane",
      "date": "2026-05-26",
      "date_basis": "published",
      "title": "Data Agents Need an Agent-facing Data Plane",
      "organization": "Open Data Circle",
      "url": "https://speakerdeck.com/opendatacircle/data-agents-need-an-agent-facing-data-plane",
      "section": "talk",
      "language": "en",
      "status": "published",
      "tags": [
        "slides",
        "data-plane",
        "governance",
        "observability"
      ]
    },
    {
      "id": "shopee-qcon-data-agent",
      "date": "2026-04-16/2026-04-18",
      "date_basis": "event",
      "title": "从 AI 取数到智能分析：企业级数据 Agent 的多阶段演进与工程化落地",
      "organization": "Shopee / QCon 北京",
      "url": "https://qcon.infoq.cn/2026/beijing/presentation/7006",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "multi-agent",
        "file-as-context",
        "sandbox",
        "evaluation"
      ]
    },
    {
      "id": "netease-qcon-dataagent",
      "date": "2026-04-16/2026-04-18",
      "date_basis": "event",
      "title": "从 Copilot 到 DataAgent：企业级智能数据开发治理平台的技术演进和实践",
      "organization": "网易 / QCon 北京",
      "url": "https://qcon.infoq.cn/2026/beijing/presentation/7030",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "data-engineering",
        "governance",
        "copilot"
      ]
    },
    {
      "id": "meituan-qcon-data-agent",
      "date": "2026-04-16/2026-04-18",
      "date_basis": "event",
      "title": "不止 ChatBI：懂归因、能洞察的美团 Data Agent 落地实践",
      "organization": "美团 / QCon 北京",
      "url": "https://qcon.infoq.cn/2026/beijing/presentation/6946",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "knowledge-graph",
        "memory",
        "semantic-layer",
        "evaluation"
      ]
    },
    {
      "id": "tongcheng-dacon-dataagent",
      "date": "2026-04-25",
      "date_basis": "event",
      "title": "DataAgent 在分析和营销场景的智能化建设探索",
      "organization": "同程旅行 / DACon 上海",
      "url": "https://www.bagevent.com/event/9108668/p/569570",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "analytics",
        "marketing",
        "multi-agent",
        "actions"
      ]
    },
    {
      "id": "tencent-cloud-trustworthy-data-agent",
      "date": "2026-07-03",
      "date_basis": "published",
      "title": "构建可信、可控、可演进的 Data Agent",
      "organization": "腾讯云大数据",
      "url": "https://developer.cloud.tencent.cn/article/2703714",
      "section": "architecture",
      "language": "zh",
      "status": "published",
      "tags": [
        "semantic-layer",
        "multi-agent",
        "skills",
        "audit"
      ]
    },
    {
      "id": "datacyber-aicon",
      "date": "2026-08-21/2026-08-22",
      "date_basis": "event",
      "title": "基于 DataCyber 的 Data Agent：构建企业数据智能中枢的工程范式",
      "organization": "数新智能 / AICon 深圳",
      "url": "https://aicon.infoq.cn/2026/shenzhen/presentation/7205",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "skills",
        "workflow",
        "policy",
        "evaluation"
      ]
    },
    {
      "id": "kuaishou-data-agent-salon",
      "date": "2026-09-19",
      "date_basis": "event",
      "title": "Data Agent：数据生产与智能决策新范式",
      "organization": "快手技术沙龙",
      "url": "https://www.ithome.com/1/005/819.htm",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "multi-agent",
        "skills",
        "trace",
        "evaluation"
      ]
    },
    {
      "id": "qcon-shanghai-track",
      "date": "2026-10-22/2026-10-24",
      "date_basis": "event",
      "title": "Agentic 时代的数据系统重构",
      "organization": "QCon 上海",
      "url": "https://qcon.infoq.cn/2026/shanghai/track/1976",
      "section": "talk",
      "language": "zh",
      "status": "agenda",
      "tags": [
        "agenda",
        "data-systems"
      ],
      "last_verified": "2026-11-02",
      "material_review": {
        "checked_on": "2026-11-02",
        "status": "event-held-material-unverified",
        "evidence_url": "https://qcon.infoq.cn/2026/shanghai/schedule",
        "note": "官网日程将该专题列于2026-10-23，现有议题页仍主要提供讲师和提纲；日程增加“点击查看全部PPT”入口，但未核验具体 Data Agent deck 文件。",
        "materials_portal_url": "https://ppt.infoq.cn/list/173"
      }
    },
    {
      "id": "qcon-shanghai-netease-das",
      "date": "2026-10-23",
      "date_basis": "event",
      "title": "DAS ：当 Data 遇见 Agent——网易云音乐大数据 AI Infra 的共生之路",
      "organization": "网易云音乐 / QCon 上海",
      "url": "https://qcon.infoq.cn/2026/shanghai/presentation/7257",
      "section": "talk",
      "language": "zh",
      "status": "agenda",
      "tags": [
        "agenda",
        "data-engineering",
        "security"
      ],
      "last_verified": "2026-11-02",
      "material_review": {
        "checked_on": "2026-11-02",
        "status": "event-held-material-unverified",
        "evidence_url": "https://qcon.infoq.cn/2026/shanghai/schedule",
        "note": "以官方日程核对正式标题与具体日期；未核验到可打开的具体PPT或录像，保留agenda。"
      }
    },
    {
      "id": "qcon-shanghai-kuaishou",
      "date": "2026-10-23",
      "date_basis": "event",
      "title": "Data Agent - 快手大数据智能化探索之路",
      "organization": "快手 / QCon 上海",
      "url": "https://qcon.infoq.cn/2026/shanghai/presentation/7230",
      "section": "talk",
      "language": "zh",
      "status": "agenda",
      "tags": [
        "agenda",
        "multi-agent",
        "context",
        "evaluation"
      ],
      "last_verified": "2026-11-02",
      "material_review": {
        "checked_on": "2026-11-02",
        "status": "event-held-material-unverified",
        "evidence_url": "https://qcon.infoq.cn/2026/shanghai/presentation/7230",
        "note": "原始议题页提供父子Agent+Skill、知识飞轮及Trace评测提纲；未核验具体公开deck或录像，保留agenda。"
      }
    },
    {
      "id": "qcon-shanghai-datus",
      "date": "2026-10-23",
      "date_basis": "event",
      "title": "如何构建 AI Native 的数据语义层：从 Data Agent 实践到可执行的 Semantic Runtime",
      "organization": "Datus / QCon 上海",
      "url": "https://qcon.infoq.cn/2026/shanghai/presentation/7342",
      "section": "talk",
      "language": "zh",
      "status": "agenda",
      "tags": [
        "agenda",
        "semantic-layer",
        "sql"
      ],
      "last_verified": "2026-11-02",
      "material_review": {
        "checked_on": "2026-11-02",
        "status": "event-held-material-unverified",
        "evidence_url": "https://qcon.infoq.cn/2026/shanghai/schedule",
        "note": "官方日程与同专题其他议题页交叉核对完整标题；具体议题页本次工具不可达，未核验deck或录像。"
      }
    },
    {
      "id": "qcon-shanghai-rednote",
      "date": "2026-10-23",
      "date_basis": "event",
      "title": "DIBP-小红书数据 Agent 的语义层建设实践",
      "organization": "小红书 / QCon 上海",
      "url": "https://qcon.infoq.cn/2026/shanghai/presentation/7343",
      "section": "talk",
      "language": "zh",
      "status": "agenda",
      "tags": [
        "agenda",
        "harness",
        "multi-agent",
        "context"
      ],
      "last_verified": "2026-11-02",
      "material_review": {
        "checked_on": "2026-11-02",
        "status": "event-held-material-unverified",
        "evidence_url": "https://qcon.infoq.cn/2026/shanghai/presentation/7343",
        "note": "原始议题页公开Context Layer生产、治理与消费等提纲；未核验deck或录像，保留agenda。"
      }
    },
    {
      "id": "qcon-shanghai-meituan-aide",
      "date": "2026-10-23",
      "date_basis": "event",
      "title": "从代码生成到可信交付：美团 AIDE 数据研发 Agent 实践",
      "organization": "美团 / QCon 上海",
      "url": "https://qcon.infoq.cn/2026/shanghai/presentation/7356",
      "section": "talk",
      "language": "zh",
      "status": "agenda",
      "tags": [
        "agenda",
        "data-engineering",
        "guardrails",
        "evaluation"
      ],
      "last_verified": "2026-11-02",
      "material_review": {
        "checked_on": "2026-11-02",
        "status": "event-held-material-unverified",
        "evidence_url": "https://qcon.infoq.cn/2026/shanghai/presentation/7356",
        "note": "原始议题页公开交付验证、专家委派、知识审核等提纲；未核验deck或录像，保留agenda。"
      }
    },
    {
      "id": "google-data-agent-kit-plugin",
      "date": "2026-10-30",
      "date_basis": "released",
      "title": "data-agent-kit-plugin",
      "organization": "GoogleCloudPlatform",
      "url": "https://github.com/GoogleCloudPlatform/data-agent-kit-plugin",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "skills",
        "plugin",
        "apache-2.0"
      ],
      "summary": "给现有Coding Agent提供Google Data Cloud的Skills与MCP工具；1.2.0补充数据库接入、BigQuery/Spark运维技能及评测套件。本轮连接安装、云端授权与完整动手案例。",
      "date_evidence": "原索引使用2026-04-02T14:10:44Z仓库创建时间；本轮改用GitHub 1.2.0 published_at 2026-10-30T02:52:43Z，不将本次发现日期当作发布时间。",
      "evidence_url": "https://github.com/GoogleCloudPlatform/data-agent-kit-plugin/releases/tag/1.2.0",
      "license": "Apache-2.0",
      "license_url": "https://github.com/GoogleCloudPlatform/data-agent-kit-plugin/blob/main/LICENSE",
      "component_type": "skills-and-tool-plugin",
      "technical_material_url": "https://docs.cloud.google.com/data-agent-kit",
      "mechanism": "插件把数据工作流程Skills与MCP工具交给宿主Agent；实际云资源访问由调用身份、IAM及具体服务权限约束。",
      "use_case": "沿用Coding Agent开展多源分析、数据工程与Notebook任务，或比较流程指引与强制授权的作用。",
      "caveat": "插件与IDE扩展是不同安装入口；免费插件不代表模型、计算和存储免费。Skills中的确认指令不等于云端权限执行；示例资源与生产配置应分别评审。本轮未调用模型或创建云资源。",
      "reading_path": [
        {
          "title": "插件安装与配置",
          "url": "https://github.com/GoogleCloudPlatform/data-agent-kit-plugin/blob/main/README.md",
          "focus": "先核对宿主客户端、ADC和插件配置。"
        },
        {
          "title": "远程MCP与授权",
          "url": "https://docs.cloud.google.com/data-agent-kit/use-mcp-servers",
          "focus": "检查工具调用角色、产品权限和服务账号模拟。"
        },
        {
          "title": "原始数据到预测Codelab",
          "url": "https://codelabs.developers.google.com/next26/gen-keynote/raw-data-forecasting",
          "focus": "跟踪文档抽取、Iceberg/BigQuery和Spark分析链路；示例授权不照搬生产。"
        },
        {
          "title": "Data Apps技术实现指南",
          "url": "https://googlecloudplatform.github.io/cloud-solutions/agentic-coding-with-data-agent-kit-demo/",
          "focus": "从项目环境、数据库能力到应用交付组织复现。"
        }
      ],
      "last_verified": "2026-11-05",
      "release_updates": [
        {
          "version": "1.2.0",
          "published_at": "2026-10-30T02:52:43Z",
          "url": "https://github.com/GoogleCloudPlatform/data-agent-kit-plugin/releases/tag/1.2.0"
        }
      ]
    },
    {
      "id": "datahub-analytics-agent",
      "date": "2026-07-29",
      "date_basis": "released",
      "title": "analytics-agent",
      "organization": "datahub-project",
      "url": "https://github.com/datahub-project/analytics-agent",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "context-graph",
        "langgraph",
        "vega-lite",
        "apache-2.0"
      ],
      "summary": "LangGraph ReAct应用连接目录上下文与查询引擎并生成Vega-Lite图表；v0.4.0增加DuckDB引擎、连接池配置和合成零售样例。本轮补齐从本地演示到目录反馈的路线。",
      "date_evidence": "原索引以2026-04-22T20:43:55Z仓库创建时间为证据；本轮改用更明确的v0.4.0 published_at 2026-07-29T19:44:12Z，并与main README分开核验。",
      "evidence_url": "https://github.com/datahub-project/analytics-agent/releases/tag/v0.4.0",
      "license": "Apache-2.0",
      "license_url": "https://github.com/datahub-project/analytics-agent/blob/main/LICENSE",
      "component_type": "agent-application",
      "technical_material_url": "https://github.com/datahub-project/analytics-agent/blob/main/README.md",
      "mechanism": "分析Agent调用目录/引擎工具生成并执行SQL，通过上下文改进建议由用户确认后回写目录；DataHub连接可选。",
      "use_case": "研究可启动的分析工作区、DuckDB本地演示和人工确认的目录反馈。",
      "caveat": "DataHub上下文质量分数不是答案正确率；元数据工具开关不替代数仓授权。release包含DDL/DML提交行为，不能把整个应用笼统描述成强制只读，需限制具体引擎身份与工具。",
      "reading_path": [
        {
          "title": "启动与开发说明",
          "url": "https://github.com/datahub-project/analytics-agent/blob/main/README.md",
          "focus": "选择CLI快速启动或合成Fiction Retail演示。"
        },
        {
          "title": "v0.4.0能力与变更",
          "url": "https://github.com/datahub-project/analytics-agent/releases/tag/v0.4.0",
          "focus": "核对DuckDB、连接池和错误处理能力。"
        },
        {
          "title": "Agent Context Kit教程",
          "url": "https://datahub.com/blog/building-autonomous-data-agents/",
          "focus": "比较嵌入SDK与远程MCP；历史代码按当前SDK修正。"
        }
      ],
      "last_verified": "2026-11-05",
      "release_updates": [
        {
          "version": "v0.4.0",
          "published_at": "2026-07-29T19:44:12Z",
          "url": "https://github.com/datahub-project/analytics-agent/releases/tag/v0.4.0"
        }
      ]
    },
    {
      "id": "altimate-code",
      "date": "2026-11-04",
      "date_basis": "released",
      "title": "altimate-code",
      "organization": "AltimateAI",
      "url": "https://github.com/AltimateAI/altimate-code",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "dbt",
        "deterministic-tools",
        "data-engineering",
        "mit"
      ],
      "summary": "v0.12.9 提供不依赖模型的 tool CLI 与 MCP server，让外部 Coding Agent 复用数据工具；dbt_lineage 返回结构化模型血缘，工作区规则经提案与 Owner 审查后进入会话，并修复配置授权下分析角色仍须禁止写工具的边界。",
      "date_evidence": "GitHub Release v0.12.9 published_at 为 2026-11-04T23:16:13Z（Asia/Shanghai 为11月5日）；date 沿用来源 UTC 日期，完整时间保留在 release_updates。",
      "evidence_url": "https://github.com/AltimateAI/altimate-code/releases/tag/v0.12.9",
      "license": "MIT",
      "license_url": "https://github.com/AltimateAI/altimate-code/blob/main/LICENSE",
      "notes": "v0.12.8 曾先加入后移除 text_label，不能据旧变更声称该工具可用；v0.12.9 的 analyst 修复是在配置授权下仍拒绝写工具。2026-11-05 核对 MIT LICENSE 与 archived=false。",
      "last_verified": "2026-11-05",
      "release_updates": [
        {
          "version": "v0.12.8",
          "published_at": "2026-10-23T19:37:02Z",
          "url": "https://github.com/AltimateAI/altimate-code/releases/tag/v0.12.8"
        },
        {
          "version": "v0.12.9",
          "published_at": "2026-11-04T23:16:13Z",
          "url": "https://github.com/AltimateAI/altimate-code/releases/tag/v0.12.9"
        }
      ]
    },
    {
      "id": "datafoundry",
      "date": "2026-06-18",
      "date_basis": "repo_created",
      "title": "DataFoundry (repository: dataagent)",
      "organization": "datagallery-ai",
      "url": "https://github.com/datagallery-ai/dataagent",
      "section": "open-source",
      "language": "en",
      "status": "early",
      "tags": [
        "semantic-layer",
        "governance",
        "self-hosted",
        "apache-2.0"
      ]
    },
    {
      "id": "aliyun-data-agent-skill",
      "date": "2026-03-11",
      "date_basis": "repo_created",
      "title": "data-agent-skill",
      "organization": "aliyun",
      "url": "https://github.com/aliyun/data-agent-skill",
      "section": "open-source",
      "language": "zh",
      "status": "published",
      "tags": [
        "mcp",
        "skills",
        "analytics",
        "no-license"
      ]
    },
    {
      "id": "aws-duckdb-data-agent",
      "date": "2026-09-02",
      "date_basis": "repo_created",
      "title": "sample-data-agent-on-duckdb",
      "organization": "aws-samples",
      "url": "https://github.com/aws-samples/sample-data-agent-on-duckdb",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "duckdb",
        "s3",
        "agentcore",
        "mit-0"
      ]
    },
    {
      "id": "microsoft-fabric-workshop",
      "date": "2026-08-28",
      "date_basis": "repo_created",
      "title": "fabric-data-agent-workshop",
      "organization": "microsoft",
      "url": "https://github.com/microsoft/fabric-data-agent-workshop",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "fabric",
        "semantic-model",
        "mlflow",
        "mit"
      ]
    },
    {
      "id": "evoontology",
      "date": "2026-09-15",
      "date_basis": "repo_created",
      "title": "EvoOntology",
      "organization": "ruc-datalab",
      "url": "https://github.com/ruc-datalab/EvoOntology",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "ontology",
        "mcp",
        "self-evolving",
        "mit"
      ]
    },
    {
      "id": "dataspace-repo",
      "date": "2026-07-21",
      "date_basis": "repo_created",
      "title": "DataSpace",
      "organization": "HKUSTDial",
      "url": "https://github.com/HKUSTDial/DataSpace",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "multimodal",
        "evaluator",
        "mit"
      ]
    },
    {
      "id": "agenticdatabench-repo",
      "date": "2026-04-09",
      "date_basis": "repo_created",
      "title": "AgenticDataBench",
      "organization": "AgenticDataBench",
      "url": "https://github.com/AgenticDataBench/AgenticDataBench",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "data-science",
        "skills",
        "apache-2.0"
      ]
    },
    {
      "id": "kddcup-data-agent-starter",
      "date": "2026-03-24",
      "date_basis": "repo_created",
      "title": "KDD Cup 2026 DataAgent-Bench Starter Kit",
      "organization": "HKUSTDial",
      "url": "https://github.com/HKUSTDial/kddcup2026-data-agents-starter-kit",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "react",
        "benchmark",
        "starter-kit",
        "no-license"
      ]
    },
    {
      "id": "automattic-agentic-analytics",
      "date": "2026-05-22",
      "date_basis": "repo_created",
      "title": "agentic-analytics",
      "organization": "Automattic",
      "url": "https://github.com/Automattic/agentic-analytics",
      "section": "open-source",
      "language": "en",
      "status": "early",
      "tags": [
        "plugin",
        "duckdb",
        "reporting",
        "no-license"
      ]
    },
    {
      "id": "spring-ai-alibaba-dataagent",
      "date": "2026-07-29",
      "date_basis": "released",
      "title": "DataAgent",
      "organization": "spring-ai-alibaba",
      "url": "https://github.com/spring-ai-alibaba/DataAgent",
      "section": "open-source",
      "language": "zh",
      "status": "active-project",
      "tags": [
        "text-to-sql",
        "python",
        "mcp",
        "apache-2.0"
      ],
      "summary": "Spring AI Alibaba Graph驱动的SQL/Python/报告应用；本轮补齐架构、三类知识配置与任务级Python沙箱阅读路线，明确人工计划反馈、执行重试与API鉴权的实现边界。",
      "date_evidence": "将原索引未附证据的2026-09-20更新日期改为可核验的GitHub 1.0.0-rc7 published_at 2026-07-29T15:44:52Z；文档按2026-11-05 main快照另行核验，不宣称本日新release。",
      "evidence_url": "https://github.com/spring-ai-alibaba/DataAgent/releases/tag/1.0.0-rc7",
      "license": "Apache-2.0",
      "license_url": "https://github.com/spring-ai-alibaba/DataAgent/blob/main/LICENSE",
      "component_type": "agent-application",
      "technical_material_url": "https://github.com/spring-ai-alibaba/DataAgent/blob/main/docs/ARCHITECTURE.md",
      "mechanism": "证据/Schema召回、表关系推理、可行性评估和规划，经StateGraph调度SQL、Python及报告节点；可在计划阶段加入人工反馈。",
      "use_case": "Java团队研究可扩展数据分析应用、知识配置与容器执行的端到端实现。",
      "caveat": "当前main高级文档明确X-API-Key尚未在后端执行权限校验；管理入口存在不等于认证已生效。1.0.0-rc7名称为RC，不能因GitHub prerelease=false推定为稳定GA；main文档和所固定tag需分别核对。",
      "reading_path": [
        {
          "title": "快速开始",
          "url": "https://github.com/spring-ai-alibaba/DataAgent/blob/main/docs/QUICK_START.md",
          "focus": "准备示例数据库、模型和数据源，再提交第一个任务。"
        },
        {
          "title": "架构设计",
          "url": "https://github.com/spring-ai-alibaba/DataAgent/blob/main/docs/ARCHITECTURE.md",
          "focus": "对照运行时节点、HITL和SQL/Python结果回传。"
        },
        {
          "title": "知识配置",
          "url": "https://github.com/spring-ai-alibaba/DataAgent/blob/main/docs/KNOWLEDGE_USAGE.md",
          "focus": "区分字段映射、业务定义和文档/问答知识。"
        },
        {
          "title": "高级功能",
          "url": "https://github.com/spring-ai-alibaba/DataAgent/blob/main/docs/ADVANCED_FEATURES.md",
          "focus": "查看API鉴权缺口、MCP与任务容器的依赖/资源边界。"
        }
      ],
      "last_verified": "2026-11-05",
      "release_updates": [
        {
          "version": "1.0.0-rc7",
          "published_at": "2026-07-29T15:44:52Z",
          "url": "https://github.com/spring-ai-alibaba/DataAgent/releases/tag/1.0.0-rc7"
        }
      ]
    },
    {
      "id": "nao",
      "date": "2026-11-03",
      "date_basis": "released",
      "title": "nao",
      "organization": "getnao",
      "url": "https://github.com/getnao/nao",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "context-as-code",
        "evaluation",
        "self-hosted",
        "mixed-license"
      ],
      "summary": "v0.3.32–33 引入实验性 codemode 并移除旧 Python interpreter；语义查询在刷新时重编译并可在图表编辑器修改，补充指标元数据同步、画像分区过滤与 GitHub Enterprise Server 接入，适合跟踪上下文到分析工作区的执行链。",
      "date_evidence": "GitHub v0.3.32 published_at 为 2026-11-03T08:54:14Z；v0.3.33 为 2026-11-03T11:55:58Z。前者含实验 codemode、移除 Python interpreter 与语义查询刷新，后者含 GitHub Enterprise Server 支持。",
      "evidence_url": "https://github.com/getnao/nao/releases/tag/v0.3.33",
      "license": "Apache-2.0 + commercial (Enterprise-marked files)",
      "license_url": "https://github.com/getnao/nao/blob/main/LICENSE",
      "license_note": "2026-11-05 核对根 LICENSE 为 Apache-2.0；LICENSE-ENTERPRISE 仍限制标记 @license Enterprise 的文件，不能将整个仓库概括为纯 Apache-2.0。GitHub API 识别为 NOASSERTION，正文及文件范围优先；archived=false。",
      "notes": "codemode 明确为 experimental；旧 Python interpreter 已移除，旧执行示例需核对版本。先前 subagents beta 针对 nao cloud，不表示所有自托管安装默认具备；新 release 不改变混合许可边界。",
      "last_verified": "2026-11-05",
      "release_updates": [
        {
          "version": "v0.3.29",
          "published_at": "2026-10-25T14:27:21Z",
          "url": "https://github.com/getnao/nao/releases/tag/v0.3.29"
        },
        {
          "version": "v0.3.30",
          "published_at": "2026-10-28T14:43:09Z",
          "url": "https://github.com/getnao/nao/releases/tag/v0.3.30"
        },
        {
          "version": "v0.3.31",
          "published_at": "2026-10-30T10:15:24Z",
          "url": "https://github.com/getnao/nao/releases/tag/v0.3.31"
        },
        {
          "version": "v0.3.32",
          "published_at": "2026-11-03T08:54:14Z",
          "url": "https://github.com/getnao/nao/releases/tag/v0.3.32"
        },
        {
          "version": "v0.3.33",
          "published_at": "2026-11-03T11:55:58Z",
          "url": "https://github.com/getnao/nao/releases/tag/v0.3.33"
        }
      ]
    },
    {
      "id": "databao-agent",
      "date": "2026-01-16",
      "date_basis": "released",
      "title": "databao-agent",
      "organization": "JetBrains",
      "url": "https://github.com/JetBrains/databao-agent",
      "section": "open-source",
      "language": "en",
      "status": "early",
      "tags": [
        "semantic-layer",
        "python",
        "eap"
      ]
    },
    {
      "id": "wrenai",
      "date": "2026-10-30",
      "date_basis": "released",
      "title": "WrenAI",
      "organization": "Canner",
      "url": "https://github.com/Canner/WrenAI",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "semantic-layer",
        "text-to-sql",
        "multi-license"
      ],
      "summary": "wren v0.17 增加 Ossie 上下文导入导出和 OpenMetadata 项目导入，让语义资产在目录与 Agent 查询层之间转换；同时修复派生指标、引用列与模型作用域问题，适合研究上下文互操作。",
      "date_evidence": "wren-v0.17.0 的 changelog 标为 2026-10-23，但 GitHub published_at 为 2026-10-27T03:02:08Z；wren-v0.17.1 published_at 为 2026-10-30T02:01:05Z，后者修复 DuckDB 2.0 Arrow table 兼容性。",
      "evidence_url": "https://github.com/Canner/WrenAI/releases/tag/wren-v0.17.1",
      "license": "Multi-license: Apache-2.0 code; CC-BY-4.0 docs; package manifests govern published package artifacts",
      "license_url": "https://github.com/Canner/WrenAI/blob/548eee3e864b11e7f11eee15ae82be178a295f22/LICENSE",
      "notes": "2026-11-06 核对发布、仓库状态、路径许可与架构原稿；发布为 wren-v0.17.1，技术文档固定至 548eee3e864b11e7f11eee15ae82be178a295f22，二者不是同一快照。",
      "last_verified": "2026-11-06",
      "release_updates": [
        {
          "version": "wren-v0.17.0",
          "published_at": "2026-10-27T03:02:08Z",
          "url": "https://github.com/Canner/WrenAI/releases/tag/wren-v0.17.0"
        },
        {
          "version": "wren-v0.17.1",
          "published_at": "2026-10-30T02:01:05Z",
          "url": "https://github.com/Canner/WrenAI/releases/tag/wren-v0.17.1"
        }
      ],
      "technical_material_url": "https://github.com/Canner/WrenAI/blob/548eee3e864b11e7f11eee15ae82be178a295f22/docs/core/reference/architecture.md",
      "component_type": "semantic-layer-and-tooling",
      "mechanism": "MDL项目定义业务模型和关联；CLI/SDK通过规划、语义展开与连接器执行，将知识文件和可重建记忆索引分开。",
      "use_case": "为现有Agent提供可版本化业务语义、SQL执行与查询证据。",
      "caveat": "架构文档仍将 golden NL-SQL eval 标为开发中；dry-plan 展示规划，dry-run 检查数据库执行条件，都不等于业务语义验收。行列安全、团队访问和商业交付能力不能由 CLI 存在直接推断。",
      "reading_path": [
        {
          "title": "官方快速开始",
          "url": "https://docs.getwren.ai/oss/get_started/quickstart",
          "focus": "用固定版本和样例先完成一次查询。"
        },
        {
          "title": "架构原稿",
          "url": "https://github.com/Canner/WrenAI/blob/548eee3e864b11e7f11eee15ae82be178a295f22/docs/core/reference/architecture.md",
          "focus": "区分技能、项目上下文、规划引擎与执行连接器。"
        },
        {
          "title": "MDL参考",
          "url": "https://github.com/Canner/WrenAI/blob/main/docs/core/reference/mdl.md",
          "focus": "追踪模型、关系、指标和知识目录的表示。"
        },
        {
          "title": "许可范围",
          "url": "https://github.com/Canner/WrenAI/blob/548eee3e864b11e7f11eee15ae82be178a295f22/LICENSE",
          "focus": "先按路径确认代码、文档和发布组件的许可。"
        }
      ],
      "verification_scope": "核对官方 release、仓库归档状态、所列根许可证及固定提交的技术文档；未部署运行、复现基准或审计依赖许可证。历史支持链接并非本轮全部重验。",
      "license_note": "根 LICENSE 按路径区分代码与文档；预置 AGPL 正文不表示现有组件均为 AGPL，发布包须核对各自 manifest。"
    },
    {
      "id": "db-gpt",
      "date": "2026-08-26",
      "date_basis": "released",
      "title": "DB-GPT",
      "organization": "eosphoros-ai",
      "url": "https://github.com/eosphoros-ai/DB-GPT",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "data-assistant",
        "platform",
        "mit"
      ],
      "summary": "DB-GPT把数据连接、ReAct规划、SQL/代码工具、Skills、上下文压缩与报告组织为应用平台；v0.8.2包含多文件分析、并行子任务和Agentic知识检索，本轮补齐规划与工具层的文档路线。",
      "date_evidence": "GitHub v0.8.2 published_at 为 2026-08-26T11:10:31Z；本轮技术文档固定至 9d50d0a55071a040a8cb7684b3b6acfa13cc7c47，与 release 提交不同，不把文档核验日期改写为版本发布日期。",
      "evidence_url": "https://github.com/eosphoros-ai/DB-GPT/releases/tag/v0.8.2",
      "license": "MIT",
      "license_url": "https://github.com/eosphoros-ai/DB-GPT/blob/9d50d0a55071a040a8cb7684b3b6acfa13cc7c47/LICENSE",
      "component_type": "agent-platform",
      "technical_material_url": "https://github.com/eosphoros-ai/DB-GPT/blob/9d50d0a55071a040a8cb7684b3b6acfa13cc7c47/docs/docs/agents/modules/context-management.md",
      "mechanism": "在模型调用前计量上下文，按观察截断、历史轮次裁剪、LLM 摘要和溢出重试逐层压缩；完整工具结果另存操作快照，提示保留可回查路径。",
      "use_case": "自托管分析工作区、SQL/代码混合任务，或研究长会话上下文管理。",
      "caveat": "sql_query 文档规定只读且只显示前 50 行，不能把显示截断当作完整结果，或把该工具的只读保证推广到 Python、Shell 等工具。上下文压缩会裁剪或概括历史，精确结果应回查快照；文档固定于 main 的后续提交，不等于 v0.8.2 的逐项验收。",
      "reading_path": [
        {
          "title": "项目与启动",
          "url": "https://github.com/eosphoros-ai/DB-GPT/blob/main/README.md",
          "focus": "确认安装方式、模型提供商和运行入口。"
        },
        {
          "title": "分析规划设计",
          "url": "https://github.com/eosphoros-ai/DB-GPT/blob/main/docs/data_analysis_planning_agent.md",
          "focus": "区分计划状态、工具接口与设计示例。"
        },
        {
          "title": "上下文管理",
          "url": "https://github.com/eosphoros-ai/DB-GPT/blob/9d50d0a55071a040a8cb7684b3b6acfa13cc7c47/docs/docs/agents/modules/context-management.md",
          "focus": "查看分层压缩、token预算和工具结果快照。"
        },
        {
          "title": "SQL工具契约",
          "url": "https://github.com/eosphoros-ai/DB-GPT/blob/9d50d0a55071a040a8cb7684b3b6acfa13cc7c47/docs/docs/agents/modules/resource/tools/sql-query.md",
          "focus": "核对只读范围与前50行显示截断。"
        }
      ],
      "last_verified": "2026-11-06",
      "release_updates": [
        {
          "version": "v0.8.2",
          "published_at": "2026-08-26T11:10:31Z",
          "url": "https://github.com/eosphoros-ai/DB-GPT/releases/tag/v0.8.2"
        }
      ],
      "verification_scope": "核对官方 release、仓库归档状态、所列根许可证及固定提交的技术文档；未部署运行、复现基准或审计依赖许可证。历史支持链接并非本轮全部重验。"
    },
    {
      "id": "dataagentbench-repo",
      "date": "2026-10-07",
      "date_basis": "updated",
      "title": "DataAgentBench",
      "organization": "ucbepic",
      "url": "https://github.com/ucbepic/DataAgentBench",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "benchmark",
        "multi-database",
        "no-license"
      ]
    },
    {
      "id": "deepeye",
      "date": "2026-05",
      "date_basis": "released",
      "title": "DeepEye",
      "organization": "HKUSTDial",
      "url": "https://github.com/HKUSTDial/DeepEye",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "typed-dag",
        "validator",
        "optimizer",
        "apache-2.0"
      ]
    },
    {
      "id": "openai-cookbook-data-analyst",
      "date": "2026-09-16",
      "date_basis": "published",
      "title": "Data Analyst app walkthrough",
      "organization": "OpenAI Cookbook",
      "url": "https://github.com/openai/openai-cookbook/tree/main/examples/agents_api/apps/data_analyst",
      "section": "open-source",
      "language": "en",
      "status": "published",
      "tags": [
        "tools",
        "read-only",
        "memory",
        "sql"
      ]
    },
    {
      "id": "datacross",
      "date": "2026-01-29",
      "date_basis": "published",
      "title": "DataCross: A Unified Benchmark and Agent Framework for Cross-Modal Heterogeneous Data Analysis",
      "organization": "Research paper",
      "url": "https://arxiv.org/abs/2601.21403",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "multimodal",
        "multi-agent"
      ]
    },
    {
      "id": "data-agents-levels",
      "date": "2026-02-04",
      "date_basis": "published",
      "title": "Data Agents: Levels, State of the Art, and Open Problems",
      "organization": "SIGMOD 2026 Tutorial",
      "url": "https://arxiv.org/abs/2602.04261",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "taxonomy",
        "survey",
        "autonomy-levels"
      ]
    },
    {
      "id": "dab-paper",
      "date": "2026-03-21",
      "date_basis": "published",
      "title": "Can AI Agents Answer Your Data Questions? A Benchmark for Data Agents",
      "organization": "UC Berkeley EPIC et al.",
      "url": "https://arxiv.org/abs/2603.20576",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "enterprise",
        "multi-database"
      ]
    },
    {
      "id": "agenticdatabench-paper",
      "date": "2026-07-02",
      "date_basis": "published",
      "title": "AgenticDataBench: A Comprehensive Benchmark for Data Agents",
      "organization": "Research paper",
      "url": "https://arxiv.org/abs/2607.01647",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "data-science",
        "skills"
      ]
    },
    {
      "id": "unidatabench",
      "date": "2026-07",
      "date_basis": "published",
      "title": "UniDataBench: Evaluating Data Analytics Agents Across Structured and Unstructured Data",
      "organization": "ACL 2026",
      "url": "https://aclanthology.org/2026.acl-long.1556/",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "structured-data",
        "unstructured-data"
      ]
    },
    {
      "id": "datasage",
      "date": "2026-07",
      "date_basis": "published",
      "title": "DataSage: Multi-agent Collaboration for Insight Discovery",
      "organization": "ACL Findings 2026",
      "url": "https://aclanthology.org/2026.findings-acl.309/",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "multi-agent",
        "insight-discovery",
        "debate"
      ]
    },
    {
      "id": "dataspace-paper",
      "date": "2026-08-04",
      "date_basis": "published",
      "title": "DataSpace: Benchmarking Data Agents for Verifiable Analytics over Heterogeneous Workspaces",
      "organization": "HKUST DIAL et al.",
      "url": "https://arxiv.org/abs/2608.03451",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "multimodal",
        "verifiable"
      ]
    },
    {
      "id": "enterprise-analytics-evaluation",
      "date": "2026-08-28",
      "date_basis": "published",
      "title": "Evaluating Enterprise Analytics Agents: An End-to-End, Trace-Backed Methodology",
      "organization": "KDD 2026 Workshop",
      "url": "https://arxiv.org/abs/2609.09182",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "evaluation",
        "traces",
        "enterprise"
      ]
    },
    {
      "id": "agentic-data-systems",
      "date": "2026-09-21",
      "date_basis": "published",
      "title": "Data Agents: Agentic Data Systems",
      "organization": "IEEE TKDE",
      "url": "https://arxiv.org/abs/2609.24137",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "architecture",
        "semantic-operators",
        "memory"
      ]
    },
    {
      "id": "learned-enterprise-comprehension",
      "date": "2026-09-21",
      "date_basis": "published",
      "title": "Learned Enterprise Data Comprehension: Compression and Routing for Data Agents",
      "organization": "Research preprint",
      "url": "https://arxiv.org/abs/2609.25286",
      "section": "research",
      "language": "en",
      "status": "early",
      "tags": [
        "context",
        "routing",
        "compression"
      ]
    },
    {
      "id": "fabric-data-agent-concepts",
      "date": "2026-05-11",
      "date_basis": "updated",
      "title": "Fabric data agent concepts",
      "organization": "Microsoft Learn",
      "url": "https://learn.microsoft.com/en-us/fabric/data-science/concept-data-agent",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "fabric",
        "orchestrator",
        "purview",
        "semantic-model"
      ]
    },
    {
      "id": "fabric-data-agent-evaluation",
      "date": "2026-09-01",
      "date_basis": "updated",
      "title": "Evaluate your data agent",
      "organization": "Microsoft Learn",
      "url": "https://learn.microsoft.com/fabric/data-science/evaluate-data-agent",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "fabric",
        "sdk",
        "evaluation"
      ]
    },
    {
      "id": "alibaba-dms-openapi",
      "date": "2026-06-05",
      "date_basis": "updated",
      "title": "Data Agent OpenAPI integration guide",
      "organization": "Alibaba Cloud DMS",
      "url": "https://www.alibabacloud.com/help/en/dms/data-agent-openapi-integration-guide",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "api",
        "integration",
        "architecture"
      ]
    },
    {
      "id": "alibaba-dataworks-agent",
      "date": "2026-10-29",
      "date_basis": "updated",
      "title": "Data Agent 简介",
      "organization": "Alibaba Cloud DataWorks",
      "url": "https://help.aliyun.com/zh/dataworks/user-guide/new-data-agent",
      "section": "living-doc",
      "language": "zh",
      "status": "living-doc",
      "tags": [
        "context-graph",
        "skills",
        "data-engineering",
        "governance",
        "workflow",
        "human-in-the-loop"
      ],
      "summary": "中文简介展开八层架构：专家套件、Context Graph、企业级 Agent 管控与多计算引擎协同，并补充自动化、工作流、IM 协作和审批模式。适合借鉴平台原生数据资产与生产权限的结合；功能受版本、地域和实际配置约束，完全访问模式不等于安全默认。",
      "date_evidence": "中文原页 meta name=last-modified 为 2026-10-29T16:06:28+08:00。11月2日复查时英文页仍显示 Oct 09；11月5日重开英文页现显示 Last Updated: Oct 29, 2026，并已包含八层架构。本轮仅校正语言版本状态，不把10月29日的内容计为11月新增。",
      "evidence_url": "https://help.aliyun.com/zh/dataworks/user-guide/new-data-agent",
      "related_materials": [
        {
          "title": "DataWorks: Data Agent introduction — English documentation",
          "url": "https://www.alibabacloud.com/help/en/dataworks/user-guide/new-data-agent",
          "date": "2026-10-29",
          "note": "2026-11-05 复查：英文页当前显示 Oct 29, 2026，含八层架构、Context Graph 与企业治理；替换上次保留的英文旧页状态，不另增条目。"
        }
      ],
      "last_verified": "2026-11-05"
    },
    {
      "id": "alibaba-sls-dataagent",
      "date": "2026-09-10",
      "date_basis": "updated",
      "title": "DataAgent core concepts",
      "organization": "Alibaba Cloud SLS",
      "url": "https://www.alibabacloud.com/help/en/sls/sls-dataagent-core-concept",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "context-model",
        "logging"
      ]
    },
    {
      "id": "alibaba-maxcompute-dataagent",
      "date": "2026-10-12",
      "date_basis": "updated",
      "title": "Multi-source, multi-engine DataAgent",
      "organization": "Alibaba Cloud MaxCompute",
      "url": "https://www.alibabacloud.com/help/en/maxcompute/user-guide/multi-source-multi-engine-dataagent",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "multi-source",
        "multi-engine",
        "files"
      ]
    },
    {
      "id": "alibaba-analyticdb-dataagent",
      "date": "2026-09-14",
      "date_basis": "updated",
      "title": "Data Agent architecture",
      "organization": "Alibaba Cloud AnalyticDB",
      "url": "https://www.alibabacloud.com/help/en/analyticdb/analyticdb-for-postgresql/data-agent/",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "architecture",
        "database",
        "analytics"
      ]
    },
    {
      "id": "china-data-agent-report-2026",
      "date": "2026-08-17",
      "date_basis": "published",
      "title": "数据智能体研究报告（2026）（征求意见稿）",
      "organization": "中国互联网协会人工智能工作委员会 / 中国信通院",
      "url": "https://www.isc.org.cn/article/30075177847877632.html",
      "section": "research",
      "language": "zh",
      "status": "published",
      "tags": [
        "report",
        "taxonomy",
        "reference-architecture",
        "draft"
      ]
    },
    {
      "id": "shopee-qcon-transcript",
      "date": "2026-07-14",
      "date_basis": "published",
      "title": "从 AI 取数到智能分析：企业级数据 Agent 的多阶段演进与工程化落地（全文实录）",
      "organization": "Shopee / InfoQ",
      "url": "https://www.infoq.cn/article/LfO7mzdhxMWTtICFuTNb",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "transcript",
        "multi-agent",
        "file-as-context",
        "evaluation"
      ]
    },
    {
      "id": "tencent-databuddy-runtime",
      "date": "2026-09-01",
      "date_basis": "published",
      "title": "数据语义驱动的企业 Agent Runtime 实践",
      "organization": "腾讯 DataBuddy",
      "url": "https://developer.cloud.tencent.cn/article/2735067",
      "section": "architecture",
      "language": "zh",
      "status": "published",
      "tags": [
        "runtime",
        "semantic-layer",
        "sandbox",
        "human-in-the-loop"
      ]
    },
    {
      "id": "yunqi-production-data-agent",
      "date": "2026-09-09",
      "date_basis": "published",
      "title": "从 Tool Calling 到 Data Agent：企业数据智能体的生产化路径",
      "organization": "云器科技",
      "url": "https://www.yunqi.tech/resource/blogs/enterprise-data-agent-production",
      "section": "architecture",
      "language": "zh",
      "status": "published",
      "tags": [
        "semantic-layer",
        "sandbox",
        "traces",
        "evaluation"
      ]
    },
    {
      "id": "tencent-cloud-data-guide-2026",
      "date": "2026-09-10",
      "date_basis": "published",
      "title": "2026 腾讯云大数据年度精选技术实践指南",
      "organization": "腾讯云大数据 / InfoQ",
      "url": "https://www.infoq.cn/minibook/W2lxNULd8rJGsXWjEgMX",
      "section": "architecture",
      "language": "zh",
      "status": "published",
      "tags": [
        "report",
        "databuddy",
        "data-platform",
        "analytics"
      ]
    },
    {
      "id": "dacon-beijing-2026-agenda",
      "date": "2026-10-23/2026-10-24",
      "date_basis": "event",
      "title": "DACon 北京 2026 Data Agent 相关议程",
      "organization": "DACon 北京",
      "url": "https://www.bagevent.com/event/9196611",
      "section": "talk",
      "language": "zh",
      "status": "agenda",
      "tags": [
        "agenda",
        "semantic-layer",
        "ontology",
        "evaluation"
      ],
      "last_verified": "2026-11-02",
      "material_review": {
        "checked_on": "2026-11-02",
        "status": "event-held-material-unverified",
        "evidence_url": "https://www.bagevent.com/event/9196611",
        "note": "官方活动页仍可打开且有详细技术提纲，但本次未核验统一公开PPT/录像。云器会后复盘作为独立新文收录，不能据此把全部DACon议程升级为published。"
      }
    },
    {
      "id": "sigmod-data-agents-slides",
      "date": "2026-06-03",
      "date_basis": "event",
      "title": "Data Agents: From Text-to-SQL to Autonomous Data Intelligence — Tutorial Slides",
      "organization": "SIGMOD 2026",
      "url": "https://2026.sigmod.org/slides/DataAgents-SIGMOD2026.pdf",
      "section": "talk",
      "language": "en",
      "status": "published",
      "tags": [
        "tutorial",
        "taxonomy",
        "slides"
      ]
    },
    {
      "id": "data-agents-under-attack",
      "date": "2026-06-07",
      "date_basis": "published",
      "title": "Data Agents Under Attack: A Systematic Study of Security Vulnerabilities",
      "organization": "Research paper",
      "url": "https://arxiv.org/abs/2606.08661",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "security",
        "threat-model",
        "attacks"
      ]
    },
    {
      "id": "dataclaweval",
      "date": "2026-07-30",
      "date_basis": "published",
      "title": "DataClawEval: Benchmarking Data Agents on End-to-End Data Engineering",
      "organization": "Research paper",
      "url": "https://arxiv.org/abs/2607.28033",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "data-engineering",
        "pyspark",
        "sql"
      ]
    },
    {
      "id": "kddcup-2026-data-agents",
      "date": "2026-08-12",
      "date_basis": "event",
      "title": "KDD Cup 2026: Data Agents for Complex Data Analysis",
      "organization": "KDD Cup 2026",
      "url": "https://dataagent.top/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "competition",
        "benchmark",
        "multimodal"
      ]
    },
    {
      "id": "dbt-agent-skills-guide",
      "date": "2026-02-05",
      "date_basis": "published",
      "title": "Make your AI better at data work with dbt's agent skills",
      "organization": "dbt Labs",
      "url": "https://docs.getdbt.com/blog/dbt-agent-skills",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "skills",
        "analytics-engineering",
        "semantic-layer",
        "workflow"
      ],
      "summary": "把建模、测试、语义层和迁移流程封装为按任务加载的 Skills；附接入与端到端工作流示例。",
      "date_evidence": "Article header: February 5, 2026.",
      "evidence_url": "https://docs.getdbt.com/blog/dbt-agent-skills"
    },
    {
      "id": "dbt-wizard-agent-harness",
      "date": "2026-10-14",
      "date_basis": "published",
      "title": "Inside the dbt Wizard harness: building an agent that thinks in DAGs",
      "organization": "dbt Labs",
      "url": "https://docs.getdbt.com/blog/wizard-agent-harness",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "harness",
        "dag",
        "validation",
        "cost-control"
      ],
      "summary": "拆解 DAG 上下文、编译反馈、专用验证 Agent、批准范围内构建与 deferral；将数据差异交给人工评审。",
      "date_evidence": "Article header: October 14, 2026.",
      "evidence_url": "https://docs.getdbt.com/blog/wizard-agent-harness"
    },
    {
      "id": "dbt-context-engineering-guide",
      "date": "2026-09-16",
      "date_basis": "published",
      "title": "dbt_context_engineering: modeling the context your AI agents read",
      "organization": "dbt Labs",
      "url": "https://docs.getdbt.com/blog/dbt-context-engineering",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "unstructured-data",
        "testing",
        "lineage"
      ],
      "summary": "将工单、通话和合同等非结构化材料建模为 Agent 可检索、可引用的上下文，沿用版本、测试与血缘管理。",
      "date_evidence": "Article header: September 16, 2026.",
      "evidence_url": "https://docs.getdbt.com/blog/dbt-context-engineering"
    },
    {
      "id": "dbt-ade-bench-methodology",
      "date": "2026-10-14",
      "date_basis": "updated",
      "title": "Introducing ADE-bench: measuring how AI agents perform data work",
      "organization": "dbt Labs",
      "url": "https://www.getdbt.com/blog/ade-bench-dbt-data-benchmarking",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "dbt",
        "sandbox",
        "data-engineering"
      ],
      "summary": "以真实 dbt 项目、DuckDB 数据与 Docker 环境测试模糊修复任务，并联合观察测试通过率、成本和运行时。",
      "date_evidence": "Current article header: Last edited on Oct 14, 2026; includes dbt v1/v2 and MCP comparison.",
      "evidence_url": "https://www.getdbt.com/blog/ade-bench-dbt-data-benchmarking",
      "notes": "Use updated, not the original publication date. Reported scores are configuration-specific publisher results."
    },
    {
      "id": "cube-agentic-analytics-harness",
      "date": "2026-09-03",
      "date_basis": "published",
      "title": "Building an Agentic Analytics Harness",
      "organization": "Cube",
      "url": "https://cube.dev/blog/building-an-agentic-analytics-harness",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "harness",
        "semantic-sql",
        "tool-design",
        "context-budget"
      ],
      "summary": "从 Semantic SQL、可恢复工具错误、结果截断提示、持久产物和权限过滤检索拆解分析 Agent 的运行机制。",
      "date_evidence": "Article header: September 3, 2026.",
      "evidence_url": "https://cube.dev/blog/building-an-agentic-analytics-harness"
    },
    {
      "id": "cube-agent-skills",
      "date": "2026-06-02",
      "date_basis": "published",
      "title": "Introducing Cube Agent Skills",
      "organization": "Cube",
      "url": "https://cube.dev/blog/introducing-cube-agent-skills",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "skills",
        "workflow",
        "version-control",
        "semantic-layer"
      ],
      "summary": "把周报、流失分析和模型变更流程存为模型项目中的 Markdown Skills，经分支、评审与部署复用。",
      "date_evidence": "Article header: June 2, 2026.",
      "evidence_url": "https://cube.dev/blog/introducing-cube-agent-skills",
      "notes": "The article explicitly lists first-release limitations; not a guarantee of deterministic answers."
    },
    {
      "id": "cube-context-needs-semantic-layer",
      "date": "2026-07-10",
      "date_basis": "published",
      "title": "The Context Layer Needs a Semantic Layer",
      "organization": "Cube",
      "url": "https://cube.dev/blog/the-context-layer-needs-a-semantic-layer",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "semantic-layer",
        "permissions",
        "memory"
      ],
      "summary": "区分可执行语义、描述性业务知识与历史记忆，说明指标和权限如何在查询编译时落实；属于厂商架构观点。",
      "date_evidence": "Article header: July 10, 2026.",
      "evidence_url": "https://cube.dev/blog/the-context-layer-needs-a-semantic-layer"
    },
    {
      "id": "hex-context-studio",
      "date": "2026-01-28",
      "date_basis": "published",
      "title": "Introducing Context Studio",
      "organization": "Hex",
      "url": "https://hex.tech/blog/introducing-context-studio/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "semantic-layer",
        "observability",
        "governance"
      ],
      "summary": "统一管理数据资产认证、确定性指标模型和业务指南；从真实对话诊断错误，预览上下文修改，再版本化发布。",
      "notes": "产品架构分享，未提供独立量化评测。",
      "date_evidence": "原文标题下明确标注 January 28, 2026。",
      "evidence_url": "https://hex.tech/blog/introducing-context-studio/"
    },
    {
      "id": "hex-context-suggestions",
      "date": "2026-04-23",
      "date_basis": "published",
      "title": "Introducing Context Suggestions",
      "organization": "Hex",
      "url": "https://hex.tech/blog/introducing-context-suggestions/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "memory",
        "review-agent",
        "feedback"
      ],
      "summary": "Review Agent 复盘代码、工具调用和分析输出，将澄清、歧义与缺失定义聚类成问题，并提出上下文变更供团队审核。",
      "notes": "展示分析、复盘与上下文更新闭环；未证明自动学习必然提升正确率。",
      "date_evidence": "原文标题下明确标注 April 23, 2026。",
      "evidence_url": "https://hex.tech/blog/introducing-context-suggestions/"
    },
    {
      "id": "hex-databench",
      "date": "2026-08-13",
      "date_basis": "published",
      "title": "Introducing DataBench",
      "organization": "Hex",
      "url": "https://hex.tech/blog/databench-agentic-analytics-benchmark/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "analytics",
        "ambiguous-intent",
        "abstention"
      ],
      "summary": "在混乱的合成企业数据仓库中设置 100 个问答与开放分析任务，覆盖指标歧义、缺失证据和拒绝作答。",
      "notes": "任务保持私有，不能视为完整公开可复现基准。评分查看对话、产物、真值与 rubric，LLM judge 三次投票仍有噪声。",
      "date_evidence": "原文作者与分类栏下明确标注 August 13, 2026。",
      "evidence_url": "https://hex.tech/blog/databench-agentic-analytics-benchmark/"
    },
    {
      "id": "preset-chatbot-production",
      "date": "2026-05-05",
      "date_basis": "published",
      "title": "Shipping Preset Chatbot: From AI Prototype to Production",
      "organization": "Preset",
      "url": "https://preset.io/blog/preset-chatbot-technical-deep-dive/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "langgraph",
        "streaming",
        "production"
      ],
      "summary": "MCP 统一多端能力，LangGraph 持久化对话；详述异步 Agent 接入同步 Web 框架、独立连接池、SSE 桥接和工具错误处理。",
      "notes": "真实产品工程经验，未给出准确率或并发容量 benchmark。",
      "date_evidence": "原文作者栏下明确标注 May 05, 2026。",
      "evidence_url": "https://preset.io/blog/preset-chatbot-technical-deep-dive/"
    },
    {
      "id": "preset-agent-skills-evaluation",
      "date": "2026-06-11",
      "date_basis": "published",
      "title": "Preset Agent Skills: Giving Your AI Agent Real Preset Expertise",
      "organization": "Preset",
      "url": "https://preset.io/blog/building-preset-agent-skills/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "skills",
        "evaluation",
        "tool-selection",
        "governance"
      ],
      "summary": "以 31 个真实 BI 操作任务成对比较有无 Skills，重复运行并确定性检查工具选择、执行顺序、审批节点与秘密保护。",
      "notes": "作者报告 30/31 项持平或改善、23 项明显改善；上下文 token 约增至 1.6 倍。结果仅适用于其 Preset 任务集。",
      "date_evidence": "原文作者栏下明确标注 June 11, 2026。",
      "evidence_url": "https://preset.io/blog/building-preset-agent-skills/"
    },
    {
      "id": "superset-mcp-technical-deep-dive",
      "date": "2026-03-31",
      "date_basis": "published",
      "title": "Apache Superset MCP Service: A Technical Deep Dive",
      "organization": "Preset / Apache Superset",
      "url": "https://preset.io/blog/superset-mcp-service-deep-dive/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "rbac",
        "preview-first",
        "token-budget"
      ],
      "summary": "MCP 直接复用 Superset Python 库与 DAO；图表修改默认预览，持久化显式执行；沿用 RBAC/RLS 并限制响应 token。",
      "notes": "BI 工具服务层参考实现，包含部署、鉴权、工具和中间件说明；不是完整自主数据分析 Agent。",
      "date_evidence": "原文作者栏下明确标注 March 31, 2026。",
      "evidence_url": "https://preset.io/blog/superset-mcp-service-deep-dive/"
    },
    {
      "id": "preset-mcp-enterprise",
      "date": "2026-03-31",
      "date_basis": "published",
      "title": "Preset MCP: From Open Source to Enterprise",
      "organization": "Preset",
      "url": "https://preset.io/blog/preset-mcp-open-source-to-enterprise/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "multi-tenant",
        "authentication",
        "observability"
      ],
      "summary": "通过外围中间件为 MCP 服务加入工作区隔离、逐请求数据库绑定、OAuth、弹性部署及可观测性，同时复用原有权限检查。",
      "notes": "与 Superset MCP 基础架构文章互补，重点是单租户服务的企业化运行。",
      "date_evidence": "原文作者栏下明确标注 March 31, 2026。",
      "evidence_url": "https://preset.io/blog/preset-mcp-open-source-to-enterprise/"
    },
    {
      "id": "motherduck-guides-context",
      "date": "2026-07-29",
      "date_basis": "published",
      "title": "Context belongs in the warehouse",
      "organization": "MotherDuck",
      "url": "https://motherduck.com/blog/context-belongs-in-the-warehouse/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context",
        "warehouse",
        "mcp",
        "semantic-layer"
      ],
      "summary": "将指标定义、关联规则和业务陷阱存为仓内 Markdown Guides，以 SQL 管理与版本化；Agent 经 MCP 按需读取，组织和个人上下文分层分发。",
      "notes": "DABStep 改善数字来自厂商特定配置，不应推广至任意企业仓库。采用正式主站 URL，避免搜索返回的 staging 域名。",
      "date_evidence": "正式主站原文标题下明确标注 Jul 29, 2026。",
      "evidence_url": "https://motherduck.com/blog/context-belongs-in-the-warehouse/"
    },
    {
      "id": "clickhouse-platform-for-agents",
      "date": "2026-03-27",
      "date_basis": "published",
      "title": "How we’re building a data platform for a new user: agents",
      "organization": "ClickHouse",
      "url": "https://clickhouse.com/blog/building-a-data-platform-for-agents",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "observability",
        "data-platform",
        "analytics"
      ],
      "summary": "以 LibreChat、ClickHouse＋MCP 和 Langfuse 构成交互、数据及观测三层架构，讨论探索型查询、上下文发现和任务级权限。",
      "notes": "正文包含路线图，计划中的能力不能写成已发布。文中引用的 DWAINE 完整构建文章实际发表于 2025-11-12。",
      "date_evidence": "原文作者栏下明确标注 Mar 27, 2026。",
      "evidence_url": "https://clickhouse.com/blog/building-a-data-platform-for-agents"
    },
    {
      "id": "clickhouse-agents-beta",
      "date": "2026-06-09",
      "date_basis": "published",
      "title": "ClickHouse Agents: Claude-powered agentic analytics, now in public beta",
      "organization": "ClickHouse",
      "url": "https://clickhouse.com/blog/clickhouse-agents-beta",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "sandbox",
        "multi-agent",
        "mcp",
        "artifacts"
      ],
      "summary": "将原生查询、MCP、沙箱代码执行、Skills、记忆、子 Agent 与分析产物整合到 Cloud RBAC 下的 LibreChat 托管运行时。",
      "notes": "公开 Beta 产品架构介绍，适合研究托管运行时能力组成；不是独立性能或安全评测。",
      "date_evidence": "原文作者栏下明确标注 Jun 9, 2026。",
      "evidence_url": "https://clickhouse.com/blog/clickhouse-agents-beta"
    },
    {
      "id": "xorq-orientation-over-reasoning",
      "date": "2026-04-22",
      "date_basis": "published",
      "title": "Orientation Over Reasoning",
      "organization": "Xorq",
      "url": "https://xorq.dev/blog/orientation-over-reasoning/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "semantic-catalog",
        "paired-evaluation",
        "context",
        "reproducibility"
      ],
      "summary": "固定模型、执行预算与评分器，在 450 个 DABStep 问题上比较原始文件与 33 个命名语义表达式目录，公开环境、日志及评分流程。",
      "notes": "作者明确承认题目已公开，绝对分数不宜作排行榜结论；50%→84% 仅是同条件对照结果。",
      "date_evidence": "原文作者行明确标注 April 22, 2026。",
      "evidence_url": "https://xorq.dev/blog/orientation-over-reasoning/"
    },
    {
      "id": "dot-dabstep-configuration",
      "date": "2026-04-30",
      "date_basis": "published",
      "title": "From 10% to 90% on Adyen’s DABstep: configuring a data warehouse agent",
      "organization": "Dot",
      "url": "https://www.getdot.ai/blog/from-10-to-90-percent-on-adyens-dabstep",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "skills",
        "configuration",
        "feedback",
        "benchmark"
      ],
      "summary": "将业务约定与重复分析路径沉淀为配置，再让 Agent 读取失败、提出修改并重新测量；用版本化实验记录保留成功和回退路径。",
      "notes": "原文限定为 30 个代表任务、14 个配置版本，90% 不是整个 DABStep 得分；旧模型和人工基线展示并非严格同配置比较。",
      "date_evidence": "原文标题下明确标注 April 30, 2026 by Alex Tatarinov。",
      "evidence_url": "https://www.getdot.ai/blog/from-10-to-90-percent-on-adyens-dabstep"
    },
    {
      "id": "fabric-data-agent-runtime",
      "date": "2026-10-13",
      "date_basis": "updated",
      "title": "Fabric data agent runtime",
      "organization": "Microsoft Learn",
      "url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-runtime",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "fabric",
        "runtime",
        "planning",
        "release-management"
      ],
      "summary": "通过标准与预览运行时理解编排、路由和查询生成的发布边界，可借鉴升级验证、回归评测和生产变更流程。",
      "date_evidence": "页面底部标注 Last updated on 2026-10-13。",
      "evidence_url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-runtime"
    },
    {
      "id": "fabric-data-agent-mcp-server",
      "date": "2026-09-04",
      "date_basis": "updated",
      "title": "Data agent as Model Context Protocol server",
      "organization": "Microsoft Learn",
      "url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-mcp-server",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "fabric",
        "mcp",
        "authentication",
        "tool-routing"
      ],
      "summary": "把已发布数据 Agent 暴露为单个 MCP 工具，附 Python 和编辑器接入；重点关注工具描述与 Bearer token 认证。",
      "date_evidence": "页面底部标注 Last updated on 2026-09-04。",
      "evidence_url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-mcp-server"
    },
    {
      "id": "fabric-foundry-fabric-iq-integration",
      "date": "2026-09-01",
      "date_basis": "updated",
      "title": "Add Fabric data agents to a Foundry agent with Fabric IQ (preview)",
      "organization": "Microsoft Learn",
      "url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-foundry-fabric-iq",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "fabric",
        "multi-agent",
        "mcp",
        "orchestration"
      ],
      "summary": "经 OneLake Catalog 发现多个领域数据 Agent，再由 Foundry 统一调用；包含跨 Agent 问题测试和长任务配置。",
      "date_evidence": "页面底部标注 Last updated on 2026-09-01。",
      "evidence_url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-foundry-fabric-iq"
    },
    {
      "id": "fabric-query-generation-best-practices",
      "date": "2026-08-28",
      "date_basis": "updated",
      "title": "Best practices for improving data agent query generation",
      "organization": "Microsoft Learn",
      "url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-configuration-best-practices",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "context",
        "text-to-sql",
        "table-grain",
        "join"
      ],
      "summary": "将上下文落实到 Schema 选择、对象描述、跨表规则和查询示例，尤其适合借鉴表粒度、先聚合后连接、财年及单位说明。",
      "date_evidence": "页面底部标注 Last updated on 2026-08-28。",
      "evidence_url": "https://learn.microsoft.com/en-us/fabric/data-science/data-agent-configuration-best-practices"
    },
    {
      "id": "google-bigquery-colab-data-science-agent",
      "date": "2026-10-20",
      "date_basis": "updated",
      "title": "Use the Colab Enterprise Data Science Agent with BigQuery",
      "organization": "Google Cloud",
      "url": "https://docs.cloud.google.com/bigquery/docs/colab-data-science-agent",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "bigquery",
        "notebook",
        "data-science",
        "python"
      ],
      "summary": "演示从 BigQuery 或 CSV 进入 Notebook 分析，覆盖探索、缺失值、预测与可视化，并可比较 SQL、DataFrames 和 PySpark 路径。",
      "date_evidence": "页面底部标注 Last updated 2026-10-20 UTC。",
      "evidence_url": "https://docs.cloud.google.com/bigquery/docs/colab-data-science-agent"
    },
    {
      "id": "google-data-engineering-agent-pipelines",
      "date": "2026-10-20",
      "date_basis": "updated",
      "title": "Use the Data Engineering Agent to build and modify data pipelines",
      "organization": "Google Cloud",
      "url": "https://docs.cloud.google.com/bigquery/docs/data-engineering-agent-pipelines",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "bigquery",
        "data-engineering",
        "dataform",
        "human-in-the-loop"
      ],
      "summary": "用自然语言生成和审阅 Dataform SQLX 流水线，包含预览与提示示例；明确区分生成代码、应用变更和执行流水线。",
      "date_evidence": "页面底部标注 Last updated 2026-10-20 UTC。",
      "evidence_url": "https://docs.cloud.google.com/bigquery/docs/data-engineering-agent-pipelines"
    },
    {
      "id": "google-data-engineering-agent-a2a-api",
      "date": "2026-10-13",
      "date_basis": "updated",
      "title": "Data Engineering Agent API: Build data pipelines with an agent",
      "organization": "Google Cloud",
      "url": "https://docs.cloud.google.com/gemini/data-agents/data-engineering-agent/api-overview",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "a2a",
        "api",
        "data-engineering",
        "streaming"
      ],
      "summary": "给出基于 A2A 的 Agent Card、流式与同步消息接口，可用于将数据工程 Agent 接入自有应用并处理权限和完成状态。",
      "date_evidence": "页面底部标注 Last updated 2026-10-13 UTC。",
      "evidence_url": "https://docs.cloud.google.com/gemini/data-agents/data-engineering-agent/api-overview"
    },
    {
      "id": "aws-data-agent-trusted-identity-propagation",
      "date": "2026-10-06",
      "date_basis": "published",
      "title": "Identity-aware AI data agents with AWS Lake Formation and Trusted Identity Propagation",
      "organization": "AWS Security Blog",
      "url": "https://aws.amazon.com/blogs/security/identity-aware-ai-data-agents-with-aws-lake-formation-and-trusted-identity-propagation/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "identity",
        "lake-formation",
        "agentcore",
        "audit"
      ],
      "summary": "贯穿 AgentCore、MCP Gateway、Lambda 和 Athena 传递真实用户身份，由 Lake Formation 授权并用 CloudTrail 核验。",
      "date_evidence": "文章署名行标注 on 06 OCT 2026。",
      "evidence_url": "https://aws.amazon.com/blogs/security/identity-aware-ai-data-agents-with-aws-lake-formation-and-trusted-identity-propagation/"
    },
    {
      "id": "snowflake-cortex-agent-from-scratch",
      "date": "2026-03-12",
      "date_basis": "updated",
      "title": "Build a Cortex Agent from Scratch with Snowflake",
      "organization": "Snowflake Developers",
      "url": "https://www.snowflake.com/en/developers/guides/build-a-cortex-agent-from-scratch-with-snowflake/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "semantic-view",
        "rag",
        "structured-data",
        "unstructured-data"
      ],
      "summary": "从样例数据、语义视图和 Cortex Search 建到双工具 Agent，再用 SQL/Python 测试，复现指标与产品文档的统一问答。",
      "date_evidence": "教程结尾标注 Updated Mar 12, 2026。",
      "evidence_url": "https://www.snowflake.com/en/developers/guides/build-a-cortex-agent-from-scratch-with-snowflake/"
    },
    {
      "id": "snowflake-cortex-agent-sdk-quickstart",
      "date": "2026-10-09",
      "date_basis": "updated",
      "title": "Getting Started with the Cortex Agent SDK",
      "organization": "Snowflake Developers",
      "url": "https://www.snowflake.com/en/developers/guides/getting-started-with-cortex-agent-sdk/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "sdk",
        "streaming",
        "sandbox",
        "data-operations"
      ],
      "summary": "五个递进示例覆盖流式结果、多轮会话、后台运行与代码沙箱，最终组成检查流水线、数据质量并按需诊断事故的运维 Agent。",
      "date_evidence": "教程结尾标注 Updated Oct 9, 2026。",
      "evidence_url": "https://www.snowflake.com/en/developers/guides/getting-started-with-cortex-agent-sdk/"
    },
    {
      "id": "databricks-genie-test-monitor",
      "date": "2026-09-17",
      "date_basis": "updated",
      "title": "Test and monitor a Genie Agent",
      "organization": "Databricks Documentation",
      "url": "https://docs.databricks.com/aws/en/genie-agents/monitor",
      "section": "evaluation",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "benchmark",
        "sql",
        "llm-judge",
        "feedback"
      ],
      "summary": "串联线上反馈、人工纠错与离线基准；区分 Chat 的 SQL 结果比较和 Agent 的 LLM 评判，适合设计持续维护的验收集。",
      "date_evidence": "AWS 版本页面顶部标注 Last updated on Sep 17, 2026。",
      "evidence_url": "https://docs.databricks.com/aws/en/genie-agents/monitor"
    },
    {
      "id": "databricks-genie-one-mcp",
      "date": "2026-10-16",
      "date_basis": "updated",
      "title": "Genie One MCP server",
      "organization": "Databricks Documentation",
      "url": "https://docs.databricks.com/aws/en/agents/mcp-tools/genie-mcp",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "mcp",
        "semantic-layer",
        "governance",
        "citations"
      ],
      "summary": "经 Unity Gateway 将 Genie 暴露为对话分析工具，按 Unity Catalog 权限访问数据并返回来源链接，适合研究具备业务语义的工具接口。",
      "date_evidence": "页面顶部标注 Last updated on Oct 16, 2026。",
      "evidence_url": "https://docs.databricks.com/aws/en/agents/mcp-tools/genie-mcp"
    },
    {
      "id": "datus-agent",
      "date": "2026-10-28",
      "date_basis": "released",
      "title": "Datus Agent",
      "organization": "Datus AI",
      "url": "https://github.com/Datus-ai/Datus-agent",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "context",
        "semantic-layer",
        "sql",
        "lineage",
        "mcp"
      ],
      "summary": "v0.4.4 将本体问答、语义模型和数据源映射接入 OntologyAnalyst，补充 Athena/HANA 等执行适配器；同时改进长目录初始化、指标解释、同步 MCP 工具隔离和数据库超时，适合研究语义与执行边界。",
      "date_evidence": "GitHub Release v0.4.4 published_at 2026-10-28T14:22:18Z。发布说明列出 OntologyAnalyst、Athena/HANA 适配器、长目录初始化与 bounded executor 等实质变更。",
      "evidence_url": "https://github.com/Datus-ai/Datus-agent/releases/tag/v0.4.4",
      "license": "Apache-2.0",
      "license_url": "https://github.com/Datus-ai/Datus-agent/blob/main/LICENSE",
      "mechanism": "按业务域组织 schema、参考 SQL、指标和业务知识，检索后调用数据库/语义工具；修正反馈回流上下文；提供 CLI、Web、API 和 MCP。",
      "use_case": "适合研究数据工程 Agent 的上下文资产化、领域子 Agent 和指标优先查询。",
      "caveat": "数据库能力依赖所安装适配器与 Dosi build；Agent 的 300 秒等待上限不证明远端查询已经终止，不能据此自动安全重试写入。v0.4.3 移除的旧语义工作流仍需与旧教程区分。",
      "technical_material_url": "https://docs.datus.ai/dev/subagent/ontology_analyst/",
      "license_note": "2026-11-02 已核对 LICENSE 正文为 Apache License Version 2.0；GitHub API 仍识别为 NOASSERTION；archived=false。",
      "component_type": "agent",
      "last_verified": "2026-11-02",
      "release_updates": [
        {
          "version": "v0.4.4",
          "published_at": "2026-10-28T14:22:18Z",
          "url": "https://github.com/Datus-ai/Datus-agent/releases/tag/v0.4.4"
        }
      ]
    },
    {
      "id": "dbt-mcp-server",
      "date": "2026-11-03",
      "date_basis": "released",
      "title": "dbt MCP Server",
      "organization": "dbt Labs",
      "url": "https://github.com/dbt-labs/dbt-mcp",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "mcp",
        "semantic-layer",
        "lineage",
        "data-engineering",
        "governance"
      ],
      "summary": "v2.10.1 将 job/run 访问描述改为 Owner 目标与共享列举策略，授权宿主需处理 JOBS_LISTING、RUNS_LISTING；移除未使用的全文文档助手，但保留 llms.txt 搜索与单页读取。此前 Mesh 发现、超时处理和环境授权仍按对应配置使用。",
      "date_evidence": "GitHub v2.10.1 published_at 为 2026-11-03T16:45:17Z，release 正文标题却写 2026-11-02；索引按可核验的公开发布时间记为11月3日，并保留差异。",
      "evidence_url": "https://github.com/dbt-labs/dbt-mcp/releases/tag/v2.10.1",
      "license": "Apache-2.0",
      "license_url": "https://github.com/dbt-labs/dbt-mcp/blob/main/LICENSE",
      "mechanism": "通过 MCP 提供本地 dbt CLI、manifest 查询、Semantic Layer、Discovery、Admin API 与 SQL 工具，让通用 Agent 复用 dbt 上下文。",
      "use_case": "适合已采用 dbt 的团队构建分析工程助手、指标问答与作业诊断工具。",
      "caveat": "它是工具服务；本地 CLI 与平台 API 的依赖不同。Semantic Layer、Discovery 和平台 SQL 等能力需要对应 dbt Platform 配置与权限。",
      "technical_material_url": "https://docs.getdbt.com/docs/dbt-ai/about-mcp",
      "component_type": "tool-server",
      "notes": "v2.10.1 调整授权宿主的访问描述与 listing policy 契约，不代表扩大为无条件账户级权限，也没有移除全部产品文档工具。Mesh discovery 仍需 multi-project 配置；2026-11-05 核对 Apache-2.0 LICENSE 与 archived=false。",
      "last_verified": "2026-11-05",
      "release_updates": [
        {
          "version": "v2.9.0",
          "published_at": "2026-10-27T15:06:49Z",
          "url": "https://github.com/dbt-labs/dbt-mcp/releases/tag/v2.9.0"
        },
        {
          "version": "v2.10.0",
          "published_at": "2026-10-30T20:00:14Z",
          "url": "https://github.com/dbt-labs/dbt-mcp/releases/tag/v2.10.0"
        },
        {
          "version": "v2.10.1",
          "published_at": "2026-11-03T16:45:17Z",
          "url": "https://github.com/dbt-labs/dbt-mcp/releases/tag/v2.10.1"
        }
      ]
    },
    {
      "id": "dbt-agent-skills",
      "date": "2026-02-05",
      "date_basis": "released",
      "title": "dbt Agent Skills",
      "organization": "dbt Labs",
      "url": "https://github.com/dbt-labs/dbt-agent-skills",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "skills",
        "data-engineering",
        "semantic-layer",
        "validation",
        "evaluation"
      ],
      "summary": "可安装的 dbt 流程知识：先检查真实数据、迭代建模、再测试验证；包含指标建模与技能 A/B 评估材料。",
      "date_evidence": "dbt Developer Blog 2026-02-05 的官方文章写明当天发布 dbt agent skills，并链接该项目与其评估材料。",
      "evidence_url": "https://docs.getdbt.com/blog/dbt-agent-skills",
      "license": "Apache-2.0",
      "license_url": "https://github.com/dbt-labs/dbt-agent-skills/blob/main/LICENSE",
      "mechanism": "将探索数据、编写模型、验证结果、编写测试等流程封装为按需加载的 Agent Skills，并提供补充参考与评估工具。",
      "use_case": "适合用已有编码 Agent 承接分析工程工作，或设计企业自己的数据技能包和验证流程。",
      "caveat": "它提供流程与知识，依赖宿主 Agent 和实际 dbt 环境执行；技能是否被自动选中也需纳入评估。",
      "technical_material_url": "https://github.com/dbt-labs/dbt-agent-skills/tree/main/evals",
      "component_type": "skills"
    },
    {
      "id": "dbt-ade-bench-repo",
      "date": "2026-09-30",
      "date_basis": "updated",
      "title": "ADE-bench",
      "organization": "dbt Labs / Benn Stancil",
      "url": "https://github.com/dbt-labs/ade-bench",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "benchmark",
        "sandbox",
        "data-engineering",
        "dbt",
        "evaluation"
      ],
      "summary": "用真实 dbt 项目、数据库、隔离容器与结果测试评测数据工作；可比较模型、MCP、技能包带来的效果与成本。",
      "date_evidence": "2026-09-30 合入 fe3c50263447232064cdd45908efc6270970118a：修正 dbt Fusion 2.x 测试输出解析，恢复评测兼容性。",
      "evidence_url": "https://github.com/dbt-labs/ade-bench/commit/fe3c50263447232064cdd45908efc6270970118a",
      "license": "Apache-2.0",
      "license_url": "https://github.com/dbt-labs/ade-bench/blob/main/LICENSE",
      "mechanism": "每个任务复制项目并创建临时沙箱，注入任务问题，交给 Agent 修改，随后以 dbt 测试、结果表对比和文件快照判断是否完成。",
      "use_case": "适合评估修改数据模型、排查管道错误等工程任务，也可用于回归比较不同工具与技能组合。",
      "caveat": "默认任务主要围绕 dbt 分析工程；评测结果不能直接外推到开放式业务洞察或可视化报告质量。",
      "technical_material_url": "https://github.com/dbt-labs/ade-bench/blob/main/docs/CLI.md",
      "component_type": "evaluation-harness"
    },
    {
      "id": "google-mcp-toolbox-databases",
      "date": "2026-11-05",
      "date_basis": "released",
      "title": "MCP Toolbox for Databases",
      "organization": "Google",
      "url": "https://github.com/googleapis/mcp-toolbox",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "mcp",
        "database",
        "execution",
        "authentication",
        "observability"
      ],
      "summary": "v1.16.0 增加通过 MCP 提供 Skills 和动态 UI 资源，并修复 BigQuery protected 写入模式：非 SELECT 若无法证明目标属于当前会话临时数据集则拒绝执行；仍需区分开发期通用 SQL 工具与生产定制工具。",
      "date_evidence": "v1.16.0 published_at 为 2026-11-05T19:56:15Z，即 Asia/Shanghai 2026-11-06 03:56:15；按本仓 release 日期规则保留 UTC 日期 2026-11-05，changelog 也标注 2026-11-05。",
      "evidence_url": "https://github.com/googleapis/mcp-toolbox/releases/tag/v1.16.0",
      "license": "Apache-2.0",
      "license_url": "https://github.com/googleapis/mcp-toolbox/blob/06ea49d9a487a72c22419b5a8d5c22c11f0d7485/LICENSE",
      "mechanism": "将数据源连接和工具声明集中在配置中，向 MCP 客户端或 Agent SDK 暴露预置查询及定制工具，并提供认证、连接池和可观测性。",
      "use_case": "适合搭建统一数据库工具服务，或将受控查询封装成可复用工具供多个数据 Agent 调用。",
      "caveat": "bigquery-execute-sql 官方定位为有人监督的开发助手工具，不建议直接用于生产 Agent；默认 readOnly=false、writeMode=allowed。protected 模式仍允许经验证的会话临时写入，不等于严格只读；生产应配置最小权限与受控工具。",
      "technical_material_url": "https://github.com/googleapis/mcp-toolbox/blob/06ea49d9a487a72c22419b5a8d5c22c11f0d7485/docs/en/documentation/introduction/_index.md",
      "notes": "本次核验 v1.16.0、Apache-2.0 与 archived=false。创建/更新/删除数据 Agent 是管理能力，不能等同查询权限；工具服务仍需上层系统提供业务语义、规划与结果验证。",
      "component_type": "tool-server",
      "last_verified": "2026-11-06",
      "release_updates": [
        {
          "version": "v1.15.0",
          "published_at": "2026-10-23T00:14:18Z",
          "url": "https://github.com/googleapis/mcp-toolbox/releases/tag/v1.15.0"
        },
        {
          "version": "v1.16.0",
          "published_at": "2026-11-05T19:56:15Z",
          "url": "https://github.com/googleapis/mcp-toolbox/releases/tag/v1.16.0"
        }
      ],
      "verification_scope": "核对官方 release、仓库归档状态、所列根许可证及固定提交的技术文档；未部署运行、复现基准或审计依赖许可证。历史支持链接并非本轮全部重验。",
      "reading_path": [
        {
          "title": "工具服务架构",
          "url": "https://github.com/googleapis/mcp-toolbox/blob/06ea49d9a487a72c22419b5a8d5c22c11f0d7485/docs/en/documentation/introduction/_index.md",
          "focus": "区分连接、工具配置、认证与上层 Agent。"
        },
        {
          "title": "BigQuery 通用 SQL 工具边界",
          "url": "https://github.com/googleapis/mcp-toolbox/blob/06ea49d9a487a72c22419b5a8d5c22c11f0d7485/docs/en/integrations/bigquery/tools/bigquery-execute-sql.md",
          "focus": "核对默认写入权限、开发期用途和 protected 模式。"
        },
        {
          "title": "protected 写入修复及测试",
          "url": "https://github.com/googleapis/mcp-toolbox/commit/426fb4e8074b873dc53b74bd259540d01684fa17",
          "focus": "核对临时表允许范围、永久目标拒绝与未知语句类型的处理。"
        }
      ]
    },
    {
      "id": "boring-semantic-layer",
      "date": "2026-08-24",
      "date_basis": "released",
      "title": "Boring Semantic Layer (BSL)",
      "organization": "boringdata / xorq-labs",
      "url": "https://github.com/boringdata/boring-semantic-layer",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "semantic-layer",
        "metrics",
        "ibis",
        "mcp",
        "verification"
      ],
      "summary": "基于 Ibis 的轻量语义层，将维度与指标封装为可查询模型；可作为 Agent 的指标工具底座，减少临时重写口径。",
      "date_evidence": "v0.3.18 published_at 2026-08-24T13:29:34Z；版本改进 authored leaf 恢复、join 叶表达式处理，并拒绝定义侧语义表达式的 raw 输出。",
      "evidence_url": "https://github.com/boringdata/boring-semantic-layer/releases/tag/v0.3.18",
      "license": "MIT",
      "license_url": "https://github.com/boringdata/boring-semantic-layer/blob/main/LICENSE",
      "mechanism": "用 Python/Ibis 定义维度、度量和语义表，通过受控指标查询编译到不同数据库后端。",
      "use_case": "适合研究指标口径约束、语义工具接口与轻量级分析 Agent 的数据底座。",
      "caveat": "它是语义组件；需要另接 Agent、认证和 UI。已有 dbt MetricFlow 模型的团队还需评估语义模型转换成本。",
      "technical_material_url": "https://boringdata.github.io/boring-semantic-layer/",
      "component_type": "semantic-layer"
    },
    {
      "id": "microsoft-rd-agent",
      "date": "2026-09-23",
      "date_basis": "released",
      "title": "R&D-Agent",
      "organization": "Microsoft Research",
      "url": "https://github.com/microsoft/RD-Agent",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "data-science",
        "multi-agent",
        "experimentation",
        "feedback",
        "evaluation"
      ],
      "summary": "以研究 Agent 提出假设、开发 Agent 编码实验、执行反馈再迭代，覆盖数据科学与建模任务，适合观察实验闭环。",
      "date_evidence": "v1.0.0 官方发布说明标注 2026-09-23；新增 LLM finetune 场景、AutoRL-Bench 集成、Web UI 与后端轨迹同步。",
      "evidence_url": "https://github.com/microsoft/RD-Agent/releases/tag/v1.0.0",
      "license": "MIT",
      "license_url": "https://github.com/microsoft/RD-Agent/blob/main/LICENSE",
      "mechanism": "将提出假设的 Research 和编码实现的 Development 分开，依据实际实验和评估反馈循环改进数据特征、模型或训练方案。",
      "use_case": "适合研究数据科学 Agent 的自动实验、代码修复和反馈循环，或复现 ML 工程任务。",
      "caveat": "项目当前明确支持 Linux，多数场景需要 Docker；不同建模任务还需相应数据与计算资源，非开箱即用业务 BI 助手。",
      "technical_material_url": "https://rdagent.readthedocs.io/en/latest/scens/data_science.html",
      "component_type": "agent"
    },
    {
      "id": "microsoft-data-formulator",
      "date": "2026-05-28",
      "date_basis": "released",
      "title": "Data Formulator",
      "organization": "Microsoft Research",
      "url": "https://github.com/microsoft/data-formulator",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "visualization",
        "analytics",
        "memory",
        "human-in-the-loop",
        "report"
      ],
      "summary": "把数据连接、加载 Agent、带记忆的 Data Thread、图表细化和报告分享串在可分支的可视工作区，适合交互式分析。",
      "date_evidence": "稳定版 0.7.0 published_at 2026-05-28T08:55:51Z；引入统一分析 Agent、数据加载 Agent、线程记忆与持久工作区。",
      "evidence_url": "https://github.com/microsoft/data-formulator/releases/tag/0.7.0",
      "license": "MIT",
      "license_url": "https://github.com/microsoft/data-formulator/blob/5477f0e236426dc8f74a498ec400414fba7fbc0f/LICENSE",
      "mechanism": "通过数据连接器和数据记忆理解多源数据，在 Data Thread 中保留分析路径，支持从历史节点分支、图表编辑和自然语言继续探索。",
      "use_case": "适合交互式 EDA、图表与报告体验原型，也适合研究用户如何审阅和纠正 Agent 的分析过程。",
      "caveat": "稳定复现仍可固定 0.7.0；GitHub 最新 tag 为 0.8b1，README 对应包版本 0.8.0b1，官方称为 beta，虽然 GitHub prerelease 字段为 false。当前开发文档默认面向单用户本地运行；匿名多用户部署应关闭数据连接器、自定义模型和密钥显示，正式团队部署需认证身份，不能把浏览器 UUID 当作安全身份。",
      "technical_material_url": "https://github.com/microsoft/data-formulator/blob/5477f0e236426dc8f74a498ec400414fba7fbc0f/DEVELOPMENT.md",
      "component_type": "workspace",
      "last_verified": "2026-11-06",
      "verification_scope": "核对官方 release、仓库归档状态、所列根许可证及固定提交的技术文档；未部署运行、复现基准或审计依赖许可证。历史支持链接并非本轮全部重验。",
      "notes": "稳定版记录保留 0.7.0 的发布日期；预览版和当前开发文档单列，不能视为稳定版已具备全部后续能力。",
      "release_updates": [
        {
          "version": "0.7.0",
          "published_at": "2026-05-28T08:55:51Z",
          "url": "https://github.com/microsoft/data-formulator/releases/tag/0.7.0"
        },
        {
          "version": "0.8b1",
          "published_at": "2026-08-15T20:54:12Z",
          "url": "https://github.com/microsoft/data-formulator/releases/tag/0.8b1",
          "prerelease": true
        }
      ],
      "related_materials": [
        {
          "title": "Beta 名称与包版本说明",
          "url": "https://github.com/microsoft/data-formulator/blob/5477f0e236426dc8f74a498ec400414fba7fbc0f/README.md",
          "note": "本仓根据发布正文和 README 标记预览版，不仅依赖 GitHub prerelease 字段。"
        }
      ]
    },
    {
      "id": "jupyter-ai",
      "date": "2026-09-03",
      "date_basis": "released",
      "title": "Jupyter AI",
      "organization": "Project Jupyter",
      "url": "https://github.com/jupyterlab/jupyter-ai",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "notebook",
        "mcp",
        "acp",
        "execution",
        "human-in-the-loop"
      ],
      "summary": "在 JupyterLab 内让 Agent 编辑、运行和调试 Notebook；以 ACP/MCP 接入工具，适合保留单元格与执行结果作为分析证据。",
      "date_evidence": "官方 v3.2.0 发布文档标注 Published on September 3, 2026；GitHub published_at 为 2026-09-03T17:43:30Z（Asia/Shanghai 为 2026-09-04）。本条采用发布方标注且与 UTC 一致的日期，不将核验日期或时区换算误作新发布。",
      "evidence_url": "https://github.com/jupyterlab/jupyter-ai/blob/995b53b651c39bf9a7744bfe0d2d85b0967b672c/docs/source/releases/v3.2.0.md",
      "license": "BSD-3-Clause",
      "license_url": "https://github.com/jupyterlab/jupyter-ai/blob/dc4d493031484480040a89e68ecea8d49f1665e2/LICENSE",
      "mechanism": "在 JupyterLab 原生聊天中连接 Agent，通过内置 MCP 和前端命令操作 Notebook 与文件，并在写入和执行前进行权限审批。",
      "use_case": "适合需要人工审阅的 Notebook 分析、数据科学协作，以及将步骤和输出留在可继续运行的工作区。",
      "caveat": "v3.2.0 的 Notebook 编辑、运行和调试可不依赖 RTC，但官方明确说明：外部直接修改磁盘上的 Notebook 不会因此实时刷新到界面，文件刷新还可能丢失滚动位置。聊天工具实时操作与任意磁盘修改自动同步是不同能力；主仓库发布还需结合子包 release notes 阅读。",
      "technical_material_url": "https://github.com/jupyterlab/jupyter-ai/blob/995b53b651c39bf9a7744bfe0d2d85b0967b672c/docs/source/releases/v3.2.0.md",
      "component_type": "workspace",
      "last_verified": "2026-11-06",
      "verification_scope": "核对官方 release、仓库归档状态、所列根许可证及固定提交的技术文档；未部署运行、复现基准或审计依赖许可证。历史支持链接并非本轮全部重验。",
      "release_updates": [
        {
          "version": "v3.2.0",
          "published_at": "2026-09-03T17:43:30Z",
          "url": "https://github.com/jupyterlab/jupyter-ai/releases/tag/v3.2.0"
        }
      ]
    },
    {
      "id": "dlt-mcp-server",
      "date": "2026-02-19",
      "date_basis": "updated",
      "title": "dlt MCP Server",
      "organization": "dltHub",
      "url": "https://github.com/dlt-hub/dlt-mcp",
      "section": "open-source",
      "language": "en",
      "status": "active-project",
      "tags": [
        "mcp",
        "data-engineering",
        "pipeline",
        "schema",
        "execution"
      ],
      "summary": "为 Agent 提供 dlt 管道、表结构、增量状态、schema diff 与 SELECT 查询工具，适合数据接入和管道诊断。",
      "date_evidence": "2026-02-19 commit b128708215abe34a91e028a2f8916516392ffb8a 将版本更新为 0.3.0，同日新增简单入口与 extras 并更新安装说明。",
      "evidence_url": "https://github.com/dlt-hub/dlt-mcp/commit/b128708215abe34a91e028a2f8916516392ffb8a",
      "license": "Apache-2.0",
      "license_url": "https://github.com/dlt-hub/dlt-mcp/blob/main/LICENSE",
      "mechanism": "把本地 dlt 管道元数据、加载状态、schema 差异、只读查询与 dlt 文档/代码检索暴露为 MCP 工具。",
      "use_case": "适合给数据工程 Agent 补充真实管道状态，使其能够诊断增量加载、表结构变动和数据接入问题。",
      "caveat": "需要已有 dlt 管道并配置目标数据库 extras；它提供数据接入工具上下文，执行和规划仍由宿主 Agent 完成。",
      "technical_material_url": "https://dlthub.com/docs/dlt-ecosystem/llm-tooling/llm-native-workflow",
      "component_type": "tool-server"
    },
    {
      "id": "vanna-agent-framework",
      "date": "2026-02-02",
      "date_basis": "released",
      "title": "Vanna 2.0",
      "organization": "Vanna AI",
      "url": "https://github.com/vanna-ai/vanna",
      "section": "open-source",
      "language": "en",
      "status": "archived",
      "tags": [
        "sql",
        "agent-loop",
        "identity",
        "governance",
        "streaming"
      ],
      "summary": "由应用提供身份和组信息，贯穿 Agent、SQL 工具与流式表格图表；兼容适配器不保证行级安全。2026-03-29 已归档，保留作架构研究案例。",
      "date_evidence": "v2.0.2 published_at 2026-02-02T14:14:20Z，修正 ChromaDB 检索时多余 embedding 初始化；GitHub 项目页显示 2026-03-29 归档。",
      "evidence_url": "https://github.com/vanna-ai/vanna/releases/tag/v2.0.2",
      "license": "MIT",
      "license_url": "https://github.com/vanna-ai/vanna/blob/365d0617c1a4567ffee1b19b40c27feb4206bfcf/LICENSE",
      "archived": true,
      "archived_date": "2026-03-29",
      "mechanism": "由应用实现 UserResolver，将经认证的身份和组信息交给 Agent 与工具注册表；工具组权限、SQL 执行与流式 UI 在同一调用链中协作。",
      "use_case": "适合研究如何把身份、SQL 执行、审计和可嵌入分析 UI 串成同一数据 Agent 调用链。",
      "caveat": "项目已于 2026-03-29 归档。LegacyVannaAdapter 共享同一 VannaBase 实例，官方迁移指南明确说明不能利用行级安全；身份传递和工具组权限不能直接当作数据库行级隔离保证。",
      "technical_material_url": "https://github.com/vanna-ai/vanna/blob/365d0617c1a4567ffee1b19b40c27feb4206bfcf/MIGRATION_GUIDE.md",
      "component_type": "agent-framework",
      "last_verified": "2026-11-06",
      "verification_scope": "核对官方 release、仓库归档状态、所列根许可证及固定提交的技术文档；未部署运行、复现基准或审计依赖许可证。历史支持链接并非本轮全部重验。",
      "release_updates": [
        {
          "version": "v2.0.2",
          "published_at": "2026-02-02T14:14:20Z",
          "url": "https://github.com/vanna-ai/vanna/releases/tag/v2.0.2"
        }
      ]
    },
    {
      "id": "adapra-paper",
      "date": "2026-10-23",
      "date_basis": "published",
      "title": "AdaPRA: Process Reward Agents with Adaptive Rubrics for Data Science",
      "organization": "Carnegie Mellon University / Microsoft",
      "url": "https://arxiv.org/abs/2610.33245",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "data-science",
        "process-supervision",
        "rubrics",
        "memory",
        "verification"
      ],
      "summary": "将跨尝试保留的工作区与自适应检查准则结合：独立过程监督 Agent 读取轨迹和产物、执行检查，只反馈相关失败项，并把任务特有错误转成后续准则；适合研究分析过程中的静默错误与验证泄漏。",
      "notes": "公开预印本。论文的 24.5/14.8 点是每任务五次尝试中 best-attempt 的平均 Kaggle 百分位增益，不是成功率；监督器也消耗额外成本。作者项目页仍写 Code (coming soon)，但已直接核验 GitHub 完整实现与 MIT LICENSE；运行输出标为正在上传，不能据此声称所有轨迹已公开。",
      "date_evidence": "arXiv submission history: [v1] Fri, 23 Oct 2026 19:46:55 UTC。",
      "evidence_url": "https://arxiv.org/abs/2610.33245",
      "technical_material_url": "https://github.com/atharva-naik/AdaPRA",
      "supporting_urls": [
        "https://atharva-naik.github.io/AdaPRA/",
        "https://github.com/atharva-naik/AdaPRA/blob/main/LICENSE"
      ],
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "atlan-analytics-agent-fde-playbook",
      "date": "2026-10-23",
      "date_basis": "updated",
      "title": "The FDE Playbook for Analytics Agents",
      "organization": "Atlan AI Labs",
      "url": "https://analytics-playbook.atlan.com/",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "semantic-layer",
        "metrics",
        "evaluation",
        "context",
        "production"
      ],
      "summary": "从封存真实问题黄金集开始，依次生成指标、按题检索上下文、从轨迹修正并设发布门禁；区分受治理指标回答与探索 SQL，附模板和故障排查。示例项目为复合场景，评测来自作者内部仓库。",
      "date_evidence": "作者 CHANGELOG.md 标注 V4.0, 2026-10-23，重组问题、方法与参考章节并新增原则页；同一日志列最早 V1.0 为 2026-09-27，因此按实质更新收录，而非用 10-27 仓库创建时间声称首次发布。",
      "evidence_url": "https://github.com/atlanai/analytics-playbook/blob/main/CHANGELOG.md",
      "caveat": "这是实施方法手册，不是完整 Agent 软件。作者内部 107 问题的分数不能外推到其他企业或当作独立排行榜；内容 CC BY 4.0，构建代码 MIT。",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "keyhole-agentic-harness-data-engineering",
      "date": "2026-10-23",
      "date_basis": "published",
      "title": "White Paper – Agentic Harness Architecture in 2026: Components, Vendors, and a Reference Design for Data Engineering",
      "organization": "Keyhole Software / Zach Gardner",
      "url": "https://keyholesoftware.com/agentic-harness/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "data-engineering",
        "harness",
        "governance",
        "verification",
        "human-in-the-loop"
      ],
      "summary": "将通用 Harness 的模型网关、最小权限工具、检查点与确定性检查器扩展到数据工程，以只读发现、计划审批、测试、PR/CI 晋级和目录登记串起交付。文中端到端流程是复合参考场景，不是单个客户实测。",
      "date_evidence": "作者官网正文标注 October 23, 2026；参考架构与 One Request, Start to Finish 部分说明数据工程扩展及其验收门禁，并明确示例为 representative composite。",
      "evidence_url": "https://keyholesoftware.com/agentic-harness/",
      "caveat": "用于参考架构与治理检查；文中跨厂商数字来自不同研究，不能合并为统一 Data Agent 排行榜。",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "kx-qsql-mcp-verification",
      "date": "2026-10-23",
      "date_basis": "published",
      "title": "Natural Language to qSQL with the MCP qSQL Suggester",
      "organization": "KX",
      "url": "https://kx.com/blog/natural-language-to-qsql-mcp-suggester/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "qsql",
        "mcp",
        "skills",
        "verification",
        "self-hosted"
      ],
      "summary": "把专用 qSQL 模型、MCP 和 Skills 组合为草稿—候选一致性—解析/qlint/执行检查链，并按需与参考查询比较结果。候选一致不代表语义正确；完整自托管仍需 q/kdb+ 许可。",
      "date_evidence": "官方文章 JSON-LD datePublished 为 2026-10-23T12:38:32+01:00（dateModified 为 2026-10-28）；KX 官方资源列表同样列出 23 October, 2026。",
      "evidence_url": "https://kx.com/blog/natural-language-to-qsql-mcp-suggester/",
      "technical_material_url": "https://github.com/KxSystems/mcp-qsql-suggester",
      "caveat": "这是专用查询生成与验证组件，不是通用企业分析 Agent；发布方的机器生成任务集与训练分布相近，不作为跨模型榜单。",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "calimem-paper",
      "date": "2026-10-25",
      "date_basis": "published",
      "title": "CaliMem: Search-Efficient Data Agent via Calibrated Memory",
      "organization": "Virginia Tech / University of Southern California / Princeton University",
      "url": "https://arxiv.org/abs/2610.34500",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "data-curation",
        "memory",
        "calibration",
        "experimentation",
        "feedback"
      ],
      "summary": "在数据整理策略搜索中先预测实验结果，再用预测与实测的偏差修正持久记忆中的结论及适用条件，避免一次失败便否定整类策略；其借鉴点是可检验、带条件的记忆，而不只是累积历史文本。",
      "notes": "公开预印本；评估覆盖生物数据整理和 LLM 后训练，不是企业问数通用基准。作者称代码已公开，正文指向 anonymous.4open.science 的匿名工件，本次未独立验证其可访问性与许可证，因此不新增为已核验开源项目。性能提升来自论文实验，尚非独立复现。",
      "date_evidence": "arXiv submission history: [v1] Sun, 25 Oct 2026 18:55:44 UTC。",
      "evidence_url": "https://arxiv.org/abs/2610.34500",
      "technical_material_url": "https://arxiv.org/html/2610.34500v1",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "clickhouse-jev-retrieval-failure-labeling",
      "date": "2026-10-27",
      "date_basis": "published",
      "title": "The Agentic Analytics Benchmark: ranking tables and labeling failures with Jev",
      "organization": "ClickHouse",
      "url": "https://clickhouse.com/blog/agentic-analytics-benchmark-jev",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "retrieval",
        "failure-analysis",
        "trace",
        "evaluation"
      ],
      "summary": "用类型化分类器做表排序和细粒度失败归因，可从 SQL 确定的事实先由代码判定。作者公开了检索提示的成对实验：组件排序表现较好，但跨 20 个模型未带来显著端到端准确率收益；不要把检索指标等同于 Agent 提升。",
      "date_evidence": "ClickHouse 原文作者栏下标注 Oct 27, 2026；区分 136 题开发集与 201 题保留评测集，并公开配套 harness 入口。",
      "evidence_url": "https://clickhouse.com/blog/agentic-analytics-benchmark-jev",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "data-marc-dynamic-data-agent-evaluation",
      "date": "2026-10-27",
      "date_basis": "published",
      "title": "Can You Trust Your Fabric Data Agent? Programmatically Testing Accuracy and Determinism",
      "organization": "Data – Marc",
      "url": "https://data-marc.com/2026/10/27/can-you-trust-your-fabric-data-agent-programmatically-testing-accuracy-and-determinism/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "fabric",
        "evaluation",
        "ground-truth",
        "regression",
        "determinism"
      ],
      "summary": "将参考查询而非固定答案存入测试集，每次先从当前数据计算期望值，再用 Fabric SDK 留存评测与步骤结果。公开 Notebook 代码示例；查询使用测试执行者身份，需对齐权限与数据快照，不能把动态答案当成无需审核的真值。",
      "date_evidence": "作者原文标题下标注 October 27, 2026；正文给出动态参考结果生成和 SDK 评测代码。",
      "evidence_url": "https://data-marc.com/2026/10/27/can-you-trust-your-fabric-data-agent-programmatically-testing-accuracy-and-determinism/",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "hex-4-agent-native-execution-architecture",
      "date": "2026-10-27",
      "date_basis": "published",
      "title": "Introducing Hex 4.0",
      "organization": "Hex",
      "url": "https://hex.tech/blog/introducing-hex-4-0/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "dag",
        "lineage",
        "human-in-the-loop",
        "execution",
        "mcp"
      ],
      "summary": "以代码文件和显式依赖 DAG 支持 Agent 构建、人类检查数字血缘；执行器并行运行独立步骤、流式返回结果，只重跑被变更影响的依赖。发布时仍为 private beta，性能图来自内部应用，不是通用对照评测。",
      "date_evidence": "Hex 产品原文标题下标注 October 27, 2026；正文分别解释代码框架、可检查依赖和新执行器，末节注明 private beta。",
      "evidence_url": "https://hex.tech/blog/introducing-hex-4-0/",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "m4bench-paper",
      "date": "2026-10-27",
      "date_basis": "published",
      "title": "M4Bench: Evaluating Procedural Specification for Clinical EHR Derivation Agents",
      "organization": "Hannes Ill et al.",
      "url": "https://arxiv.org/abs/2610.37459",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "benchmark",
        "data-engineering",
        "skills",
        "semantic-validation",
        "sandbox"
      ],
      "summary": "用 MIMIC-IV/eICU 的二十八个数据派生任务检验 Agent 能否遵守编码、时间窗、缺失值规则与输出粒度；以操作规格、Skills、原始参考 SQL 和模式变体作对照，区分知识完整性与交付形式的影响。",
      "notes": "公开预印本。reward 提升是与参考结果的部分一致性，不等于全任务完成：with-skill 仅 25/280 次 Codex 运行通过全部严格检查，no-skill 为 13/280。基准代码为 MIT；MIMIC-IV/eICU 原始数据、任务数据库与参考输出不随代码分发，完整复现仍需 PhysioNet 资质及相应数据使用协议。",
      "date_evidence": "arXiv submission history: [v1] Tue, 27 Oct 2026 15:59:13 UTC；代码与复现说明链接由论文 Comments 明确提供。",
      "evidence_url": "https://arxiv.org/abs/2610.37459",
      "technical_material_url": "https://github.com/hannesill/m4/tree/m4bench-neurips2026-camera-ready/benchmark",
      "supporting_urls": [
        "https://github.com/hannesill/m4/blob/m4bench-neurips2026-camera-ready/benchmark/REPRODUCIBILITY.md",
        "https://github.com/hannesill/m4/blob/main/LICENSE"
      ],
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "yunqi-dacon-knowledge-ontology-recap",
      "date": "2026-10-27",
      "date_basis": "published",
      "title": "从数据底座到 AI 知识底座，云器在 DACon 带来了什么？｜附演讲材料",
      "organization": "云器科技 / DACon 北京",
      "url": "https://www.yunqi.tech/resource/blogs/dacon-ai-lakehouse",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "recap",
        "ontology",
        "knowledge-engineering",
        "semantic-layer",
        "observability"
      ],
      "summary": "官方会后复盘把抽取、语义对齐、评估认证与运行反馈串成知识生产链，并说明 AI ETL、Semantic View、SingSight 的分工。公开的是复盘正文；完整 PPT 需扫码小程序获取，本次未验证可直接下载的幻灯片或录像。",
      "date_evidence": "官方原文标注 2026年10月27日，正文回顾 10月23—24日 DACon 北京活动；该日期为复盘文章发布日期，不是 PPT 公布时间。",
      "evidence_url": "https://www.yunqi.tech/resource/blogs/dacon-ai-lakehouse",
      "material_access": "public-recap; deck-request-via-qr-unverified",
      "event_date": "2026-10-23/2026-10-24",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "atlan-post-ai-data-stack-context-conference",
      "date": "2026-10-28",
      "date_basis": "event",
      "title": "The Post-AI Data Stack: A blueprint for analytics agents",
      "organization": "Atlan / Context Conference / Workday / DocuSign / Upwork",
      "url": "https://atlan.com/context-conference/post-ai-data-stack/",
      "section": "talk",
      "language": "en",
      "status": "published",
      "tags": [
        "semantic-layer",
        "ontology",
        "skills",
        "evaluation",
        "human-in-the-loop"
      ],
      "summary": "36:55 公开视频与逐字稿，拆解指标、Ontology、Skills 三类上下文，以及引导生成、版本治理和评测迭代；客户讨论认证指标优先、标注推断与人工回流。内部准确率陈述仅作发布方经验。",
      "date_evidence": "大会原始页面标注 Builder session + customer panel Oct 28, 2026 36:55，并提供 Play 播放入口、12 个章节和可定位时间的完整 Transcript。",
      "evidence_url": "https://atlan.com/context-conference/post-ai-data-stack/",
      "material_type": [
        "video",
        "transcript"
      ],
      "caveat": "视频与逐字稿已公开，相关实施手册另作持续文档条目；分享中的内部评测不等于独立可复现 benchmark。",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "datus-dosi-executable-business-context",
      "date": "2026-10-28",
      "date_basis": "published",
      "title": "Introducing DOSI: Executable Business Context for AI Agents",
      "organization": "Datus",
      "url": "https://datus.ai/blog/datus-dosi-executable-business-context/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "semantic-layer",
        "ontology",
        "mcp",
        "lineage",
        "verification"
      ],
      "summary": "将经审核的 Ossie 语义模型编译为 SQL，通过 MCP 等接口执行指标；用稳定错误码、关联歧义与 fan-out 检查降低盲目重试，并让结果关联定义、查询和血缘。业务定义、源数据及模型兼容性仍需人工验证。",
      "date_evidence": "Datus 原文首行同时标注发布与更新日期 Oct 28, 2026；功能段解释 MCP 错误契约、join 检查及可检查 SQL/血缘。",
      "evidence_url": "https://datus.ai/blog/datus-dosi-executable-business-context/",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "cube-data-model-factory",
      "date": "2026-10-29",
      "date_basis": "published",
      "title": "Building the Data Model Factory",
      "organization": "Cube",
      "url": "https://cube.dev/blog/data-modeling-is-becoming-system-design",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "semantic-layer",
        "context",
        "feedback",
        "human-in-the-loop",
        "data-engineering"
      ],
      "summary": "把语义建模当成可维护系统：监测模式、查询日志和对话，生成模型变更 PR，由工程师定义验收并独立审查；Semantic SQL 组合既有受治理指标。属于作者方案与实践观察，不代表无需人工决定业务口径，也不是公开对照评测。",
      "date_evidence": "Cube 原文标题下标注 October 29, 2026；正文描述生产反馈、PR 审查、验收标准与 Semantic SQL 示例。",
      "evidence_url": "https://cube.dev/blog/data-modeling-is-becoming-system-design",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "motherduck-dive-apps-agent-actions",
      "date": "2026-10-29",
      "date_basis": "published",
      "title": "Introducing Dive Apps: Superpowers for AI Builders",
      "organization": "MotherDuck",
      "url": "https://motherduck.com/blog/dive-apps-superpowers-for-ai-builders/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "mcp",
        "execution",
        "human-in-the-loop",
        "data-apps",
        "write-back"
      ],
      "summary": "通过 MCP 让 Agent 构建数据应用，Dive 交互触发 serverless Python Flights，支持调用 Agent harness、写回源系统及人工介入。是产品方架构和示例说明，不是完整权限/审批安全证明；写操作需单独设计授权与回滚。",
      "date_evidence": "MotherDuck 原文标题下标注 Oct 29, 2026；首节解释 Dive 到 Flights 的触发关系，并链接写回、邮件及 harness 示例。",
      "evidence_url": "https://motherduck.com/blog/dive-apps-superpowers-for-ai-builders/",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "pena-fabric-data-agent-repeatability",
      "date": "2026-10-29",
      "date_basis": "published",
      "title": "Fabric Data Agent Accuracy Is a Distribution, Not a Number",
      "organization": "Michael John Peña",
      "url": "https://michaeljohnpena.com/blog/2026-10-29-fabric-data-agent-determinism-real-nem-data/",
      "section": "evaluation",
      "language": "en",
      "status": "published",
      "tags": [
        "fabric",
        "evaluation",
        "determinism",
        "trace",
        "regression"
      ],
      "summary": "对 24 个问题、三种表述和多次全新会话保留重写、路由、查询与答案，并比较确定性数值判定和 SDK judge，报告准确率及 pass^k。结果仅覆盖一个电力数据集、Standard runtime 与指定预览 SDK，不能外推为所有 Fabric Agent 的成绩。",
      "date_evidence": "作者原文标题前标注 October 29, 2026；正文列出数据、运行环境、重复次数和 harness 代码，并明确范围限制。",
      "evidence_url": "https://michaeljohnpena.com/blog/2026-10-29-fabric-data-agent-determinism-real-nem-data/",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "yunqi-semantic-view-trusted-context",
      "date": "2026-10-29",
      "date_basis": "published",
      "title": "让 AI Agent 真正理解业务：云器 Semantic View 如何提供统一语义与可信数据上下文",
      "organization": "云器科技",
      "url": "https://www.yunqi.tech/resource/blogs/semantic-view-ai-agent",
      "section": "architecture",
      "language": "zh",
      "status": "published",
      "tags": [
        "semantic-layer",
        "context",
        "metrics",
        "governance",
        "verification"
      ],
      "summary": "以可发现、可查询的 Semantic View 复用指标、粒度与关联定义，通过内省命令先确认口径再计算。明确区分对象 SELECT 授权与行列隔离，强调澄清/停止仍需 Agent 策略，语义层不自动修复源数据质量；文中零售案例是流程示例而非客户实测。",
      "date_evidence": "云器官方原文标题下标注 2026年10月29日，作者为产品负责人郭天琪；正文公开 semantic_view()、SHOW SEMANTIC METRICS、DESC EXTENDED 等机制与限制。",
      "evidence_url": "https://www.yunqi.tech/resource/blogs/semantic-view-ai-agent",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "hex-governed-mode",
      "date": "2026-10-30",
      "date_basis": "released",
      "title": "Governed mode",
      "organization": "Hex Technologies",
      "url": "https://learn.hex.tech/docs/agent-management/governed-mode",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "governance",
        "context",
        "semantic-layer",
        "permissions"
      ],
      "summary": "按资产类型和状态引导 Agent 使用仓库表、语义模型与既有项目，可采用仅语义模型或自定义组合。文档明确它不是硬安全边界，不能代替数据连接和数仓权限。",
      "date_evidence": "官方独立更新日志 Governed mode for the Hex Agent 标注 October 30, 2026，并链接到该配置文档；日期对应功能发布，不将页面抓取日期视为更新时间。",
      "evidence_url": "https://learn.hex.tech/changelog/2026-10-30",
      "caveat": "来源选择策略与访问授权必须分开；关闭某类 Agent 来源不会撤销用户原有的数据查询权限。",
      "added_on": "2026-11-02",
      "last_verified": "2026-11-02"
    },
    {
      "id": "snowflake-agent-gateway-security-governance",
      "date": "2026-11-03",
      "date_basis": "published",
      "title": "Build Trust in AI Agents with Governed Business Context and AI Security",
      "organization": "Snowflake",
      "url": "https://www.snowflake.com/en/blog/ai-agents-governance-ai-security-cost-control/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "identity",
        "mcp",
        "governance",
        "observability",
        "cost-control",
        "sandbox"
      ],
      "summary": "从控制面拆解 Agent 接入：Cortex AI Gateway 集中策略，MCP 工具携带最终用户身份并在执行前检查权限；再结合隔离执行、运行观测与费用治理。适合对照语义定义治理与实际动作授权应分别落实的位置。",
      "date_evidence": "Snowflake 原文标题前标注 Nov 3, 2026；Govern, observe and optimize every agent action 章节说明 Gateway、执行前策略和沙箱的发布阶段。",
      "evidence_url": "https://www.snowflake.com/en/blog/ai-agents-governance-ai-security-cost-control/",
      "caveat": "Gateway/Tools 为 public preview，Cortex Sandboxes 为 private preview；治理文章不构成所有攻击均可阻断的证明。预算目标、费用告警与异常检测不应直接解释成所有任务的强制停止机制。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "snowflake-business-ontology-stewardship",
      "date": "2026-11-03",
      "date_basis": "published",
      "title": "Meaning Before Metrics: How We Built a Governed Business Ontology for AI Agents",
      "organization": "Snowflake Engineering",
      "url": "https://www.snowflake.com/en/blog/engineering/business-ontology-context/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "ontology",
        "semantic-layer",
        "context",
        "human-in-the-loop",
        "governance"
      ],
      "summary": "以带业务域的概念、类型化关系和资产引用组织业务定义；修改先形成提案，经 steward 批准才替换 Agent 使用的版本。Ontology 解释含义，Semantic Views 执行计算，Cortex Sense 提供推断线索；三者互补，不是强制串行配置。",
      "date_evidence": "Snowflake Engineering 原文标题前标注 Nov 3, 2026；正文公开概念图、域权限、审批及 business_ontology 检索工具的设计。",
      "evidence_url": "https://www.snowflake.com/en/blog/engineering/business-ontology-context/",
      "caveat": "文章公开不表示产品已全面开放：文首标注 Business Ontology 为 private preview，文末说明 public preview soon。示例为概念说明，不可直接执行或导入；子域权限不自动继承，资产引用本身不执行指标。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "snowflake-coco-cloud-agents-automations",
      "date": "2026-11-03",
      "date_basis": "published",
      "title": "Snowflake CoCo: Agentic Development on Data Without Limits",
      "organization": "Snowflake",
      "url": "https://www.snowflake.com/en/blog/snowflake-coco-cloud-agents-automations/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "data-engineering",
        "sandbox",
        "automation",
        "mcp",
        "memory",
        "governance"
      ],
      "summary": "托管 Cloud Agents 在沙箱中推进长任务，以 Git 分支和 PR 交付变更；自动化由定时、事件或 webhook 触发，以创建者身份运行并记录每次轨迹。Profiles 组织模型、Skills、工具与上下文，项目级记忆支持跨会话复用。",
      "date_evidence": "Snowflake 原文标题前标注 Nov 3, 2026；Cloud Agents 与 Automations 段落说明运行时、触发方式、创建者身份及审计。",
      "evidence_url": "https://www.snowflake.com/en/blog/snowflake-coco-cloud-agents-automations/",
      "caveat": "MCP Gateway Tools、Memory 与动态模型路由均标注 public preview；不能把整组能力统称 GA。后台任务继承创建者授权，PR 和工具审批仍需按具体配置核验。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "vidueu-data-curation-agent-paper",
      "date": "2026-11-03",
      "date_basis": "published",
      "title": "VidUEU-Agent: A Data-Curation Agent for Multimodal Understanding, Editing, and Unified Tasks",
      "organization": "Tengju Ru et al.",
      "url": "https://arxiv.org/abs/2611.04418",
      "section": "research",
      "language": "en",
      "status": "published",
      "tags": [
        "data-curation",
        "multimodal",
        "training-feedback",
        "data-quality"
      ],
      "summary": "将原始视频依次做全局感知与低质量镜头过滤、关键帧/多模态信号挖掘、元数据构建和指令样本路由；再依据下游训练反馈调整数据分布。适合研究数据整理 Agent 的生产—测量—再整理闭环。",
      "date_evidence": "arXiv Submitted on 3 Nov 2026；Submission history v1 为 2026-11-03T07:02:55Z。",
      "evidence_url": "https://arxiv.org/abs/2611.04418",
      "caveat": "这是多模态训练数据整理预印本，不是企业 BI/问数框架。VidUEU-100K 规模与训练收益属于论文自述；本轮仅收录论文，未另行宣称代码或数据集的开放许可已核验。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "hengshi-agent-tasks-analytics-assets",
      "date": "2026-11-04",
      "date_basis": "published",
      "title": "从问数到分析资产：衡石 Data Agent 的任务拆解、辅助创作与持续优化",
      "organization": "衡石科技",
      "url": "https://www.hengshi.com/blog/hengshi-data-agent-tasks-analytics-assets.html",
      "section": "architecture",
      "language": "zh",
      "status": "published",
      "tags": [
        "task-planning",
        "semantic-layer",
        "visualization",
        "tool-feedback",
        "recovery"
      ],
      "summary": "把数据发现、HQL 指标、多步查询、建模与看板创作接成分析交付链；通过工具返回继续推进任务。恢复时分别检查查询、页面修改和资产保存状态，先读取已有资源再补做缺失步骤，减少重试造成重复资产。",
      "date_evidence": "衡石官网技术博客原文标题下标注 2026/11/4；正文给出任务拆解、工具协作、失败恢复和分析资产验收方法。",
      "evidence_url": "https://www.hengshi.com/blog/hengshi-data-agent-tasks-analytics-assets.html",
      "caveat": "销售分析流程是任务设计示例，不是客户实测或固定内部协议。不同能力不必机械串行；可调用范围、写入确认和保存方式依实际版本与入口核验，不能推断存在统一事务或自动回滚。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "oceanbase-ontology-driven-enterprise-data-agents",
      "date": "2026-11-04",
      "date_basis": "published",
      "title": "Ontology-Driven Enterprise Data Agents: From Retrieving Numbers to Explaining Why",
      "organization": "OceanBase / Macro Ji",
      "url": "https://en.oceanbase.com/blog/ontology-driven-enterprise-data-agents",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "ontology",
        "semantic-layer",
        "progressive-context",
        "verification",
        "human-in-the-loop"
      ],
      "summary": "DataPilot 以业务对象、关系、属性和指标定义组织本体，按业务域逐步收窄候选，再沿关系寻找分析维度并用明细验证。先用 Schema、历史 SQL、ETL 和文档生成场景草稿，再由业务专家通过真实问题修订。",
      "date_evidence": "OceanBase 原文标注 Published on November 4, 2026，当前 Updated on 2026-11-05；索引使用首次发表日期，不将后续页更新重复计数。",
      "evidence_url": "https://en.oceanbase.com/blog/ontology-driven-enterprise-data-agents",
      "caveat": "物理可关联不等于业务关系，变化贡献也不证明因果。文中示例解释方法，不是可复现实验成绩；主动探索不等于已授权写数据或触发外部流程。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "datahub-agent-context-kit-walkthrough",
      "date": "2026-04-07",
      "date_basis": "published",
      "title": "Building Autonomous Data Agents with DataHub Agent Context Kit",
      "organization": "DataHub / Nick Adams",
      "url": "https://datahub.com/blog/building-autonomous-data-agents/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "context-graph",
        "mcp",
        "sdk",
        "snowflake",
        "langchain"
      ],
      "summary": "通过目录检索、字段语义、样例 SQL 与血缘工具，为查询规划补齐上下文；比较直接嵌入 SDK 与远程 MCP 两种集成，并以 Snowflake 和 LangChain 展开教程。上下文发现与数仓查询各有工具边界。",
      "date_evidence": "作者原页标注 04.07.26，即2026-04-07；正文含 Agent Context Kit 工具表、Snowflake 接入及 LangChain 示例。",
      "evidence_url": "https://datahub.com/blog/building-autonomous-data-agents/",
      "caveat": "4月教程反映当时的产品与 SDK 接口，代码中的历史命名/拼写需按当前 SDK 文档核对，未在本仓执行。文中 Snowsight 当时不支持 MCP 的描述不能当作11月状态；元数据 mutation 与数据库写权限须分别限制。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "volcengine-private-dataagent-v321",
      "date": "2026-07-01",
      "date_basis": "updated",
      "title": "V3.21.0 — 数据智能体 DataAgent（私有化）",
      "organization": "火山引擎",
      "url": "https://docs.volcengine.com/docs/data-agent-private/v3-21-0?lang=zh",
      "section": "living-doc",
      "language": "zh",
      "status": "living-doc",
      "tags": [
        "retrieval-scope",
        "tools",
        "skills",
        "governance",
        "private-deployment"
      ],
      "summary": "3.21 系列文档（含3.21.1修订）公开知识库/全域资产检索配置、会话级范围切换、SQL/Python/联网工具独立开关与指定知识库的 OpenAPI。适合研究数据发现范围、可用工具和接口凭证如何分层配置。",
      "date_evidence": "原页列出基础版本发布时间2026-04-15，并在文末标注最近更新时间2026.07.01 15:06:25；本条使用文档更新日期，不把3.21.1各项修订都推定为4月15日发布。",
      "evidence_url": "https://docs.volcengine.com/docs/data-agent-private/v3-21-0?lang=zh",
      "caveat": "引擎嵌入版与标准版能力不同；检索优先范围不等于数据库硬授权，资产配置还注明T+1更新。文档日期不能作为所有地域、购买版本均已具备功能的证明。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "paras-seven-ai-analysis-checks-slides",
      "date": "2026-07-10",
      "date_basis": "published",
      "title": "Catch 7 AI Analysis Mistakes Before They Drive Decisions",
      "organization": "Paras Doshi / Maven Lightning Lesson",
      "url": "https://www.insightextractor.com/p/maven-session-slide-deck-download",
      "section": "talk",
      "language": "en",
      "status": "published",
      "tags": [
        "slides",
        "verification",
        "metric-contract",
        "grain",
        "causality"
      ],
      "summary": "18页讲义以虚构分析案例检查决策问题、指标口径、粒度、时间覆盖、来源对账、因果表述与交付状态，要求保存查询和数值证据；适合转成业务分析 Agent 的验收问题清单。",
      "date_evidence": "作者发布页标注 Jul 10, 2026；公开PDF共18页，题名及作者与页面一致。",
      "evidence_url": "https://www.insightextractor.com/p/maven-session-slide-deck-download",
      "material_type": [
        "slides"
      ],
      "materials_url": "https://www.insightextractor.com/api/v1/file/162a46cb-90f0-42da-9c25-6bde530522be.pdf",
      "caveat": "已核验可读讲义；页面提到的录像未单独核验。案例数字明确为虚构教学数据，阈值与检查严度需按任务调整，不是统一行业标准。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "paras-three-analytics-harnesses-slides",
      "date": "2026-07-16",
      "date_basis": "published",
      "title": "Build 3 AI Harnesses for Trustworthy Analytics",
      "organization": "Paras Doshi / Maven Lightning Lesson",
      "url": "https://www.insightextractor.com/p/session-slide-and-recording-build",
      "section": "talk",
      "language": "en",
      "status": "published",
      "tags": [
        "slides",
        "harness",
        "metric-contract",
        "validation",
        "evidence",
        "memory"
      ],
      "summary": "15页讲义把上下文、校验和证据复用拆成三个 harness，以版本化指标文件、独立检查职责和含失效条件的证据记录串联；展示可审阅的文件布局，适合设计分析 Agent 的交付门禁。",
      "date_evidence": "作者发布页标注 Jul 16, 2026；公开PDF为15页，首页标注 MAVEN LIVE JUL16，已读取全文并查看文件布局页。",
      "evidence_url": "https://www.insightextractor.com/p/session-slide-and-recording-build",
      "material_type": [
        "slides"
      ],
      "materials_url": "https://www.insightextractor.com/api/v1/file/c04ec353-b782-4432-b631-1a0d42353cbb.pdf",
      "caveat": "示例 Stop hook 只输出提醒，不是强制拦截；reopen_if 也需实际实现检查，写入YAML不会自动失效。另一个模型不天然构成独立真值，需确定性校验、查询对账与人工判断。本轮未独立验证录像。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "aws-multimodal-data-agent-orchestration",
      "date": "2026-08-04",
      "date_basis": "published",
      "title": "Multi-Agent Multimodal Data Analysis on AWS – Part 2: Multi-Agent Orchestration and Predictive Analytics",
      "organization": "AWS for Industries",
      "url": "https://aws.amazon.com/blogs/industries/multi-agent-multimodal-data-analysis-on-aws-part-2-multi-agent-orchestration-and-predictive-analytics/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "multimodal",
        "supervisor",
        "a2a",
        "mcp",
        "governance",
        "data-science"
      ],
      "summary": "以 Strands 与 AgentCore 部署专门的数据/知识 Agent，MCP 连接不同模态的数据服务，Supervisor 通过 A2A 调用并综合结果；提供身份校验、共享客户端和超时的编排细节。前篇负责目录、订阅与数据准备。",
      "date_evidence": "AWS Part 2 与前篇均标注04 AUG 2026；Part2正文逐项说明专门Agent、Supervisor、Cognito与A2A部署。",
      "evidence_url": "https://aws.amazon.com/blogs/industries/multi-agent-multimodal-data-analysis-on-aws-part-2-multi-agent-orchestration-and-predictive-analytics/",
      "related_materials": [
        {
          "title": "Part 1: Data Governance and Visualization",
          "url": "https://aws.amazon.com/blogs/industries/multi-agent-multimodal-data-analysis-on-aws-part-1-data-governance-and-visualization/",
          "date": "2026-08-04",
          "note": "先读目录、订阅授权与Synthea合成数据准备，再读多Agent编排。"
        }
      ],
      "caveat": "使用Synthea相关合成数据演示研究架构，不是临床有效性验证。本仓未部署云资源、运行模型或复现预测成绩，服务配置、数据授权与成本需另行评估。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "volcengine-analysis-agent-build-video",
      "date": "2026-08-20",
      "date_basis": "published",
      "title": "分析洞察快速入门-04 搭建流程概览",
      "organization": "火山引擎数智平台 VeDI",
      "url": "https://developer.volcengine.com/videos/7566240392669528073",
      "section": "talk",
      "language": "zh",
      "status": "published",
      "tags": [
        "video-page",
        "tutorial",
        "configuration",
        "analytics"
      ],
      "summary": "官方培训页面按管理员、分析师和技术人员的分工介绍企业分析智能体从前置准备到搭建配置的总体流程，适合作为产品文档前的入门导航。",
      "date_evidence": "公开页面元数据 publish_time 为1787192306，即2026-08-20T02:18:26Z；搜索页显示2026-08-20，标题与发布方均匹配。",
      "evidence_url": "https://developer.volcengine.com/videos/7566240392669528073",
      "material_type": [
        "video-page"
      ],
      "material_review": {
        "checked_on": "2026-11-05",
        "status": "page-and-description-verified",
        "note": "已核验官方视频页面、发布日期与简介；未独立测试完整播放或获取逐字稿。"
      },
      "caveat": "作为官方视频入口收录，不把页面可访问写成完整回放已逐段验证；具体参数与执行能力需结合当前产品版本文档。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "datahub-secupi-agent-runtime-governance",
      "date": "2026-09-10",
      "date_basis": "published",
      "title": "Enhancing Agent Governance with DataHub and SecuPi: From Trusted Context to Runtime Control",
      "organization": "DataHub / Stephen Goldbaum",
      "url": "https://datahub.com/blog/agent-governance-with-datahub-and-secupi/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "identity",
        "context-graph",
        "runtime-policy",
        "masking",
        "audit"
      ],
      "summary": "将 DataHub 的语义、血缘和质量上下文，与 SecuPi 的用户身份保留、字段级屏蔽和查询时策略执行分开，串起谁提问、用何上下文、访问哪些数据与哪些字段被屏蔽的审计链。",
      "date_evidence": "作者原页标注09.10.26，即2026-09-10；正文明确上下文层与运行时授权的分工。",
      "evidence_url": "https://datahub.com/blog/agent-governance-with-datahub-and-secupi/",
      "caveat": "属于厂商集成方案，Context Intelligence/Context Hub与SecuPi有产品配置依赖；不把文中客户准确率作为本仓复现结果，也不将预告的9月22日webinar自动标为已核验录像。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "aws-atscale-agentcore-semantic-layer",
      "date": "2026-10-20",
      "date_basis": "published",
      "title": "Build a semantic layer for agentic AI on AWS with AtScale and Amazon Bedrock AgentCore",
      "organization": "AWS / AtScale",
      "url": "https://aws.amazon.com/blogs/machine-learning/build-a-semantic-layer-for-agentic-ai-on-aws-with-atscale-and-amazon-bedrock-agentcore/",
      "section": "architecture",
      "language": "en",
      "status": "published",
      "tags": [
        "semantic-layer",
        "mcp",
        "identity",
        "agentcore",
        "sml",
        "deployment"
      ],
      "summary": "将 Git 中的 SML 维度、层级与指标通过 MCP 提供给 Strands Agent；拆解入口JWT、出口凭证、AtScale角色及查询时过滤，计算下推Redshift。可与同系列图谱/虚拟化方案比较数据放置与授权路径。",
      "date_evidence": "AWS原文署名行标注20 OCT 2026，示例以AtScale C2026.7为版本依据；正文含语义建模、MCP注册、身份路径与EKS部署。",
      "evidence_url": "https://aws.amazon.com/blogs/machine-learning/build-a-semantic-layer-for-agentic-ai-on-aws-with-atscale-and-amazon-bedrock-agentcore/",
      "related_materials": [
        {
          "title": "Stardog semantic layer + AgentCore",
          "url": "https://aws.amazon.com/blogs/machine-learning/build-a-semantic-layer-for-agentic-ai-on-aws-with-stardog-and-amazon-bedrock-agentcore/",
          "date": "2026-07-10",
          "note": "同一C360问题，图谱与跨源联邦路线；MCP路径有产品API可用性前提。"
        },
        {
          "title": "Denodo semantic layer + AgentCore",
          "url": "https://aws.amazon.com/blogs/machine-learning/build-a-semantic-layer-for-agentic-ai-on-aws-with-denodo-and-amazon-bedrock-agentcore/",
          "note": "同系列数据虚拟化路线，作为对照材料，不另增加条目。"
        }
      ],
      "caveat": "SML规范开放不等于AtScale产品免费或开源；示例将分析数据预先汇集到Redshift，并不是每次查询直接跨多个引擎。未部署、压测或复现厂商效果。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "google-dak-remote-mcp-governance-guide",
      "date": "2026-10-26",
      "date_basis": "updated",
      "title": "Use MCP servers — Google Cloud Data Agent Kit",
      "organization": "Google Cloud",
      "url": "https://docs.cloud.google.com/data-agent-kit/use-mcp-servers",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "mcp",
        "iam",
        "impersonation",
        "model-armor",
        "tools"
      ],
      "summary": "说明远程MCP连接、MCP Tool User与产品数据权限、服务账号模拟、Model Armor检查/阻断模式及按工具属性编写IAM策略。适合把Agent的过程指引与云端实际授权分开设计。",
      "date_evidence": "Google官方文档末尾标注 Last updated 2026-10-26 UTC；本条是该文档快照，不把日期当作全部MCP功能的首次发布日。",
      "evidence_url": "https://docs.cloud.google.com/data-agent-kit/use-mcp-servers",
      "caveat": "MCP调用权限不替代具体服务的数据权限；IDE扩展与Coding Agent插件入口不同，配置方式不能照搬。Model Armor支持地域和跨项目检查行为需逐项核对。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "sqlbot-chatbi-mcp-application",
      "date": "2026-10-29",
      "date_basis": "released",
      "title": "SQLBot",
      "organization": "DataEase / FIT2CLOUD",
      "url": "https://github.com/dataease/SQLBot",
      "section": "open-source",
      "language": "zh",
      "status": "active-project",
      "tags": [
        "chatbi",
        "rag",
        "mcp",
        "embedding",
        "workspace-isolation",
        "modified-gpl"
      ],
      "summary": "面向嵌入式ChatBI的应用与MCP提供方，以术语、SQL示例、RAG及工作空间组织问数和图表；v1.10.3增加动态向量检索策略并修正数据源、审计和会话鉴权问题。",
      "date_evidence": "GitHub Release v1.10.3 published_at为2026-10-29T04:24:34Z；不是本日新发布。2026-11-05核对archived=false及LICENSE正文。",
      "evidence_url": "https://github.com/dataease/SQLBot/releases/tag/v1.10.3",
      "license": "FIT2CLOUD modified GPLv3 (additional conditions)",
      "license_url": "https://github.com/dataease/SQLBot/blob/main/LICENSE",
      "license_note": "LICENSE说明以GPLv3为基础并附加LOGO、版权信息及贡献条款；GitHub API为NOASSERTION，因此不标为无附加条件的标准GPLv3，也不据公开源码自动认定OSI开源。",
      "component_type": "chatbi-application-and-mcp-provider",
      "technical_material_url": "https://github.com/dataease/SQLBot/blob/main/README.md",
      "mechanism": "RAG与可维护术语/SQL示例提供问数上下文，应用执行查询与可视化，并通过嵌入和MCP向外提供能力。",
      "use_case": "研究中文ChatBI、企业嵌入和工具接入，或比较应用层隔离与数据库权限的关系。",
      "caveat": "是ChatBI应用/工具提供方，不能据此假定具备完整自主分析验证闭环。README部署示例包含privileged选项，需与生产权限配置分开评审；许可边界以实际分发组件为准。",
      "release_updates": [
        {
          "version": "v1.10.3",
          "published_at": "2026-10-29T04:24:34Z",
          "url": "https://github.com/dataease/SQLBot/releases/tag/v1.10.3"
        }
      ],
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "databricks-genie-agent-api-guide",
      "date": "2026-10-30",
      "date_basis": "updated",
      "title": "Use the Genie Agents API",
      "organization": "Databricks",
      "url": "https://docs.databricks.com/aws/en/genie-agents/conversation-api",
      "section": "living-doc",
      "language": "en",
      "status": "living-doc",
      "tags": [
        "api",
        "oauth",
        "governance",
        "agent-management",
        "ci-cd"
      ],
      "summary": "区分有状态Chat、Agent mode与管理API，说明OAuth主体、Agent ACL、Unity Catalog权限，以及如何把示例SQL、指令和基准题随Agent配置跨工作区维护。适合自建应用接入与配置交付。",
      "date_evidence": "官方文档顶端标注 Last updated on Oct 30, 2026；旧 /genie/conversation-api 已重定向到 /genie-agents/conversation-api，本条使用当前规范URL。",
      "evidence_url": "https://docs.databricks.com/aws/en/genie-agents/conversation-api",
      "caveat": "查询接口与创建/更新Agent的管理接口权限不同；U2M与服务主体M2M应按调用主体授权。接口接通不证明业务答案准确，端点、报告流与功能阶段仍以对应REST文档为准。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    },
    {
      "id": "hengshi-multiturn-context-preferences-learning",
      "date": "2026-11-02",
      "date_basis": "published",
      "title": "让多轮问数保持口径一致：Data Agent 的上下文、偏好与学习开关",
      "organization": "衡石科技",
      "url": "https://www.hengshi.com/blog/hengshi-data-agent-context-preferences-learning.html",
      "section": "architecture",
      "language": "zh",
      "status": "published",
      "tags": [
        "multi-turn",
        "context",
        "memory",
        "preferences",
        "regression"
      ],
      "summary": "分开维护当前分析条件、个人展示偏好、正式指标知识和后台学习；用同会话/新会话/不同角色对照检查条件延续，再以原问题和相邻问法复测修正。提供人工回归记录示例。",
      "date_evidence": "官网原文标注2026/11/2；此前11月3日转载日期不是首次发表日。本轮按专题补录11月2日原文，不把它称为11月5日新发布。",
      "evidence_url": "https://www.hengshi.com/blog/hengshi-data-agent-context-preferences-learning.html",
      "caveat": "SELF_EVOLUTION_ENABLED是独立管理员配置项，默认关闭不等于本轮对话不能承接上下文；实际值依部署版本核对。JSON是本地人工验收示例，不是产品会话API；不能承诺越用越准。",
      "added_on": "2026-11-05",
      "last_verified": "2026-11-05"
    }
  ],
  "refresh_history": [
    {
      "id": "2026-11-06-project-evidence",
      "kind": "evidence-refresh",
      "as_of": "2026-11-06",
      "window_start": "2026-11-06",
      "previous_as_of": "2026-11-05",
      "base_commit": "1dbf24e299420c718893ef45429ae9a127f16f36",
      "previous_total": 168,
      "added_total": 0,
      "updated_total": 6,
      "added_ids": [],
      "updated_ids": [
        "wrenai",
        "vanna-agent-framework",
        "db-gpt",
        "google-mcp-toolbox-databases",
        "microsoft-data-formulator",
        "jupyter-ai"
      ],
      "agenda_reviewed_ids": [],
      "review_note": "定向复核六个已有项目的 release、根许可、归档状态与固定提交技术文档；无新增条目。MCP Toolbox 更新至 v1.16.0，Data Formulator 明确 beta，Vanna 修正行级安全措辞；不表示其余条目或历史支持链接全部重验，也没有部署和基准复现。"
    },
    {
      "id": "2026-11-05-project-docs-talks",
      "kind": "thematic-enrichment",
      "as_of": "2026-11-05",
      "window_start": "2026-01-01",
      "previous_as_of": "2026-11-05",
      "base_commit": "e6686da961fbf417c15d342ee3d29f97c6abab61",
      "previous_total": 156,
      "added_total": 12,
      "updated_total": 5,
      "added_ids": [
        "datahub-agent-context-kit-walkthrough",
        "volcengine-private-dataagent-v321",
        "paras-seven-ai-analysis-checks-slides",
        "paras-three-analytics-harnesses-slides",
        "aws-multimodal-data-agent-orchestration",
        "volcengine-analysis-agent-build-video",
        "datahub-secupi-agent-runtime-governance",
        "aws-atscale-agentcore-semantic-layer",
        "google-dak-remote-mcp-governance-guide",
        "sqlbot-chatbi-mcp-application",
        "databricks-genie-agent-api-guide",
        "hengshi-multiturn-context-preferences-learning"
      ],
      "updated_ids": [
        "spring-ai-alibaba-dataagent",
        "db-gpt",
        "wrenai",
        "datahub-analytics-agent",
        "google-data-agent-kit-plugin"
      ],
      "agenda_reviewed_ids": [],
      "review_note": "专题补录2026年技术文档与公开分享，不声称全部为上次上传后的新发布；五个既有项目补充阅读路线与证据，支持链接不另计项目数。",
      "metadata_changes": [
        {
          "id": "spring-ai-alibaba-dataagent",
          "field": "date_basis/date",
          "before": "updated/2026-09-20",
          "after": "released/2026-07-29",
          "reason": "原日期未附证据；改用可核验RC发布记录。"
        },
        {
          "id": "datahub-analytics-agent",
          "field": "date_basis/date",
          "before": "repo_created/2026-04-22",
          "after": "released/2026-07-29",
          "reason": "已有明确release，替换较弱的创建时间证据。"
        },
        {
          "id": "google-data-agent-kit-plugin",
          "field": "date_basis/date",
          "before": "repo_created/2026-04-02",
          "after": "released/2026-10-30",
          "reason": "已有明确release，替换较弱的创建时间证据。"
        }
      ]
    },
    {
      "as_of": "2026-11-05",
      "window_start": "2026-11-03",
      "previous_as_of": "2026-11-02",
      "base_commit": "0334da5b7a7df8369bd4b2c820f7792addc7e02c",
      "previous_total": 150,
      "added_total": 6,
      "updated_total": 3,
      "corrected_total": 1,
      "added_ids": [
        "snowflake-agent-gateway-security-governance",
        "snowflake-business-ontology-stewardship",
        "snowflake-coco-cloud-agents-automations",
        "vidueu-data-curation-agent-paper",
        "hengshi-agent-tasks-analytics-assets",
        "oceanbase-ontology-driven-enterprise-data-agents"
      ],
      "updated_ids": [
        "altimate-code",
        "nao",
        "dbt-mcp-server"
      ],
      "corrected_ids": [
        "alibaba-dataworks-agent"
      ],
      "agenda_reviewed_ids": [],
      "review_note": "复查31个已收录公开实现的近期 release；不把日常提交计为新发布。会议议程沿用11月2日的材料核验状态，本轮未宣称完成全量会后材料复查。"
    },
    {
      "as_of": "2026-11-02",
      "window_start": "2026-10-23",
      "previous_as_of": "2026-10-22",
      "base_commit": "228d833f557175f3ce06242c413f532df59c52d9",
      "previous_total": 133,
      "added_total": 17,
      "updated_total": 7,
      "added_ids": [
        "adapra-paper",
        "atlan-analytics-agent-fde-playbook",
        "keyhole-agentic-harness-data-engineering",
        "kx-qsql-mcp-verification",
        "calimem-paper",
        "clickhouse-jev-retrieval-failure-labeling",
        "data-marc-dynamic-data-agent-evaluation",
        "hex-4-agent-native-execution-architecture",
        "m4bench-paper",
        "yunqi-dacon-knowledge-ontology-recap",
        "atlan-post-ai-data-stack-context-conference",
        "datus-dosi-executable-business-context",
        "cube-data-model-factory",
        "motherduck-dive-apps-agent-actions",
        "pena-fabric-data-agent-repeatability",
        "yunqi-semantic-view-trusted-context",
        "hex-governed-mode"
      ],
      "updated_ids": [
        "datus-agent",
        "dbt-mcp-server",
        "wrenai",
        "nao",
        "google-mcp-toolbox-databases",
        "altimate-code",
        "alibaba-dataworks-agent"
      ],
      "agenda_reviewed_ids": [
        "qcon-shanghai-track",
        "qcon-shanghai-netease-das",
        "qcon-shanghai-kuaishou",
        "qcon-shanghai-datus",
        "qcon-shanghai-rednote",
        "qcon-shanghai-meituan-aide",
        "dacon-beijing-2026-agenda"
      ]
    }
  ]
}
