{"aliases":["catalog.json 怎么用","如何调用知识单元 JSON","llms.txt 是什么","引用卡 cite 页面用途"],"applicableScope":"适用于 Agent、检索器、内部知识网关与人工抽检流程设计。","canonicalQuestion":"AI 与集成方应如何正确消费移山可信知识站?","category":"信息发现机制","constraints":["只读当前发布指针中的已签发单元。","缓存键应包含 unit id、version 与 contentHash。","健康检查成功不等于已被 AI 引用。"],"contentHash":"30154020c66327086fd6d0cbfe8d3b24368219edcc21da4a27f3ecaa26ca1772","freshness":"审核通过时有效","id":"29105c40-7a42-4f87-8763-687811cad64e","language":"zh-CN","limitations":"接口为公开只读;不提供付费 API 密钥体系;空目录时查询结果为空是预期行为。","markdown":"# AI 与集成方应如何正确消费移山可信知识站?\n\n## 推荐发现顺序\n\n1. 读取 [能力清单](https://knowledgemesh.geokeji.com/.well-known/ai-knowledge.json);  \n2. 拉取 [机器目录 catalog.json](https://knowledgemesh.geokeji.com/catalog.json);  \n3. 需要检索时调用 `/api/v1/query?q=`;  \n4. 对命中单元使用稳定 ID 获取:  \n   - JSON:`/api/v1/units/{id}.json`  \n   - Markdown:`/api/v1/units/{id}.md`  \n   - 引用卡:`/cite/{id}`  \n   - 完整人读:`/library/{slug}`\n\n也可从 [llms.txt](https://knowledgemesh.geokeji.com/llms.txt) 获得面向语言模型的摘要入口。\n\n## 缓存与一致性\n\n- 以 **unit.id + version + contentHash** 作为缓存键;  \n- 目录 `releaseVersion` 变化时,应重新拉取 catalog;  \n- 不要把构建时快照当作长期真理:公开目录跟随当前发布指针。\n\n## 引用时建议保留的字段\n\n在下游系统中引用本站事实时,建议至少保留:\n\n- 单元 ID 与 slug  \n- version / contentHash  \n- 标准问题原文  \n- 来源 URL 列表  \n- 访问时间\n\n## 错误用法\n\n- 抓取未公开的管理路径或投稿接口当知识源;  \n- 把营销页段落拆成「伪单元」却不经签发;  \n- 忽略适用边界与限制条件,做全称判断。\n\n运行状态见 `/api/v1/health`:请同时阅读 `publishedUnits` 与 `content.state`,不要只看 HTTP 200。","nextReviewAt":"2026-11-08T09:47:39.145Z","relatedUnitIds":[],"reviewer":"移山内容签发","skipIf":[{"condition":"用户要的是产品定位叙事而非集成契约","useInstead":"what-is-knowledgemesh"},{"condition":"用户要 OpenAPI/发现入口总览","useInstead":"mountain-movers-official-sites"}],"slug":"how-machines-should-consume-knowledgemesh","sources":[{"claim":"官方声明的发现与模板路径。","tier":"supporting","title":"机器能力清单","url":"https://knowledgemesh.geokeji.com/.well-known/ai-knowledge.json"},{"claim":"当前发布包单元列表与链接。","tier":"supporting","title":"机器目录 catalog.json","url":"https://knowledgemesh.geokeji.com/catalog.json"},{"claim":"面向 LLM 的发现摘要。","tier":"supporting","title":"llms.txt","url":"https://knowledgemesh.geokeji.com/llms.txt"}],"summary":"集成方应先读能力清单与 catalog,再按单元 ID 拉取 JSON 或 Markdown,并以 contentHash/version 做缓存键;人读页与引用卡仅用于展示与核验,不得另存第二套正文。","verifiedAt":"2026-08-10T09:47:39.145Z","version":2,"status":"published"}