{"id":"7c1bf744-eb72-435d-8d1f-fc9edf9c0664","entityType":"agent","slug":"clawhub-itxiaohao-lingzao","name":"灵造","canonicalUrl":"https://www.xpersona.co/agent/clawhub-itxiaohao-lingzao","canonicalPath":"/agent/clawhub-itxiaohao-lingzao","generatedAt":"2026-10-10T10:43:38.195Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-10T07:35:37.553Z","emptyReason":null},"description":"跨平台创作者研究与自媒体运营 Skill Skill: 灵造 Owner: itxiaohao Summary: 跨平台创作者研究与自媒体运营 Skill Tags: latest:0.1.106 Version history: v0.1.106 | 2026-09-01T02:55:53.103Z | user Release Lingzao Skill 0.1.106: adds direct WeChat Channels share-link support for short-video copy extraction while preserving existing public research boundaries. v0.1.105 | 2026-08-20T05:43:54.807Z | user Release Lingzao Skill 0.1.105: adds complete Xiaohongshu image-note body im","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.6K downloads reported by the source. Last updated 10/10/2026.","installCommand":"clawhub skill install s1771g8qxavq78skbwwxtfg9z9889mm0:lingzao","sourceUrl":"https://clawhub.ai/itxiaohao/lingzao","homepage":"https://clawhub.ai/itxiaohao/skills/lingzao","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/itxiaohao/lingzao","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/itxiaohao/skills/lingzao","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":64,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"跨平台创作者研究与自媒体运营 Skill Skill: 灵造 Owner: itxiaohao Summary: 跨平台创作者研究与自媒体运营 Skill Tags: latest:0.1.106 Version history: v0.1.106 | 2026-09-01T02:55:53.103Z | user R"},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-10T07:35:37.553Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T07:35:37.553Z","emptyReason":null},"stars":null,"forks":null,"downloads":1587,"packageName":null,"latestVersion":"0.1.106","tractionLabel":"1.6K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T07:35:37.552Z","emptyReason":null},"lastUpdatedAt":"2026-10-10T07:35:37.553Z","lastCrawledAt":"2026-10-10T07:35:37.552Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-11T07:35:37.552Z","lastVerifiedAt":null,"highlights":[{"version":"0.1.106","createdAt":"2026-09-01T02:55:53.103Z","changelog":"Release Lingzao Skill 0.1.106: adds direct WeChat Channels share-link support for short-video copy extraction while preserving existing public research boundaries.","fileCount":55,"zipByteSize":619481},{"version":"0.1.105","createdAt":"2026-08-20T05:43:54.807Z","changelog":"Release Lingzao Skill 0.1.105: adds complete Xiaohongshu image-note body images, WeChat Channels guidance, scope-first conversations, explicit extract-video-copy operation recovery, and direct Douyin short-link detail routing.","fileCount":55,"zipByteSize":619457},{"version":"0.1.101","createdAt":"2026-08-09T10:56:34.242Z","changelog":"Release Lingzao Skill 0.1.101: removes implicit cross-process image request persistence and credential-derived fingerprints; ambiguous image requests now resume only with an explicit client_request_id.","fileCount":59,"zipByteSize":687646},{"version":"0.1.100","createdAt":"2026-08-09T03:25:50.875Z","changelog":"Release Lingzao Skill 0.1.100: adds durable image-generation retry recovery, one-to-one reference mapping, and concise insufficient-credit guidance.","fileCount":55,"zipByteSize":622862},{"version":"0.1.94","createdAt":"2026-07-20T09:08:18.655Z","changelog":"Release Lingzao Skill 0.1.94: shares the Lingzao feature usage manual once after a confirmed install or update.","fileCount":50,"zipByteSize":315980},{"version":"0.1.93","createdAt":"2026-07-17T02:12:44.956Z","changelog":"Release Lingzao Skill 0.1.93: adds TikTok, YouTube, and Instagram to six platform-neutral creator-research commands, improves cursor and target safety, and updates creator workflow playbooks.","fileCount":50,"zipByteSize":315456},{"version":"0.1.69","createdAt":"2026-06-30T10:18:41.907Z","changelog":"Release Lingzao Skill 0.1.69: adds the SkillHub-ready Chinese main Skill, brand Brief and benchmark-copy workflows, and packaged Lingzao logo assets for agent metadata.","fileCount":43,"zipByteSize":277364},{"version":"0.1.66","createdAt":"2026-06-27T04:12:20.914Z","changelog":"Release Lingzao Skill 0.1.66: adds get-article-detail --output for saving full WeChat article details to local Markdown before chat summarization.","fileCount":42,"zipByteSize":265737}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s1771g8qxavq78skbwwxtfg9z9889mm0:lingzao","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-10T10:43:38.191Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-itxiaohao-lingzao/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-10T07:35:37.553Z","emptyReason":null},"readme":"Skill: 灵造\n\nOwner: itxiaohao\n\nSummary: 跨平台创作者研究与自媒体运营 Skill\n\nTags: latest:0.1.106\n\nVersion history:\n\nv0.1.106 | 2026-09-01T02:55:53.103Z | user\n\nRelease Lingzao Skill 0.1.106: adds direct WeChat Channels share-link support for short-video copy extraction while preserving existing public research boundaries.\n\nv0.1.105 | 2026-08-20T05:43:54.807Z | user\n\nRelease Lingzao Skill 0.1.105: adds complete Xiaohongshu image-note body images, WeChat Channels guidance, scope-first conversations, explicit extract-video-copy operation recovery, and direct Douyin short-link detail routing.\n\nv0.1.101 | 2026-08-09T10:56:34.242Z | user\n\nRelease Lingzao Skill 0.1.101: removes implicit cross-process image request persistence and credential-derived fingerprints; ambiguous image requests now resume only with an explicit client_request_id.\n\nv0.1.100 | 2026-08-09T03:25:50.875Z | user\n\nRelease Lingzao Skill 0.1.100: adds durable image-generation retry recovery, one-to-one reference mapping, and concise insufficient-credit guidance.\n\nv0.1.94 | 2026-07-20T09:08:18.655Z | user\n\nRelease Lingzao Skill 0.1.94: shares the Lingzao feature usage manual once after a confirmed install or update.\n\nv0.1.93 | 2026-07-17T02:12:44.956Z | user\n\nRelease Lingzao Skill 0.1.93: adds TikTok, YouTube, and Instagram to six platform-neutral creator-research commands, improves cursor and target safety, and updates creator workflow playbooks.\n\nv0.1.69 | 2026-06-30T10:18:41.907Z | user\n\nRelease Lingzao Skill 0.1.69: adds the SkillHub-ready Chinese main Skill, brand Brief and benchmark-copy workflows, and packaged Lingzao logo assets for agent metadata.\n\nv0.1.66 | 2026-06-27T04:12:20.914Z | user\n\nRelease Lingzao Skill 0.1.66: adds get-article-detail --output for saving full WeChat article details to local Markdown before chat summarization.\n\nv0.1.65 | 2026-06-26T03:11:56.356Z | user\n\nRelease Lingzao Skill 0.1.65: adds WeChat official-account article commands and same-stage peer horizontal diagnosis workflow, keeps Skill routing aligned with the 2026-06-26 main release.\n\nv0.1.63 | 2026-06-24T02:49:22.727Z | user\n\n- Added the Xiaohongshu operation task tree playbook for routing creator-operation requests.\n- Updated image intake and generation playbooks, including reference-image handling and visual quality gates.\n- Updated Skill metadata/index for the new public workflow guidance.\n\nv0.1.62 | 2026-06-22T12:46:33.181Z | user\n\n- Added six new playbooks for advanced creator-content workflows, including creator case analysis, image generation, note breakdown, cross-platform distribution, and content readiness checks.\n- Updated documentation to reference the new playbooks: creator-case-general-analysis-framework.md, image-generation-agent-integration-guide.md, image-generation-execution-workflow.md, mother-content-cross-platform-distribution.md, pre-publish-readiness-check.md, and single-note-breakdown-workflow.md.\n- Introduced more detailed guidance for cross-platform content distribution, image generation workflows, and note analysis.\n- Removed the deprecated skill-card.md file.\n\nv0.1.40 | 2026-06-19T00:20:45.174Z | user\n\n- Added prompt-based creator image generation feature when the user explicitly requests image creation.\n- Updated skill description to include image asset generation.\n- Removed outdated documentation file: skill-card.md.\n\nv0.1.39 | 2026-06-18T02:37:50.598Z | user\n\n- Major upgrade: Adds a comprehensive set of 23 agent playbooks for Xiaohongshu creator workflows (strategy, keyword insight, account diagnosis, note review, visual/cover generation, etc.) under playbooks/.\n- Playbooks enable structured, workflow-based answers for creator research, note publishing, account benchmarking, and audience checks, instead of isolated lookups.\n- Expands skill coverage to search suggestions, public-comments lookup, and detailed agent SOPs for each workflow.\n- Updates SKILL.md usage guidance: instructs agents when to use playbooks, how to guide users concretely, and how to sync results to user knowledge bases (ima, Obsidian, 飞书) only upon request.\n- Deprecated the old skill-card.md; documentation and workflow references now reside in SKILL.md and playbooks/.\n- Keeps public promises focused: No assurance of guaranteed growth, monetization, or copying; supports only public-content research and workflow automation.\n\nv0.1.9 | 2026-06-08T07:01:07.621Z | auto\n\n- Added detailed usage instructions, clarifying when to use each command (search, profile lookup, analysis, extraction, etc).\n- Updated workflow guidance for creator profile lookups and analysis; clarified when to use `get-user-info`, `get-user-posted-notes`, and `analyze-user-profile`.\n- Provided explicit command-line usage examples for each function, including recommended defaults (e.g., use `--limit 20` with `analyze-user-profile`).\n- Included troubleshooting and update steps for installation, configuration, and wrapper repair.\n- Guidance now instructs not to use `get-user-info` and `get-user-posted-notes` in fixed pairs unless both sets of data are required.\n- Ensured distinction between basic and deep analysis use cases, especially around subtitle/copy artifact extraction and usage.\n\nArchive index:\n\nArchive v0.1.106: 55 files, 619481 bytes\n\nFiles: agents/openai.yaml (3992b), agents/openclaw.yaml (273b), assets/lingzao-logo.png (292012b), index.md (17109b), playbooks/account-report-evidence-visual-contract.md (9858b), playbooks/atian-creator-judgment-framework.md (10985b), playbooks/audience-persona-fit-check.md (5699b), playbooks/beginner-account-start-and-topic-radar.md (25119b), playbooks/benchmark-account-discovery-quality-gate.md (23742b), playbooks/brand-brief-to-content-workflow.md (11836b), playbooks/comparable-account-breakdown-report-template.md (16279b), playbooks/content-knowledge-base-workflow.md (27596b), playbooks/copy-paste-prompt-scope-boundary.md (10036b), playbooks/creator-case-general-analysis-framework.md (14268b), playbooks/draft-rewrite-and-benchmark-workflow.md (13577b), playbooks/image-generation-agent-integration-guide.md (20364b), playbooks/image-generation-execution-workflow.md (11935b), playbooks/keyword-insight-report-template.md (10946b), playbooks/keyword-to-publishable-content-package.md (41649b), playbooks/lingzao-progressive-interaction-map.md (42396b), playbooks/monetization-path-judgment-library.md (10874b), playbooks/mother-content-cross-platform-distribution.md (7847b), playbooks/post-publish-data-review-workflow.md (12962b), playbooks/pre-publish-readiness-check.md (7212b), playbooks/product-judgment-and-feedback-loop.md (8756b), playbooks/publishing-keyword-design-check.md (9503b), playbooks/reference-image-graphic-note-workflow.md (5291b), playbooks/research-scope-guard.md (5065b), playbooks/retention-and-follow-up-loop.md (12830b), playbooks/router-cases.json (3817b), playbooks/router-index.json (21352b), playbooks/self-account-diagnosis-report-template.md (27899b), playbooks/self-account-peer-horizontal-diagnosis.md (13270b), playbooks/single-note-breakdown-workflow.md (20268b), playbooks/track-difficulty-judgment-library.md (12236b), playbooks/travel-handdrawn-map-visual-workflow.md (7458b), playbooks/visual-generation-and-cover-workflow.md (11172b), playbooks/visual-reference-style-library.md (23833b), playbooks/wechat-benchmark-fit-and-writing-workflow.md (10567b), playbooks/weekly-content-motherpack-distributor.md (9459b), playbooks/xhs-content-compliance-risk-gate.md (5886b), playbooks/xhs-operation-task-tree.md (12978b), playbooks/xhs-platform-management-risk-baseline.md (5688b), playbooks/xhs-profile-bio-design.md (7960b), playbooks/xhs-title-design-check.md (7825b), playbooks/zero-beginner-onboarding-gate.md (9225b), scripts/check_playbook_router.py (5785b), scripts/check_version.py (275b), scripts/configure.py (3015b), scripts/lingzao_client.py (99047b), scripts/setup.sh (1824b), skill-card.md (2905b), SKILL.md (51196b), VERSION (8b), _meta.json (128b)\n\nFile v0.1.106:SKILL.md\n\n---\nname: lingzao\ndescription: 灵造是给 WorkBuddy、OpenClaw、Codex 等 Agent 使用的跨平台创作者研究与自媒体运营 Skill。安装免费，可先做选题、标题、封面、账号诊断、发布检查、复盘，以及公众号对标适配判断和原创写作；查询公开内容、评论、短视频文案、公众号文章数据或生成图片时需要配置灵造 API Key。\n---\n\n# 灵造：跨平台创作者研究与自媒体运营 Skill\n\n灵造是一个主 Skill，不需要拆成标题、封面、账号诊断、图片生成等多个 Skill。\n安装后，WorkBuddy、OpenClaw、Codex 等 Agent 会先把你的问题路由到合适的\n创作者运营 playbook；只有当你需要查询公开内容、读取评论、提取短视频文案、\n查看公众号文章数据或生成图片时，才需要配置灵造 API Key。\n\n## 安装完成后的首次提示\n\n当当前对话刚刚完成灵造 Skill 的安装或更新时，只有确认安装成功后，才在最终回复中\n主动告诉用户一次下面的使用手册；安装失败、尚未验证成功或普通后续对话不要重复发送：\n\n> 灵造已安装完成。你可以查看《灵造功能使用手册》：\n> https://my.feishu.cn/docx/Y2HQdj5mzoFx4vxfij3cl9TRnjh?from=from_copylink\n> 快速了解灵造的功能和使用方法～\n\n## 从这里开始\n\n| 你现在想做 | 可以直接这样问 Agent |\n| --- | --- |\n| 找内容方向 | “用灵造帮我围绕这个关键词做小红书、抖音、TikTok、Instagram 或 YouTube 选题，给我 10 个可发方向。” |\n| 找对标账号 | “帮我找这个赛道值得学习的对标账号，并说明每个账号适合学什么。” |\n| 拆一条笔记或视频 | “分析这条内容为什么有效，拆成标题、封面、结构、评论需求和可复用模板。” |\n| 改标题和封面 | “基于我的草稿，给我 3 个最强标题和 5 个小红书封面方向。” |\n| 做发布前检查 | “发布前帮我检查标题、封面、前 3 行、关键词和用户点击理由。” |\n| 做发布后复盘 | “根据这条内容的数据和评论，帮我判断下次要调整什么。” |\n| 做每周内容包 | “用灵造把我这一周的素材整理成 5 个母题，并分发成小红书、公众号、播客和短口播。” |\n| 校准公众号对标 | “我发几篇喜欢的公众号文章和一篇自己的内容，你先判断适不适合我学，不适合再补找对标，然后帮我写成自己的文章。” |\n| 做图片素材 | “先帮我设计封面/配图方向；如果需要生成图片，再按我确认的方向生成。” |\n| 保存长结果 | “把这份分析整理成 Word、网页预览或知识库 Markdown 版本。” |\n\n## 免费能做什么\n\n不配置 API Key 时，灵造仍然可以作为创作者运营路由和 playbook 使用。适合：\n\n- 判断账号定位、赛道难度、内容主线和商业路径。\n- 设计小红书标题、封面方向、发布关键词和图文结构。\n- 改写草稿、拆解用户已经提供的内容材料、做发布前检查。\n- 根据用户提供的数据截图或复盘信息，输出下一步实验建议。\n- 把用户提供的一周素材整理成 5 个母题，并规划小红书、公众号、播客、\n 短口播、社群和知识库分发。\n- 把长分析整理成 Word、网页预览或知识库 Markdown 结构。\n\n## 什么时候需要 API Key\n\n当 Agent 需要让灵造服务实际查询或生成内容时，需要到\n 配置 API Key，包括：\n\n- 搜索小红书、抖音、TikTok、Instagram、YouTube 或视频号公开内容和公开创作者；用结果辅助关键词/选题扩展。\n- 查看创作者主页、近期公开内容、主页深度分析和对标账号证据。\n- 打开小红书、抖音、TikTok、Instagram、YouTube 或视频号单条公开内容详情，读取一级公开评论。\n- 打开公众号公开文章详情，查看公开文章数据，扩展相关文章。\n- 提取公开短视频口播文案、字幕或 transcript。\n- 根据提示词和参考图生成创作者封面、配图或海报素材。\n\n用户已给出明确任务且属于小范围时，直接执行；首轮最多做 5 次外部查询。\n如果需要扩大关键词、账号、内容详情、评论分页、文案提取或生图数量，说明新增的业务范围并请用户确认。\n\n## 调用公开数据工具前\n\n- 小红书、抖音、TikTok、Instagram 或 YouTube 内容链接：看内容用详情工具，看评论用评论工具，不要当主页链接。视频号 `/sph/` 分享链接可用于详情；评论必须先从详情取得纯数字内容 ID。\n- 小红书、抖音、TikTok、Instagram 或 YouTube 主页链接：普通主页查看或基础主页分析先用\n `get-user-posted-notes`；只有用户明确要粉丝数、简介、关注数、总获赞等主页资料时\n 才用 `get-user-info`；深度主页分析看 `analyze-user-profile`。\n- 用户只给昵称、账号名、抖音号或数字 ID 时，不要自己拼 URL；先用\n `search-users` 找创作者，再用返回的主页链接或 ID 调主页工具。\n- 视频号主页只接受 `search-users` 返回的 finder ID，不接受主页 URL，也不支持 `analyze-user-profile`。\n- 抖音主页工具需要可用的主页 URL 或 `search-users` 返回的 `MS4w...` 形式 ID；\n 视频短链看单条内容时直接用 `get-note-detail`，需要口播、字幕或 transcript 时才用\n `extract-video-copy`；不要把视频短链用于主页分析。\n- YouTube 主页工具只接受 `search-users` 返回的 channel ID 或 `/channel/UC...`\n URL；不要把 `@handle`、`/c/` 或 `/user/` 直接传给主页工具，也不要自动解析。\n- TikTok 主页工具接受 canonical `https://www.tiktok.com/@handle` 或\n `search-users` 返回的 ID；单条内容接受 canonical `/@handle/video/`、\n `/@handle/photo/` 或显式 `--platform tiktok --note-id `。不要传\n `vm.tiktok.com`/`vt.tiktok.com` 短链或裸 `@handle`。\n- TikTok V1 不支持 `analyze-user-profile`。需要主页资料和近期内容时，按需分别调用\n `get-user-info` 与 `get-user-posted-notes`，不要隐藏组合调用。\n- Instagram 主页工具接受 canonical `https://www.instagram.com//` 或\n `search-users` 返回的十进制字符串 ID；内容工具接受 canonical `/p/`、\n `/reel/`、`/reels/`、`/tv/`。评论命令的裸 `--note-id` 是\n shortcode，不是十进制 media ID。Instagram V1 不支持 `analyze-user-profile`，\n 不要把主页资料与近期内容隐藏组合调用。\n- 视频号创作者先用 `search-users --platform wechat_channels`，主页工具只复用返回的\n `v2_...@finder` ID。视频搜索返回的 `export/...` ID 可立即用于详情；评论必须使用\n 详情返回的纯数字内容 ID。视频号口播文案提取直接使用公开\n `https://weixin.qq.com/sph/...` 分享链接，不需要先调用详情。视频号不支持深度主页分析、\n 下载或解密。\n- 如果 API 返回 `agent_action`、`suggested_capabilities` 或 `expected_input`，\n 先按这些字段改调工具；仍不确定时问用户要主页链接或笔记/视频链接。\n\n## 常见问题\n\n**我没有 API Key，还能用吗？**\n可以。先用灵造做选题判断、标题封面、账号诊断、草稿修改、发布检查和复盘。\n等需要查公开内容、评论、短视频文案、公众号文章数据或生成图片时，再配置 API Key。\n\n**为什么 SkillHub 里显示需要 API Key？**\n因为灵造包含在线公开内容查询和图片生成能力。安装主 Skill 免费，但深度查询和\n生成动作需要已开通的在线服务和 API Key。\n\n**WorkBuddy 用户应该怎么用？**\n优先安装这一个 `lingzao` 主 Skill。装好后直接把任务说给 WorkBuddy，例如\n“帮我找对标账号”“帮我拆这条笔记”“帮我做发布前检查”。需要查公开数据时，\n再按灵造网页教程配置 API Key。\n\n**灵造能保证爆款、涨粉或变现吗？**\n不能。灵造只做公开内容研究、运营判断和工作流辅助。输出用于帮助你做判断和\n复盘，不是保证结果，也不能用于复制他人内容。\n\n**网络或服务失败怎么办？**\n先保留当前问题和链接，不要重复扩大查询范围。检查 `doctor`、API Key 和\n网络状态；图片生成或短视频文案提取这类异步任务可能需要等待轮询完成。\n如果灵造返回服务暂时不可用或响应超时，只用固定话术告诉用户：“灵造服务暂时\n不可用，请稍后重试。”如果返回了 `error_id`，可以附上 `error_id`，方便后续排查。\n如果 CLI 返回了明确的用户可见原因和下一步，保留该指引，不要改写成其他故障，也不要自动重试。\n\n## Agent Playbooks\n\nFor higher-level creator strategy tasks, use the playbooks in\n`/playbooks/` before answering. They turn Lingzao's public-content\ntools into creator workflows instead of isolated lookups.\n\n## Playbook Routing Contract\n\nBefore reading a playbook, read `/playbooks/router-index.json`. Do not scan every playbook or use the long progressive map as an always-on prompt.\n\nRoute in this order:\n\n1. Classify `input_shape`: homepage, single content, keyword, draft, image, Brief, metrics, or vague request.\n2. Classify `platform`: Xiaohongshu, Douyin, TikTok, Instagram, YouTube, WeChat, cross-platform, or unknown.\n3. Classify `content_stage`: no content, in progress, finished before publishing, published, or recurring system.\n4. Classify `intent`: direction, benchmark, diagnosis, production, visual, publish check, review, distribution, or knowledge base.\n5. Classify `requested_output`: chat judgment, report, publishable copy, image brief, saved files, or reusable library.\n\nThen select:\n\n- exactly 1 primary playbook when a workflow is needed\n- no more than 2 gate/support playbooks\n- 0 primary playbooks when a direct CLI command or simple answer is sufficient\n\nIf confidence is high, load only the selected files and proceed. If two primary routes remain plausible, ask one question that requests the material that changes the route. Do not show users the internal playbook list.\n\nEach entry in `router-index.json` is the centralized route card for one playbook:\n\n- `role`: router, primary, gate, or support\n- `category` and `platforms`\n- user-like `signals`\n- `required_inputs`\n- expected `outputs`\n- `avoid_when`\n- allowed `companions`\n\nUse the specialized routers only when needed:\n\n- vague or link-only input -> `progressive-interaction`\n- unclear Xiaohongshu operation stage -> `xhs-operation-tree`\n- online lookup with expandable scope -> add `research-scope-guard`\n- final Xiaohongshu-facing content -> add the relevant management/compliance gate\n- formal evidence-backed report -> add `report-evidence-contract`\n\nNever select a gate or support playbook as the main workflow. Never load more files merely because they are related.\n\nThe registry is complete only when this command passes:\n\n```bash\npython3 /scripts/check_playbook_router.py\n```\n\n`router-cases.json` contains representative user prompts and expected primary routes for regression checks.\n\nKeep public wording focused on creator-content research and workflow support.\nDo not promise viral growth, guaranteed monetization, full monitoring, bulk data\nexport, or copying another creator's content.\n\n## Before Returning Xiaohongshu Copy\n\nBefore returning any final Xiaohongshu-facing title, cover copy, page text,\nbody/caption, publishing keywords, pinned comment, comment guidance, spoken\nscript, Vlog storyboard, Brand Brief deliverable, one-stop package, or\nXiaohongshu section of a cross-platform package, run\n`playbooks/xhs-platform-management-risk-baseline.md` first, then\n`playbooks/xhs-content-compliance-risk-gate.md`.\n\nIf the draft contains off-platform diversion, WeChat/private-contact guidance,\nincentivized comment interaction, exaggerated guarantees, or sensitive\nunsupported claims, do not leave those lines in the publishable version. Show a\nshort risk note and rewrite them into a safer Xiaohongshu version. Never\npromise platform approval; say the rewrite lowers risk.\n\nFor commercial or product-related Xiaohongshu outputs, keep the order:\n\n1. public value first\n2. product or brand name after the reader benefit is clear\n3. no off-platform diversion action in the publishable Xiaohongshu copy\n\n## Install And Online Capability Entry\n\nLingzao is installed as one free main Skill. Users do not need to install\nseparate title, keyword, account-diagnosis, benchmark, cover, or review skills.\nAfter installation, this main Skill routes the user's request to the right\nplaybook.\n\nThere are two user acquisition paths:\n\n1. Community/course users:\n - They may already have A Tian's course, install link, setup steps, and\n API Key setup instructions.\n - Keep the in-chat explanation short: install the Skill, open the Lingzao web\n dashboard, follow the tutorial, enable online access, copy the API Key, then\n run setup.\n\n2. Public-platform users from Xiaohongshu, Douyin, or other public content:\n - Do not require them to open the web dashboard and pay before they\n understand what Lingzao can do.\n - Let them install the free main Skill first.\n - Then explain the online entry in friendly language: the local\n playbooks can help judge drafts, titles, covers, directions, and\n publishing plans; when they need Lingzao to search public content, inspect\n accounts, open note/article details, read comments, inspect article data,\n extract video copy, or generate creator image assets, they need to open\n the Lingzao web dashboard, follow the tutorial, enable online access, and\n configure an API Key.\n\nPresent the web dashboard as the user's learning and setup hub:\n\n- learn how to install and configure Lingzao\n- learn how to ask Agent better questions instead of waiting in a group chat\n- learn how to use Skill workflows for self-media operation\n- learn account diagnosis, benchmark breakdown, title/keyword, pre-publish, and\n post-publish review workflows\n- enable online access and get the API Key when they need public-content lookup or\n image generation\n\nUse this wording when a user has installed the Skill but has not configured an\nAPI Key yet:\n\n你已经装好灵造 Skill 了。安装本身是免费的，它会先帮你判断你现在是在找方向、拆账号、写内容、做封面、配关键词，还是复盘数据。\n如果你要继续查小红书、抖音、TikTok、Instagram、YouTube 或公众号公开内容、找对标账号、看账号主页、打开内容或文章详情、看评论区、查看公众号文章数据、提取短视频文案或生成创作者图片素材，就需要到灵造网页版开通在线服务并配置 API Key。\n你可以打开 https://lingzao.atian.vip 看安装教程和使用教程，里面也会教你怎么用 Agent 做自媒体运营、怎么问问题、怎么用这些 Skill。需要查公开内容或生成图片的时候，再在网页里获取 API Key，配置好以后回来继续问，我会接着刚才的问题往下做。\n\nFrame the two capability layers as:\n\n- free install = get the workflow brain and routing layer\n- web dashboard = tutorial, usage examples, self-media operation lessons, and\n API Key setup\n- online access = unlock public-content lookup, image generation, and deeper\n research actions\n\nKnowledge sync handoff:\n\n- After a useful Lingzao research result or diagnosis report, do not sync it\n automatically. Ask first: 要不要把这份结果同步到你的知识库？可以选择\n ima / Obsidian / 飞书 / 暂不同步。\n- If the user chooses a target, prepare a clean Markdown version and ask the\n current Agent environment to use the user's configured knowledge tool.\n- For ima, call the installed ima Skill or ima knowledge-base tool if the user\n has configured one.\n- For Obsidian, use the user's Obsidian CLI, Obsidian Skill, or approved vault\n workflow to write Markdown under a user-approved `Lingzao/` path.\n- For 飞书, use the user's Lark/Feishu CLI or Skill with user authorization to\n create or update a document.\n- Do not ask for or store ima, Obsidian, or Feishu credentials inside Lingzao.\n Synchronized content should contain only the user-approved report, public\n links, and useful conclusions; leave out credentials and details the user does\n not need.\n\nProfile workflow:\n\n- If the user asks for a creator homepage or a basic homepage analysis, use `get-user-posted-notes` by default. It returns recent posts and enough author/post data for a basic read.\n- If the user sends a Xiaohongshu short link such as `xhslink.com/m/...`, or a\n copied share sentence such as `@... 查看Ta的主页>> https://xhslink.com/m/...`,\n extract the short link, normalize bare links to `https://...`, and read the\n surrounding words before choosing a command. Do not classify the short link by\n path alone. If the context says account, homepage, creator, profile,\n benchmark, account diagnosis, homepage diagnosis, `Ta的主页`, or recent posts,\n treat it as a creator-homepage request and call\n `get-user-posted-notes --url \"https://\"`.\n- If a Xiaohongshu short link has no context, ask whether the user wants creator\n homepage recent posts or one-post detail before making an online request. If the\n context says this note, comments, copy, transcript, one-post breakdown, or is\n a normal note share sentence with a title snippet plus `前往【小红书】一探究竟吧`,\n treat it as a one-post candidate, not a homepage. One-post words such as\n `这条` or `这篇` take priority over generic diagnosis wording. Do not default\n to `get-note-detail`; first confirm it is a single post and ask for the final\n note URL or note_id plus whether it is 图文 or 视频 when needed.\n- Only add `get-user-info` when the user specifically needs full profile-level stats such as bio, follower count, following count, total likes, total collections, or total note count.\n- Use `analyze-user-profile` for Xiaohongshu deeper homepage copy/script/subtitle analysis, recent post text, covers, commercial signals, or product-note signals. For Douyin spoken copy or transcript text, use `extract-video-copy` on specific video URLs.\n- YouTube V1 does not support `analyze-user-profile`. Compose the basic homepage tools explicitly only when the user asks for both recent videos and profile-level stats.\n- Do not call `get-user-info` and `get-user-posted-notes` as a fixed pair unless the user asks for both profile-level stats and recent-post analysis.\n- Do not force a full account diagnosis when the homepage has too few public\n posts. Route by visible sample size:\n - 0 posts: no account diagnosis; switch to beginner start/account setup\n guidance.\n - 1-2 posts: homepage first impression plus single-post feedback only.\n - 3-5 posts: starter-account mini diagnosis.\n - 6-9 posts: light account analysis.\n - 10+ posts: standard account analysis can be offered.\n - 20+ posts: standard deep diagnosis can use `analyze-user-profile --limit 20`\n after confirming that deeper scope.\n - 40+ posts: deep diagnosis, creator distillation, or knowledge-base\n distillation can use `--limit 40` after confirming that deeper scope.\n\nPost drill-down workflow:\n\n- Xiaohongshu list-style commands (`search-notes`, `get-user-posted-notes`,\n `analyze-user-profile`) return `xhs_note_type` on each note item when\n Lingzao can identify whether it is 图文 or 视频.\n- When continuing from one of those note items to `get-note-detail`, pass the\n returned `xhs_note_type` directly as `--xhs-note-type`; do not infer the type\n from the URL.\n- If a Xiaohongshu note item has no `xhs_note_type`, ask the user whether it is\n 图文 or 视频 before calling `get-note-detail`. `get-note-comments` can still\n be called without this type.\n- If `get-note-detail` returns `NOTE_NOT_FOUND_OR_INACCESSIBLE`, do not retry\n the same request or probe the other Xiaohongshu type automatically. Go back to\n the source list/homepage result and reuse its `xhs_note_type`, or ask the user\n for the correct type or a public URL.\n\n## Setup\n\nResolve this `SKILL.md` directory as ``, then run setup once:\n\n```bash\nbash \"/scripts/setup.sh\" --base-url \"https://your-lingzao-domain.com\"\n```\n\nEnvironment variables override saved config:\n\n```bash\nexport LINGZAO_API_KEY=\"lgz_xxx\"\nexport LINGZAO_BASE_URL=\"https://your-lingzao-domain.com\"\n```\n\nCheck the connection:\n\n```bash\n~/.lingzao/bin/lingzao doctor\n```\n\nBefore using Lingzao commands, check whether the skill has an update:\n\n```bash\n~/.lingzao/bin/lingzao check-version\n```\n\nIf an update is available, stop the current Lingzao operation and update the skill first. Do not continue using an outdated Lingzao Skill for search, profile, subtitle, or extraction work.\n\nTo update the skill, rerun the installer. For `npx skills`, try:\n\n```bash\nnpx skills add https://assets-tian.midao.site/skills/lingzao --skill lingzao -g --copy\n```\n\nUpdating keeps the saved API config in `~/.lingzao/config.json`; no API key setup is needed again.\n\nIf `~/.lingzao/bin/lingzao` is missing or points to the wrong directory, repair the command wrapper:\n\n```bash\nbash ~/.agents/skills/lingzao/scripts/setup.sh --skip-doctor\n```\n\nIf `~/.agents/skills/lingzao` does not exist, find the directory that contains `lingzao`'s `SKILL.md`, then run `scripts/setup.sh --skip-doctor` from that directory.\n\n## Before Calling\n\nBefore running a command with meaningful filters, ask the user for the relevant\nparameters if they did not already specify them.\n\n- Track external Lingzao commands internally for the current user request. Keep\n the first pass to at most 5 lookups or one confirmed image batch. If more\n keywords, accounts, details, comment pages, transcripts, profile depth, or\n images are needed, show the exact added actions and wait for scope confirmation.\n- For broad creator or benchmark-account searches (`search-users`, \"找对标账号\",\n \"找参考博主\", \"找同赛道账号\"), do not start with a wide search. First ask or\n state a narrow starter scope: follower range, track/topic, account format,\n city/local scope when relevant, recent-update requirement, recent-hit\n requirement, and starter result count. Recommend starting with 3 accounts,\n then expanding only after the user confirms the direction. This keeps the\n research focused and avoids returning 100-follower seed accounts or huge mature\n accounts when the user asked for a specific stage.\n- If the user asks how to write prompts for Lingzao or gives a broad copy-paste\n request, use `copy-paste-prompt-scope-boundary.md` first. Provide a\n ready-to-copy prompt that includes the smallest useful scope instead of\n telling the user to add broad instructions by themselves.\n- If the user says they know nothing about self-media, are starting from zero,\n do not know what to post, or only say they want to make money, use\n `zero-beginner-onboarding-gate.md` before any search. Do not call online\n lookup first. Start with a free life-signal intake, give the lowest creator\n cognition, and move them to one concrete first task.\n- For `search-notes`, ask for sorting, note type, and time range before calling:\n sort can be `general`, `most_liked`, `popularity_descending`,\n `comment_descending`, or `collect_descending`; note type can be `不限`,\n `视频笔记`, `图文笔记`, or `直播笔记`; time range can be `不限`, `一天内`,\n `一周内`, or `半年内`.\n- Douyin and TikTok `search-notes` currently support only `general`, `most_liked`, and\n `popularity_descending`. Do not pass `comment_descending` or\n `collect_descending` for Douyin or TikTok searches.\n- Douyin and TikTok `search-notes` note type currently supports only `不限`, `视频笔记`,\n and `图文笔记`. Do not pass `直播笔记` for Douyin or TikTok searches.\n- YouTube `search-notes` supports only `--sort general`, `--note-type 不限|视频笔记`,\n and `--time-filter 不限|一天内|一周内`; use the returned opaque `next_cursor`\n with `--cursor`, repeat the same keyword and filters, and do not infer internal\n pagination fields. Changing a filter invalidates the cursor before another lookup.\n- For `get-note-comments`, ask whether the user wants latest comments or\n liked-count sorting before calling Xiaohongshu. Use `--sort latest` for latest\n comments and `--sort most_liked` for Xiaohongshu liked-count sorting.\n- Douyin, TikTok, and Instagram comments currently support only `latest`;\n TikTok uses the service default order. Do not ask for or pass\n `--sort most_liked` on these platforms.\n- YouTube comments support `latest` and `most_liked`; only top-level comments\n are returned. Reuse `next_cursor` unchanged and repeat the same `--sort` on\n every next-page request; omitting it after `most_liked` defaults to `latest`\n and invalidates the cursor before another lookup.\n- Instagram content search is Reels-only. Use `--sort general`,\n `--note-type 视频笔记`, and `--time-filter 不限`; `--note-type 不限` remains a\n compatibility input but is executed and reported as 视频笔记. Do not use\n `search-notes` for account records; use `search-users`.\n- Instagram profile, posted-note, and detail results may include public avatar,\n cover, carousel-image, and video URLs from the current response. Current\n `search-notes` returns Reels identity, canonical URL, author identity, and\n author avatar only, so do not expect it to supplement text, metrics, or\n content media. These URLs can expire; use or save needed public\n references promptly and do not treat them as permanent asset storage.\n- For TikTok and Instagram `search-notes`, `search-users`,\n `get-user-posted-notes`, and\n `get-note-comments`, pass the returned `data.page.next_cursor` unchanged with\n `--cursor` to fetch one next page. Repeat the original search keyword and\n filters, creator, or content item for that cursor; never reuse it for another\n request identity. Never parse the opaque cursor or hide multi-page fanout.\n TikTok cursors created before Skill `0.1.92` and Instagram search cursors\n created before Skill `0.1.95` are invalid: discard them and restart from the first\n page. If Lingzao returns `PAGINATION_CURSOR_STALE`, also discard\n that cursor and restart from the first page; do not loop it.\n- Xiaohongshu list-style commands (`search-notes`, `get-user-posted-notes`,\n `analyze-user-profile`) return `xhs_note_type` on each note item when\n Lingzao can identify whether it is 图文 or 视频. When continuing from one of\n those note items to `get-note-detail`, pass the returned value directly as\n `--xhs-note-type`; do not infer the type from the URL. If a Xiaohongshu note\n item has no `xhs_note_type`, ask the user whether it is 图文 or 视频 before\n calling `get-note-detail`. If `get-note-detail` returns\n `NOTE_NOT_FOUND_OR_INACCESSIBLE`, do not retry the same request or probe the\n other Xiaohongshu type automatically. `get-note-comments` can still be called\n without this type.\n- If the user explicitly says to use defaults, proceed with the documented\n defaults instead of asking again.\n\nAfter a successful research command, tell the user the estimated time saved\nshown in the CLI Markdown output. If you called multiple Lingzao research\ncommands for one user request, summarize the total once. Do not show time-saved\nlanguage for `doctor`, `check-version`, failed commands, or JSON-only automation\nflows.\n\n## Commands\n\n### Search Notes\n\n```bash\n~/.lingzao/bin/lingzao search-notes --platform xhs --keyword \"AI写作\"\n~/.lingzao/bin/lingzao search-notes --platform xhs --keyword \"AI写作\" --sort most_liked\n~/.lingzao/bin/lingzao search-notes --platform xhs --keyword \"AI生图\" --sort collect_descending --note-type \"视频笔记\" --time-filter \"一周内\"\n~/.lingzao/bin/lingzao search-notes --platform douyin --keyword \"AI生图\" --sort most_liked --note-type \"视频笔记\"\n~/.lingzao/bin/lingzao search-notes --platform youtube --keyword \"creator workflow\" --sort general --note-type \"视频笔记\" --time-filter \"一周内\"\n~/.lingzao/bin/lingzao search-notes --platform tiktok --keyword \"AI gadgets\" --sort most_liked --note-type \"视频笔记\"\n~/.lingzao/bin/lingzao search-notes --platform tiktok --keyword \"AI gadgets\" --sort most_liked --note-type \"视频笔记\" --cursor \"next_cursor_from_previous_response\"\n~/.lingzao/bin/lingzao search-notes --platform instagram --keyword \"creative coding\" --sort general --note-type \"视频笔记\" --time-filter \"不限\"\n~/.lingzao/bin/lingzao search-notes --platform wechat_channels --keyword \"人工智能\" --sort general --note-type \"视频笔记\" --time-filter \"一周内\"\n```\n\nUse this when the user wants public notes around a topic.\nBefore calling, ask the user for `--sort`, `--note-type`, and `--time-filter`\nwhen they have not specified those preferences.\nInstagram content search always means Reels search; choose `--note-type 视频笔记`.\nFor TikTok pagination, repeat the same keyword, sort, note type, and time filter\nwith the returned cursor.\n`search-suggestions` has been retired. For keyword expansion or topic discovery,\nuse `search-notes` for content ideas or `search-users` for creator discovery.\n\n### Search Creators\n\n```bash\n~/.lingzao/bin/lingzao search-users --platform xhs --keyword \"母婴博主\"\n~/.lingzao/bin/lingzao search-users --platform douyin --keyword \"AI生图\"\n~/.lingzao/bin/lingzao search-users --platform youtube --keyword \"creator workflow\"\n~/.lingzao/bin/lingzao search-users --platform tiktok --keyword \"tech.bytes\"\n~/.lingzao/bin/lingzao search-users --platform instagram --keyword \"creative coding\"\n~/.lingzao/bin/lingzao search-users --platform wechat_channels --keyword \"央视新闻\"\n```\n\nUse this when the user wants creators in a topic or niche.\nFor TikTok pagination, repeat the same keyword with the returned cursor.\nWhen continuing from `search-users` to profile verification, pass the returned\n`users[].id` with `--platform xhs --user-id ...`, `--platform douyin --user-id ...`,\n`--platform tiktok --user-id ...`, or `--platform instagram --user-id ...`.\nFor WeChat Channels, reuse the exact finder ID with\n`--platform wechat_channels --user-id \"v2_...@finder\"`.\nFor YouTube, the returned ID is a\ncanonical channel ID; reuse it with `--platform youtube --user-id ...` and\ntreat `handle` as display metadata only.\nThe output may include RED ID and follower count for screening, but RED ID is\ndisplay metadata only. Do not extract Xiaohongshu RED ID values from bios or\nbuild `/user/profile/` URLs.\n\n### Get Creator Profile\n\n```bash\n~/.lingzao/bin/lingzao get-user-info --url \"https://www.xiaohongshu.com/user/profile/...\"\n~/.lingzao/bin/lingzao get-user-info --platform xhs --user-id \"63c21e0f000000002801a1bb\"\n~/.lingzao/bin/lingzao get-user-info --platform douyin --user-id \"MS4wLjABAAAA...\"\n~/.lingzao/bin/lingzao get-user-info --platform youtube --user-id \"UC...\"\n~/.lingzao/bin/lingzao get-user-info --url \"https://www.tiktok.com/@creator\"\n~/.lingzao/bin/lingzao get-user-info --url \"https://www.instagram.com/creator/\"\n~/.lingzao/bin/lingzao get-user-info --platform wechat_channels --user-id \"v2_...@finder\"\n```\n\nUse this when the user provides a creator profile URL or platform user ID and needs full profile-level stats. For Douyin bare user IDs, use the profile `sec_user_id`. For YouTube, use a channel ID or `/channel/UC...` URL; if the user only has a handle, call `search-users` first. For basic homepage analysis, prefer `get-user-posted-notes` and avoid calling both commands by default.\n\n### Get Creator Recent Posts\n\n```bash\n~/.lingzao/bin/lingzao get-user-posted-notes --url \"https://www.xiaohongshu.com/user/profile/...\"\n~/.lingzao/bin/lingzao get-user-posted-notes --platform xhs --user-id \"63c21e0f000000002801a1bb\"\n~/.lingzao/bin/lingzao get-user-posted-notes --platform douyin --user-id \"MS4wLjABAAAA...\" --limit 20\n~/.lingzao/bin/lingzao get-user-posted-notes --platform youtube --user-id \"UC...\" --limit 20\n~/.lingzao/bin/lingzao get-user-posted-notes --platform tiktok --user-id \"\" --limit 20\n~/.lingzao/bin/lingzao get-user-posted-notes --platform tiktok --user-id \"\" --cursor \"next_cursor_from_previous_response\"\n~/.lingzao/bin/lingzao get-user-posted-notes --platform instagram --user-id \"\" --limit 20\n~/.lingzao/bin/lingzao get-user-posted-notes --platform instagram --user-id \"\" --cursor \"next_cursor_from_previous_response\"\n~/.lingzao/bin/lingzao get-user-posted-notes --platform wechat_channels --user-id \"v2_...@finder\" --limit 20\n```\n\nUse this when the user wants to understand what a creator has posted recently. Use this by default for basic creator homepage analysis. Douyin, TikTok, Instagram, YouTube, and WeChat Channels support `--limit 20` at most per public call. WeChat Channels requires the finder ID returned by `search-users`; profile URLs are not accepted. YouTube reads the Videos list only and does not add a separate Shorts request. If the response has `next_cursor`, reuse it with `--cursor`; for TikTok, Instagram, or WeChat Channels, repeat the same creator ID. If the user asks for full profile-level stats, add `get-user-info`; if the user asks for Xiaohongshu post copy, scripts, captions, or transcript text across recent posts, use `analyze-user-profile` instead. For Douyin transcript text, use `extract-video-copy` on selected video URLs. TikTok, Instagram, YouTube, and WeChat Channels V1 do not support `analyze-user-profile`.\n\n### Analyze Creator Profile\n\n```bash\n~/.lingzao/bin/lingzao analyze-user-profile --url \"https://www.xiaohongshu.com/user/profile/...\" --limit 20\n~/.lingzao/bin/lingzao analyze-user-profile --platform xhs --user-id \"63c21e0f000000002801a1bb\" --limit 40\n~/.lingzao/bin/lingzao analyze-user-profile --platform douyin --user-id \"MS4wLjABAAAA...\" --limit 20\n```\n\nUse this when the user wants deeper creator profile data, including post text, covers, commercial signals, and profile-level content signals. For Xiaohongshu, it also includes subtitle/script previews. For Douyin, it does not extract homepage subtitles or transcript text; use `extract-video-copy` on selected video URLs when the user needs spoken copy.\nUse `--limit 20` by default. The default Markdown output shows readable subtitle previews when the platform provides them.\nShort-window repeats with the same request parameters may reuse the recent successful result. Use `--force-new` only when the user explicitly needs a fresh run, and do not loop it: repeated forced refreshes in the short protection window may be rejected.\nIf Douyin profile insight sections are temporarily unavailable, the API and CLI can show `partial_data`, `warnings`, or `unavailable_sections`. Explain that homepage works data still returned successfully, and do not treat the missing insight section as proof that there is no data.\n\nImportant for Xiaohongshu: the complete profile subtitle/copy Markdown artifact is a top-level response field, not a per-note subtitle URL. Always check:\n\n`data.artifacts.subtitle_markdown.status`\n`data.artifacts.subtitle_markdown.url`\n\nDo not search only inside `items[]`. If `data.artifacts.subtitle_markdown.status == \"ready\"` and `url` exists, download it before deep script or subtitle analysis:\n\n```bash\ncurl -L \"$subtitle_markdown_url\" -o /tmp/lingzao-profile-subtitles.md\n```\n\nUse the downloaded Markdown file for complete subtitle/copy analysis. Use `--format json` when the user needs the structured fields. JSON includes `data.artifacts.subtitle_markdown.url` for the complete Markdown file when available, and inline `items[].text.subtitle.content/plain_text` are preview-sized to keep the response readable. If the artifact is unavailable, use the inline subtitle fields. For Douyin, expect `data.artifacts.subtitle_markdown.status == \"unsupported\"` and use the returned profile insights plus selected-video extraction instead.\n\n### Get Post Detail\n\n```bash\n~/.lingzao/bin/lingzao get-note-detail --url \"https://www.xiaohongshu.com/explore/...\" --xhs-note-type image\n~/.lingzao/bin/lingzao get-note-detail --platform xhs --note-id \"69690331000000001a02266a\" --xhs-note-type video\n~/.lingzao/bin/lingzao get-note-detail --platform douyin --note-id \"7372484715782352169\"\n~/.lingzao/bin/lingzao get-note-detail --url \"https://v.douyin.com/\"\n~/.lingzao/bin/lingzao get-note-detail --url \"https://www.youtube.com/watch?v=...\" --content-type video\n~/.lingzao/bin/lingzao get-note-detail --platform youtube --note-id \"...\" --content-type short\n~/.lingzao/bin/lingzao get-note-detail --url \"https://www.youtube.com/shorts/...\"\n~/.lingzao/bin/lingzao get-note-detail --url \"https://www.tiktok.com/@creator/video/7349541381817355521\"\n~/.lingzao/bin/lingzao get-note-detail --url \"https://www.instagram.com/reel//\"\n~/.lingzao/bin/lingzao get-note-detail --platform instagram --note-id \"\"\n~/.lingzao/bin/lingzao get-note-detail --platform wechat_channels --note-id \"export/...\"\n~/.lingzao/bin/lingzao get-note-detail --url \"https://weixin.qq.com/sph/...\"\n```\n\nThe `/shorts/` URL form preserves Short type automatically. For a bare ID,\n`watch?v=` URL, or `youtu.be/` URL, pass the `content_type` returned by search as\n`--content-type video|short`; Lingzao does not guess type from duration.\nYouTube channel/profile URLs are not content-detail inputs. Use\n`get-user-info` or `get-user-posted-notes`; for `@handle`, `/c/`, or `/user/`\nURLs, use `search-users` first to obtain the canonical channel ID.\n\nUse this when the user asks to analyze one public post.\nFor Douyin, pass a standard post URL, numeric `aweme_id`, or a valid HTTPS\n`v.douyin.com/` URL directly. Do not open or expand the short link\nfirst. If Lingzao reports that the short-link format is invalid, ask for the\noriginal HTTPS share URL instead of retrying another detail form automatically.\nFor Xiaohongshu details, pass `--xhs-note-type image` for 图文 and\n`--xhs-note-type video` for 视频. If the note came from `search-notes`,\n`get-user-posted-notes`, or `analyze-user-profile`, reuse that item's\n`xhs_note_type` value. If detail returns `NOTE_NOT_FOUND_OR_INACCESSIBLE`,\ndo not switch `--xhs-note-type` and retry automatically; confirm the source\nitem type or ask the user.\nFor a Xiaohongshu image note, default Markdown shows `正文图片（N 张）` followed\nby every ordered body-image link. Use `--format json` for structured\n`data.item.media.images`. Public image links may expire; do not describe them\nas downloaded, proxied, or permanently stored by Lingzao.\n\n### Get Post Comments\n\n```bash\n~/.lingzao/bin/lingzao get-note-comments --url \"https://www.xiaohongshu.com/explore/...\"\n~/.lingzao/bin/lingzao get-note-comments --url \"https://www.xiaohongshu.com/explore/...\" --sort most_liked\n~/.lingzao/bin/lingzao get-note-comments --platform xhs --note-id \"69690331000000001a02266a\"\n~/.lingzao/bin/lingzao get-note-comments --platform douyin --note-id \"7372484715782352169\"\n~/.lingzao/bin/lingzao get-note-comments --platform tiktok --note-id \"7349541381817355521\" --limit 20\n~/.lingzao/bin/lingzao get-note-comments --url \"https://www.instagram.com/p//\" --limit 20\n~/.lingzao/bin/lingzao get-note-comments --platform instagram --note-id \"\" --cursor \"next_cursor_from_previous_response\"\n~/.lingzao/bin/lingzao get-note-comments --url \"https://www.douyin.com/jingxuan?modal_id=...\" --cursor \"next_cursor_from_previous_response\"\n~/.lingzao/bin/lingzao get-note-comments --url \"https://youtu.be/...\" --sort most_liked --limit 20\n~/.lingzao/bin/lingzao get-note-comments --platform wechat_channels --note-id \"\" --sort latest --limit 20\n```\n\nUse this when the user asks for public comments on one post. The first version returns top-level comments only. Use `--sort most_liked` for Xiaohongshu or YouTube liked-count sorting; Douyin, TikTok, Instagram, and WeChat Channels support only `latest`, with TikTok using service-default order. If the response has `data.page.next_cursor`, pass that opaque value unchanged with `--cursor` to fetch one next page. For TikTok or Instagram, repeat the same content URL or ID; for WeChat Channels, repeat the same numeric object ID; for YouTube, repeat the same `--sort` with every cursor request.\nBefore calling Xiaohongshu comments, ask whether the user wants latest comments\nor liked-count sorting. For Douyin, TikTok, Instagram, and WeChat Channels comments, use only `--sort latest`;\ndo not pass `--sort most_liked`.\n\n### Get WeChat Official-Account Articles\n\n```bash\n~/.lingzao/bin/lingzao get-article-detail --url \"https://mp.weixin.qq.com/s/...\"\n~/.lingzao/bin/lingzao get-article-detail --url \"https://mp.weixin.qq.com/s/...\" --output /tmp/article.md\n~/.lingzao/bin/lingzao get-article-stats --url \"https://mp.weixin.qq.com/s/...\"\n~/.lingzao/bin/lingzao get-related-articles --url \"https://mp.weixin.qq.com/s/...\"\n```\n\nUse these when the user provides a public WeChat official-account article URL\nand asks to analyze the article, inspect public engagement metrics, or expand\nfrom that article to related public articles. The first version is URL-only.\nAn empty related-articles list is a valid response.\nDo not use these commands for account article history, account listing, or\nmulti-page fanout unless Lingzao adds a separate capability.\n\nFor full article analysis, prefer `get-article-detail --output /tmp/article.md`.\nThe command saves the complete article text as a local Markdown file and prints\nonly the file path plus a short summary in chat. Read the saved Markdown file\nfor detailed analysis instead of asking the CLI to paste the full article body\ninto the conversation.\n\n### Extract Short-Video Copy\n\n```bash\n~/.lingzao/bin/lingzao extract-video-copy --url \"https://www.xiaohongshu.com/explore/...\"\n~/.lingzao/bin/lingzao extract-video-copy --url \"https://v.douyin.com/...\"\n~/.lingzao/bin/lingzao extract-video-copy --url \"https://weixin.qq.com/sph/...\"\n~/.lingzao/bin/lingzao extract-video-copy --operation-id \"<上一次打印的 UUID>\" --url \"https://v.douyin.com/...\"\n```\n\nUse this when the user asks for short-video spoken copy, transcript, subtitles, or口播文案.\nXiaohongshu, Douyin, and WeChat Channels public video links are supported. For\nWeChat Channels, pass the canonical `https://weixin.qq.com/sph/...` share link\ndirectly; do not call detail first or attempt download/decryption.\nThe CLI prints a stable extraction request ID before submitting. If the response\nis ambiguous, interrupted, or not consumed, repeat the same command within 24\nhours with `--operation-id ` and keep every `--url` unchanged. Omit\n`--operation-id` for a new extraction intent. Never reuse an old operation ID\nwith different URLs and do not invent an automatic retry loop.\nIf one item reports that the video is too large, do not retry that URL. Explain\nwhich item failed, preserve any successful results in the same batch, keep the\nCLI-provided failure guidance, and ask for a shorter video link.\n\n### Generate Image\n\n```bash\n~/.lingzao/bin/lingzao generate-image --prompt \"一张小红书封面图，主题是 AI 生图新手避坑，干净明亮，中文大标题留白\" --output /tmp/lingzao-image.png\n~/.lingzao/bin/lingzao generate-image --prompt \"极简产品海报，白底，柔和阴影\" --size 1024x1536 --output /tmp/poster.png\n~/.lingzao/bin/lingzao generate-image --prompt \"参考两张图，保留人物风格，把产品界面换成灵造首页截图\" --size 1536x2048 --image /tmp/style.png --image /tmp/product.png --output /tmp/poster.png\n~/.lingzao/bin/lingzao generate-image --prompt \"每张参考封面分别改成 AI 工作台主题，替换原人物身份、原文字和品牌\" --count 3 --reference-mode one_to_one --image /tmp/top-1.png --image /tmp/top-2.png --image /tmp/top-3.png --size 1024x1536 --output /tmp/poster.png\n~/.lingzao/bin/lingzao generate-image --prompt \"批量生成 3 张封面草稿\" --count 3 --size 1024x1536 --output /tmp/poster.png\n~/.lingzao/bin/lingzao generate-image --prompt-file /tmp/lingzao-prompt.txt --output /tmp/poster.png\n```\n\nUse this only when the user asks to generate a creator image asset. For normal\nresearch, do not call image generation automatically.\n\nWhen the user wants N images from the same prompt, call `generate-image` once\nwith `--count N` for N=2..5. Do not loop the same prompt as multiple\n`--count 1` calls. The CLI prints a stable request ID before submitting that\nbatch. If a POST response is ambiguous or polling is interrupted, the Agent\nmust save that UUID and repeat the same command with\n`--client-request-id `; keep the prompt, size, count, output format,\nreference mode, and reference images unchanged. Omit `--client-request-id` for\nevery new generation intent. Do not reuse an old ID for new content and do not\ninvent another network-retry loop. The server retains idempotency and the\none-active-batch limit. If the user wants distinct concepts, vary the prompt\nfor each concept or use one counted batch for same-prompt variants.\nWhen each reference image should produce its own corresponding output, pass the\nreferences in output order, set `--count` to the same number, and add\n`--reference-mode one_to_one`. The CLI rejects mismatched counts before the API\nrequest. One-to-one batches support 1-4 reference images; `count=5` remains\navailable only for prompt-only or shared-reference generation. Without that\noption, repeated `--image` inputs are shared references that jointly influence\nevery output.\n\nBefore calling `generate-image`, run the minimal intake gate. If the user only\nsays something like \"给我做一张某某海报图\" or provides only a broad topic, do\nnot generate immediately. Ask for the two visual anchors first:\n\n1. 你有没有参考图？可以发 1-3 张你喜欢的封面/海报/图文截图。\n2. 你有没有想要的配色？比如明亮白底、绿色清爽、黑金高级、蓝色科技感。\n\nIf those are still unclear, ask at most one extra route-changing question, such\nas the publishing platform/size, exact on-image text, or whether the user wants\npeople/no people. Only proceed directly without asking when the user already\nprovided enough constraints: topic + platform/format + visual style/reference\nor color + on-image text/material.\nUse `--image` for local reference images; repeat it for multiple images. The\nSkill uploads those files directly to Lingzao for the current request, so the\nuser does not need to upload them elsewhere first. Supported reference image\nformats are png, jpeg, and webp.\nFor long, Chinese, or multiline prompts, prefer writing the prompt to a UTF-8\ntext file and passing `--prompt-file /path/to/prompt.txt`, or pipe the prompt\nwith `--prompt-stdin`, to avoid shell quoting or command-line encoding issues.\n\n#### Reference Image Handling\n\nFor Codex, WorkBuddy, and other agent runtimes:\n\n- `--image` accepts local filesystem paths only. If the user provides a\n reference image through a chat attachment, pasted image, screenshot, or input\n box, first materialize that image as a local file before calling the CLI.\n Preserve the original supported image format when saving the file.\n- Use a per-run temporary directory for runtime-provided images, for example\n `/tmp/lingzao-image-inputs//ref-1.png` and\n `/tmp/lingzao-image-inputs//ref-2.png`. Use absolute paths in the CLI\n call.\n- If the user already provided a stable local path, such as a file under\n `/Users/...`, you may pass that path directly. If the runtime-provided image\n lives in a temporary attachment path, copy it into the per-run temp directory\n first.\n- Do not proactively convert image formats. If the input image is already png,\n jpeg, or webp and its file size is reasonable, pass it as-is. Do not convert\n png to webp or jpeg just because an example path uses a different extension.\n- Only when a reference image is larger than 2 MB, create a smaller copy in the\n temp directory and pass that copy with `--image`. Keep the file extension and\n actual image bytes consistent. If resizing or compression fails, use the\n original supported image file instead of trying another format.\n- Do not overwrite the user's original image file. Do not store reference\n images in the repo. If the runtime cannot save an uploaded or pasted image to\n a local path, ask the user to save the image locally and provide the path.\n- In the prompt, state what should be borrowed from the reference images, such\n as layout, color palette, product shape, character style, or composition. Do\n not say only \"reference this image\" when a more specific instruction is\n possible.\n\nExample with a runtime-provided reference image:\n\n```bash\nmkdir -p /tmp/lingzao-image-inputs/run-001 /tmp/lingzao-image-outputs/run-001\n~/.lingzao/bin/lingzao generate-image \\\n --prompt \"参考这张图的排版和明亮色彩，生成一张小红书封面图，主题是 AI 生图新手避坑，中文大标题留白\" \\\n --size 1024x1024 \\\n --image /tmp/lingzao-image-inputs/run-001/ref-1.png \\\n --output /tmp/lingzao-image-outputs/run-001/result.png\n```\n\nThe command creates a Lingzao async batch and automatically polls the returned\nstatus URL until the background job finishes or the command timeout is reached.\nImage generation can take several minutes; `--timeout` can extend waiting for\nlarge or slow batches, but does not shorten the built-in per-image polling\nwindow. For one image, `--output` writes the result to the exact path you\nprovide. For `--count` greater than 1, `--output /tmp/poster.png` writes every\nsuccessful image as numbered files such as `/tmp/poster-1.png`,\n`/tmp/poster-2.png`, and so on. Default Markdown output requires `--output` so\ngenerated images are saved locally. If a direct API caller receives\n`GENERATION_IN_PROGRESS` with a returned `poll_url`, that active batch belongs\nto another intent: poll it only until the concurrency slot is free, then submit\nthe current request again with its original `client_request_id`. Do not return\nthe other batch as the current request's result. If no `poll_url` is returned,\nwait briefly and retry with the same ID. The CLI handles both cases\nautomatically. Use `--format json` only when you need structured automation\ndata.\n\n## Usage Notes\n\n- For profile and post URLs, pass the URL directly when possible.\n- For direct IDs, include `--platform`. For Xiaohongshu follow-up profile checks,\n prefer the 24-character `users[].id` returned by `search-users`; RED ID is\n display metadata only.\n- Omit `--limit` unless the user asks for a specific count.\n- Search notes default to comprehensive sorting, all note types, and all time; use `--sort`, `--note-type`, and `--time-filter` when the user asks for ranked or filtered note search.\n- Use `--format json` only when another tool needs structured output.\n- Default output is Markdown for agents to read and summarize.\n- If the API key or account needs attention, ask the user to open the Lingzao dashboard.\n\nFile v0.1.106:_meta.json\n\n{\n \"ownerId\": \"kn72byzxyctvxnvxm8a4d9qmvx8892za\",\n \"slug\": \"lingzao\",\n \"version\": \"0.1.106\",\n \"publishedAt\": 1788231353103\n}\n\nFile v0.1.106:index.md\n\n# Lingzao Skill Index\n\n## Purpose\n\nThis folder contains the Lingzao Skill package for Agent runtimes. Lingzao helps\nAgents route creator-operation work, prepare creator research, run public\ncontent lookups when the user has configured access, and generate creator image\nassets within a confirmed task scope.\n\n## Package Files\n\n- `SKILL.md`: main Agent instructions and command guidance.\n- `VERSION`: current Skill package version.\n- `agents/`: Agent metadata.\n- `assets/lingzao-logo.png`: packaged brand icon used by Agent metadata.\n- `playbooks/`: creator-operation workflows used before answering.\n- `playbooks/router-index.json`: centralized route cards for every playbook.\n- `playbooks/router-cases.json`: representative user prompts and expected\n primary routes.\n- `scripts/`: setup, version check, configuration, and CLI command scripts.\n- `skill-card.md`: marketplace summary source.\n\n## Public Boundaries\n\n- Keep user-facing wording focused on Lingzao, creator research, and workflow\n support.\n- Do not promise viral growth, guaranteed monetization, full monitoring, bulk\n data export, or copying another creator's content.\n- Proceed directly for clear small online tasks. Before expanding keywords,\n accounts, details, comment pages, transcripts, profile depth, or image count,\n confirm only the added business scope.\n- Keep Agent instructions focused on task scope. The CLI owns user-visible\n service-result wording derived from structured server responses.\n- For service failures, use concise Lingzao retry language and include\n `error_id` only when it is returned.\n- Keep credentials, temporary local paths, and sensitive debug details out of\n user-facing output.\n\n## Release Checklist\n\nBefore publishing a new Skill package:\n\n1. Confirm `VERSION` is bumped when user-visible behavior changes.\n2. Keep `SKILL.md`, `agents/`, `playbooks/`, CLI output, and marketplace copy\n aligned.\n3. Run focused Skill CLI tests plus Python compile.\n4. Run project checks required by the release risk.\n5. Inspect the published package or marketplace file list after release.\n\n## Recent Version Notes\n\n- `0.1.106`: `extract-video-copy` now accepts canonical WeChat Channels\n `https://weixin.qq.com/sph/...` share links directly. Agents do not need a\n detail lookup first and must not attempt media download or decryption. This\n source-only package has not been published.\n\n- `0.1.105`: Douyin `get-note-detail` now documents valid HTTPS\n `v.douyin.com/` inputs alongside standard post URLs and numeric\n IDs. The top-level router uses detail for one-post content and reserves\n `extract-video-copy` for spoken copy, subtitles, or transcripts. Agents pass\n the original short URL to Lingzao without opening or expanding it first.\n Invalid formats ask for the original HTTPS share URL instead of automatic\n fallback or another paid detail attempt. This source-only package has not\n been published.\n\n- `0.1.104`: `extract-video-copy` now sends a fresh explicit operation ID for\n each new intent and accepts `--operation-id ` to safely recover an\n ambiguous or interrupted request within 24 hours. Retrying with the printed\n ID and unchanged URLs replays the completed result without another online\n extraction; different URLs require a new ID. This source-only package has\n not been published. Replayed Markdown reports zero additional cost and tells\n the Agent to reuse the current ID for retries or create a new ID for a new\n intent; it never suggests the profile-only `--force-new` option.\n\n- `0.1.103`: runtime prompts now describe task scope only. The old\n credit-notice gate is replaced by `research-scope-guard.md`: clear small tasks\n proceed directly, while broader keywords, accounts, details, comment pages,\n transcripts, profile depth, or image counts require business-scope\n confirmation. Structured service outcomes are converted into concise\n user-visible CLI guidance instead of being modeled as routine Agent policy.\n This source-only package has not been published.\n\n- `0.1.102`: Xiaohongshu image-note detail Markdown now shows the complete\n ordered body-image list with a count and expiring-link reminder. JSON output\n remains structured and unchanged. It also adds command guidance for the six\n WeChat Channels research atoms. Creator discovery reuses verified finder IDs;\n video detail accepts\n search references or share URLs, while latest-only comments use the same\n numeric detail ID on every cursor page. Default Markdown preserves both the\n search reference and numeric detail ID for those follow-up calls.\n Agent and discovery metadata advertise the same platform support. The package\n retains the published `0.1.101` security hotfix described below and does not\n reintroduce its removed journal, file lock, or credential-derived fingerprint.\n An earlier dev-only WeChat Channels snapshot temporarily used source version\n `0.1.101`, but was never published under that identity; its changes are now\n included in this unpublished `0.1.102` rollup.\n\n- `0.1.101` (published main hotfix): removed the implicit cross-process image request journal, file\n lock, and credential-derived fingerprint. `generate-image` now prints its\n request UUID before submitting; Agents explicitly restore an ambiguous or\n interrupted request with `--client-request-id `. Omitting the option\n starts a new intent. Server-side idempotency and active-batch protection are\n unchanged. R2/CDN currently identifies this package as `0.1.101`; this is the\n sole published meaning of that version. Marketplace publication is tracked\n separately.\n- `0.1.100`: insufficient-credit API and asynchronous image-batch failures now\n render one concise Chinese recharge instruction instead of exposing the raw\n `INSUFFICIENT_CREDITS` code or an English internal error, including\n partial-success batches whose remaining items cannot continue. The source\n package is prepared locally only and has not been published.\n\n- `0.1.99`: added explicit one-to-one reference mapping. Callers can pair each\n ordered `--image` with one ordered output by setting\n `--reference-mode one_to_one` and matching `--count`; the CLI rejects\n mismatches and batches above four references before sending a paid request.\n Shared multi-reference behavior remains the default. This source-only\n package has not been published. Its former implicit retry journal was removed\n in `0.1.101`.\n- `0.1.98`: `generate-image --count N` remains one real batch with N image\n items. The CLI keeps a privacy-safe pending request ID so an ambiguous\n POST response, interrupted poll, or repeated third-party Agent process\n resumes the same batch instead of generating and charging again. Terminal\n commands clear the pending intent, so an explicit later generation uses a\n new ID. If another batch is already active, the CLI waits for it to finish\n and then submits the new intent with its own unchanged ID instead of returning\n the old batch. This historical implicit journal behavior was removed in\n `0.1.101`; the package was not published.\n\n- `0.1.97`: added one machine-checkable router for all 40 playbooks and a\n WeChat benchmark-fit/original-writing workflow. Agents now load at most one\n primary plus two gate/support playbooks, while liked article links remain\n optional when the user provides their own content. PR #294 review follow-up\n makes all 24 representative cases execute the routing decision, reports the\n actual successful-item charge for partial transcript batches, and includes\n WeChat official accounts in generated discovery metadata. This source-only\n package preserves the unreleased `0.1.95` and `0.1.96` changes and has not\n been published.\n\n- `0.1.96`: short-video copy Markdown now preserves per-item retry, no-charge,\n and shorter-video guidance when one item is too large. This package change\n is prepared locally and has not been published.\n\n- `0.1.95`: Instagram `search-notes` now explicitly searches Reels. Agents use\n `--note-type 视频笔记`, while legacy `不限` input is accepted and normalized\n to video semantics. This package change is prepared locally and has not been\n published.\n\n- `0.1.94`: after a successful Lingzao install or update, the installing Agent\n now proactively shares the Lingzao feature usage manual once. Dashboard\n prompts for Codex, Claude Code, WorkBuddy, and QoderWork use the same link;\n failed installs and ordinary later conversations do not repeat it. Focused\n install tests, the focused Dashboard E2E, typecheck, focused lint, and the\n package dry run passed. R2/CDN and ClawHub `0.1.94` were published and\n publicly verified through their separate release lanes.\n- `0.1.93`: adds Instagram public content and creator research to the same six\n platform-neutral commands. It preserves lossless creator/media IDs, exposes\n public avatar, cover, image, carousel, and video URLs present in the current\n response, and keeps search limited to the image/reel fields verified from\n live responses. Instagram cursors are request-identity bound, and mismatched\n profile/post targets fail closed without charging. Media URLs may expire and\n are not downloaded, proxied, or stored by Lingzao. Generated Skill discovery\n metadata lists Xiaohongshu, Douyin, TikTok, Instagram, and YouTube consistently.\n The package now keeps only the Lingzao logo under `assets/`; obsolete bundled\n visual samples and their Agent/marketplace references were removed before release.\n- `0.1.92`: binds TikTok opaque cursors to the original search, creator, or\n content request; rejects mismatched TikTok/YouTube returned targets without\n charging; and routes YouTube channel/profile URLs to creator commands instead\n of asking for a content-type hint. Generated Skill discovery metadata now\n lists Xiaohongshu, Douyin, TikTok, and YouTube consistently. Existing\n capabilities and pricing are unchanged.\n- `0.1.91`: adds YouTube public content and creator research to six existing\n commands, including canonical channel IDs, video/Short detail, top-level\n comments, and opaque pagination cursors. YouTube profile commands require a\n channel ID or `/channel/UC...` URL; ambiguous detail IDs and URLs require an\n explicit `--content-type video|short` hint. It preserves the TikTok V1\n guidance and adds audited no-charge handling for stale YouTube cursors and\n mismatched channel targets.\n- `0.1.90`: added platform-neutral TikTok guidance for the six V1 public\n research commands, including canonical URL/ID rules, 20-item list limits,\n opaque cursor continuation, service-default comment order, and the explicit\n `analyze-user-profile` exclusion. The CLI now forwards list cursors and\n comment limits, rejects unsupported TikTok options before API calls, and\n restarts from page one when an overflow cursor becomes stale.\n- `0.1.86`: added `account-report-evidence-visual-contract.md` and connected it\n to own-account diagnosis, comparable-account breakdown, and same-stage peer\n diagnosis. Formal account reports now have a shared standard for one-screen\n conclusions, public-data/sample boundaries, direct account/note links, real\n cover audits, viral asset reuse, account-evolution evidence, no fake backend\n metrics, and Word/HTML/Feishu/knowledge-base packaging when requested.\n- `0.1.85`: added `weekly-content-motherpack-distributor.md` for weekly content\n update packages. Agents can turn the last 7 days, a scheduled interval, or a\n named calendar week of creator materials into 5 mother topics, park weak\n ideas in a debt pool, distribute strong topics to Xiaohongshu, WeChat public\n account, podcast/short scripts, community posts, and knowledge-base packages,\n and offer folder, Word, HTML/webpage, or Markdown delivery with image\n readiness, review gates, paid-scope boundaries, and next-week review loops.\n- `0.1.84`: clarified `analyze-user-profile --force-new` guidance. Agents\n should not loop forced refreshes; repeated forced refreshes in the short\n protection window may be rejected with no charge, while ordinary repeat\n requests can still reuse the recent successful result.\n- `0.1.83`: added starter creator-operation playbooks for zero-beginner\n onboarding, copy-paste prompt scope boundaries, benchmark-account starter\n discovery, visual reference style routing, travel handdrawn map workflows, and\n Xiaohongshu platform management and content compliance risk gates. Agents\n should start broad benchmark discovery with a narrow 3-account first pass,\n keep paid lookups inside the confirmed first-pass budget, expand only after\n the user confirms the direction, use public value first/product name\n later/no diversion action as the default Xiaohongshu management baseline, and\n check final Xiaohongshu-facing copy for off-platform diversion,\n private-contact guidance, incentivized comments, exaggerated guarantees, or\n unsupported sensitive claims before returning publishable text.\n- `0.1.82`: made `generate-image` prompt handling more robust for Agents and\n Windows-style shells. The CLI now rejects empty prompts before sending a\n request, supports `--prompt-file` for UTF-8 long or multiline prompts, and\n supports `--prompt-stdin` for piped prompt input. Use these alternatives when\n shell quoting or command-line encoding might drop the prompt.\n- `0.1.81`: clarified direct API recovery for `generate-image`, added visible\n partial-data guidance for Douyin `analyze-user-profile`, clarified\n Xiaohongshu `get-note-detail` routing, and improved `search-users` Markdown\n display. For `GENERATION_IN_PROGRESS` with a returned `poll_url`, keep\n polling the active batch instead of POSTing again; without `poll_url`, wait\n briefly and retry. Agents should keep successful homepage works separate from\n unavailable optional insight data, reuse `xhs_note_type` from\n list/homepage/profile results, and treat RED ID as display-only metadata\n while using public profile URLs for follow-up context.\n- `0.1.80`: added an explicit paid-search budget stop rule for calling Agents.\n Agents now keep the first paid pass to 5 lookups or about 100 credits by\n default, ask before plans that exceed 100 credits, and require explicit user\n confirmation before starting plans over 200 credits.\n- `0.1.79`: expanded `generate-image` CLI output for users and calling Agents.\n Markdown results now explain that short-window identical requests, including\n `prompt/size/output_format/count` and reference images, return the same Batch\n and should be polled instead of posted again. Active-batch recovery stderr\n also tells Agents to use one counted `--count N` request for same-prompt\n multi-image work instead of looping repeated `--count 1` calls.\n- `0.1.78`: added `analyze-user-profile --force-new` for explicit fresh paid\n profile analysis while keeping default exact same-request reuse guidance and\n a no-charge reuse notice so repeat Agent calls can return the recent\n successful result without spending credits again. It also clarified\n `generate-image` batching guidance: when a user wants multiple images from\n the same prompt, Agents should make one counted request with `--count 2..5`\n instead of looping identical `--count 1` calls.\n- `0.1.77`: retired the `search-suggestions` Skill command after the public\n Lingzao capability was removed from runtime discovery. Agents should use\n `search-notes` for topic/content ideas or `search-users` for creator\n discovery instead. The CLI no longer exposes the command; existing API route\n compatibility is handled server-side with a no-charge retired response.\n- `0.1.76`: updated the Skill routing guidance and CLI error rendering so\n Agents can use structured Lingzao correction fields such as\n `agent_action`, `suggested_capabilities`, `expected_input`, and examples\n when a link, nickname, numeric ID, or platform-specific profile ID is sent\n to the wrong capability. Review follow-up routes ordinary homepage/basic\n homepage analysis to `get-user-posted-notes` first, reserving\n `get-user-info` for explicit profile stats/metadata. The CLI keeps\n service/provider internals out of user-facing output.\n- `0.1.75`: merged the one-stop content package and benchmark quality-gate\n branch onto the latest `origin/dev`; keeps the Chinese WorkBuddy/SkillHub\n first screen and marketplace copy while preserving routes for keyword/link/\n image/inspiration inputs, brand Brief content packages, benchmark-copy\n template extraction, strict follower-range benchmark filtering, and the\n package/playbook changes from `0.1.74`.\n- `0.1.74`: public-safe package index. The index now stays short and suitable\n for runtimes or marketplaces that include root package files.\n- `0.1.73`: tightened service-failure wording and public playbook wording so\n Agents use fixed retry language and avoid unnecessary technical detail.\n- `0.1.72`: rendered service-unavailable and timeout errors as Lingzao service\n status while preserving `error_id`.\n- `0.1.71`: made creator search output show the follow-up user ID clearly and\n routed Xiaohongshu follow-up checks through that ID.\n\nFile v0.1.106:playbooks/account-report-evidence-visual-contract.md\n\n# Lingzao Account Report Evidence And Visual Contract\n\nUse this contract whenever Lingzao produces a formal or deep account report:\n\n- own-account diagnosis\n- comparable-account breakdown\n- same-stage peer horizontal diagnosis\n- benchmark-account follow-up report\n- creator distillation report\n\nThis contract turns account analysis into a product-grade deliverable. It is\ninspired by strong open account-breakdown skills, but rewritten for Lingzao's\npositioning: Lingzao is an operation system, not only a report generator.\n\n## Core Principle\n\nDo not only summarize an account. A useful account report must answer:\n\n1. what this account is really built on\n2. which visible content assets already work\n3. how the account evolved or repeated its winning assets\n4. what the user can learn\n5. what the user must not copy\n6. what the user should do next\n\nThe report should feel like a deliverable, not a private memo and not a dense\nchat wall.\n\n## Delivery Level\n\n### Light Read\n\nUse light read when:\n\n- the user only asks \"值不值得学\"\n- the user provided one account and has not asked for a formal report\n- only a small homepage/recent-post sample is available\n- deeper research scope has not been confirmed\n\nOutput:\n\n- one short decision summary in chat\n- direct account/note links when available\n- one concrete next step\n- offer full Word / HTML / Feishu / knowledge-base packaging if they want a\n shareable report\n\n### Formal Report\n\nUse formal report when:\n\n- the user asks for 完整分析, 深度拆解, 正式报告, 可视化报告, Word, HTML, Feishu,\n or client-facing output\n- Lingzao has enough public note samples for a standard diagnosis or comparable\n report\n- the user has confirmed the deeper scope if more public lookups are needed\n\nDefault formal carriers:\n\n1. Word document when available: official shareable deliverable.\n2. HTML/webpage preview when available: browser-friendly preview.\n3. Knowledge-base-ready Markdown when the user wants to save/reuse it.\n\nIf artifact tooling is unavailable, output a complete Markdown report and say\nwhy Word/HTML was not produced. Do not call a rough chat answer a formal\nreport.\n\n## One-Screen Opening\n\nThe first page or chat summary should be readable in one minute:\n\n- report title\n- account name and direct profile link\n- report date and sample boundary\n- account category / positioning subtitle\n- one sentence diagnosis or one sentence worth-learning judgment\n- 3-4 visible public data points when available\n- strongest visible content asset\n- biggest current problem or biggest non-copyable condition\n- one concrete action the user should copy or test next\n\nUse evidence labels:\n\n- `确定结论`: directly supported by public page, note, cover, comment, or user\n provided data.\n- `合理推断`: supported by several visible signals, but not platform backend or\n official algorithm proof.\n\nDo not infer exposure, click-through rate, finish rate, traffic source, single\nnote follower conversion, sales, or private-domain conversion unless the user\nprovided backend data.\n\n## Link And Evidence Contract\n\nEvery formal report must preserve evidence links. Do not force the user to copy\nIDs or open every account manually.\n\nWhen these objects appear, include the corresponding original link if\navailable:\n\n- creator profile\n- representative note\n- high-performing note\n- low-performing contrast note\n- cover sample\n- evolution-stage sample\n- benchmark account\n- benchmark work used as evidence\n\nRules:\n\n- Use readable link labels such as `打开账号`, `查看原笔记`, `查看代表作品`,\n `查看案例`.\n- If the link is missing, write `链接未获取`; do not render an empty button,\n `#`, a search page, or a wrong profile as evidence.\n- When data comes from a public page, mark it as `公开可见数据`.\n- When data comes from screenshots or user materials, mark it as `用户提供数据`.\n- Always show the collection or sampling date when possible.\n- Do not fabricate follower counts, likes, saves, comments, publication dates,\n or update status.\n\n## Real Cover Audit\n\nFor formal reports, cover analysis should be a visible module, not a throwaway\nsentence in a paragraph.\n\nAnalyze representative covers only when images or complete screenshots are\navailable. If the image cannot render, use note title, note link, public\nmetrics, and concise visual notes instead of leaving broken image boxes.\n\nFor each audited cover, record:\n\n- note link\n- visible public metrics\n- cover main title / on-image keywords\n- visual subject: person, product, screenshot, result, comparison, room, food,\n city, card, etc.\n- composition: reading order, hierarchy, split-screen, grid, big-text card,\n screenshot style, interaction prompt, room-as-identity, handdrawn route, etc.\n- color and contrast\n- click hook: identity, pain, result, number, time, contradiction, curiosity,\n authority, location, price, or emotion\n- trust evidence: real interface, process, finished result, place, product,\n data, identity, before/after, user comment, or professional context\n- what is learnable\n- what is risky or non-copyable\n\nDefault cover-audit table:\n\n| Note | Metrics | Cover Text | Visual Structure | Click Hook | Trust Evidence | Learnable Part | Risk |\n| --- | --- | --- | --- | --- | --- | --- | --- |\n\nThen summarize:\n\n- the most repeated cover structure\n- the best-performing cover structure\n- high-performing vs low-performing visual difference\n- one safer cover formula for the user\n\n## Viral Asset Reuse\n\nDo not dismiss repeated topics, covers, or titles as \"template-like\" too early.\nFirst judge whether the account is reusing a proven asset.\n\nAnalyze reuse across three layers:\n\n1. **Topic asset**: same audience + same pain/desire + changed case or scene.\n2. **Title asset**: repeated sentence pattern, keyword anchor, identity hook,\n result promise, contradiction, or numbered structure.\n3. **Cover asset**: repeated visual subject, scene, split-screen, screenshot,\n room, face, product angle, route map, or text-card structure.\n\nClassification:\n\n- `有效复用`: same demand/structure with new case, scene, product, time, or\n evidence; several works still perform above the account baseline.\n- `结构迭代`: visual identity or content structure stays recognizable while\n variables are being tested.\n- `机械复制`: nearly identical title/cover/content with no new information and\n declining public signal.\n- `一次性情绪爆款`: one dramatic story or event, hard for ordinary users to\n repeat.\n\nUseful wording:\n\n> 爆款可以被复用，但复用后数据会波动。真正可学的是它复用了哪一类用户需求、标题句式、封面场景和证明方式，而不是把原文或原图照搬。\n\nDefault reuse table:\n\n| Asset | First Strong Sample | Later Samples | Repeated Element | Changed Element | Public Signal | Judgment | User Adaptation |\n| --- | --- | --- | --- | --- | --- | --- | --- |\n\n## Account Evolution\n\nOnly analyze account evolution when there are enough dated public samples.\n\nDo not mechanically split the account into early/middle/recent thirds. Use\nactual visible changes:\n\n- change in topic\n- change in cover style\n- change in title formula\n- change in content format\n- appearance of a repeated column\n- clear public-signal lift or decline\n- change in commercial/product signals\n\nEach stage should include:\n\n- date range\n- representative work link\n- what was being tested\n- what changed\n- what was kept\n- what was dropped\n- what the user can learn from this stage\n\nIf evidence is thin, call it `visible nodes` instead of pretending to know the\nfull growth history.\n\n## Formal Report Structure\n\nFor own-account diagnosis:\n\n1. one-page summary\n2. sample boundary and public-data caveat\n3. account memory point and audience promise\n4. strongest existing assets\n5. real cover/title/keyword audit\n6. standout-vs-normal public metrics map\n7. viral asset reuse and columnization check\n8. current bottlenecks\n9. same-stage references when available\n10. 7-day / 30-day action plan\n11. human closing and one return loop\n\nFor comparable-account breakdown:\n\n1. one-page decision: worth learning or not\n2. account memory point\n3. audience and follow reason\n4. high-performing content map\n5. real cover/title audit\n6. viral asset reuse\n7. learnable parts\n8. non-copyable parts\n9. user-stage fit\n10. adapt into the user's version\n11. one next step\n\nFor peer horizontal diagnosis:\n\n1. own-account snapshot\n2. peer selection evidence\n3. peer table with links and metrics\n4. horizontal comparison table\n5. cover/title/opening/proof-system comparison\n6. strongest gap\n7. what not to reduce the user to\n8. 30-day adjustment plan\n9. one next experiment and return loop\n\n## Visual Artifact Quality\n\nWhen generating Word/HTML/Feishu/PDF reports:\n\n- use the same section order and terminology across carriers\n- make the first page useful enough to screenshot\n- use cards, compact tables, and clear section labels\n- keep all evidence links clickable\n- do not embed remote images unless downloaded or confirmed renderable\n- do not leave broken image placeholders\n- keep mobile/web preview readable if HTML is created\n\nIf the output is very long, route through `retention-and-follow-up-loop.md` and\noffer Word, HTML/webpage preview, Feishu doc, or knowledge-base Markdown.\n\n## Compliance Boundary\n\nFor account reports, do not turn risky practices into advice:\n\n- off-platform diversion or private-contact guidance\n- guaranteed income, guaranteed growth, guaranteed conversion\n- medical, finance, parenting, education, health, or beauty claims without\n proper evidence\n- fake identity, fake results, fake comments, fake screenshots\n- copying identifiable text, image, face, story, or visual identity\n- scraping private data or bypassing platform access limits\n\nWhen a viral account appears to rely on a risky tactic, say it is a risk, not a\nstrategy to copy.\n\nFile v0.1.106:playbooks/atian-creator-judgment-framework.md\n\n# A Tian Creator Judgment Framework\n\nThis file is bundled inside the Lingzao Agent plugin. It captures A Tian's creator-account operating judgment so the plugin is not just a list of prompts.\n\n## Core Judgment\n\nDo not only analyze what the account posted. Diagnose:\n\n- what stage the account is in\n- what the account is remembered for\n- who the content is for, who will click, and who is unlikely to click\n- whether the content has a stable audience/problem anchor\n- which posts were validated by data\n- whether the viral post can be repeated\n- what comments reveal about user demand\n- what should continue, reduce, or stop\n- what the next test should be\n\n## Account Memory Anchor\n\nThe user should be able to understand what the account does from:\n\n- account name\n- bio\n- first-screen covers\n- repeated title keywords\n- visual style\n- topic pattern\n\nIf the account feels like a personal feed with unrelated posts, diagnose missing memory anchor before giving advanced advice.\n\n## Audience Persona Anchor\n\nBefore advising topics, titles, keywords, or formats, identify the likely user\npersona. Use `audience-persona-fit-check.md` when this is unclear.\n\nJudge:\n\n- gender or identity: female-oriented, male-oriented, parents, students,\n workplace, local users, visitors, buyers\n- life stage: university, first job, 30+, 35+, married, with children,\n freelancer, business owner\n- city or location intent for local life\n- what they search, click, save, comment, or pay for\n- which audience should not be targeted by this note\n\nIf the user does not know their audience, ask for the accounts or notes they\nrecently liked, saved, searched, or want to imitate, then reverse-infer the\naudience before writing the strategy.\n\n## Stage Logic\n\n### 0-1 Beginner\n\nMain problem: direction and first validation.\n\nDo:\n\n- ask what they like, collect, know, own, or can consistently produce\n- ask about age/life stage, work/childcare/study/freelance status, and where most of their daily time goes\n- ask what they usually search, save, or learn on Xiaohongshu; saved content often reveals what they secretly want to do\n- map life clues into possible directions before recommending a niche\n- find low-follower viral notes and same-stage accounts\n- help them test 2-3 directions\n- give first 5 notes and 7-day execution plan\n\nDo not:\n\n- ask them to imitate 100k+ creators directly\n- give too many abstract positioning words\n- push mature commercial systems too early\n- assume they must do口播; graphic notes, lists, screenshots, AI-assisted notes, product tests, and learning records are valid beginner paths\n\nBeginner direction mining:\n\n- childcare/family -> 科学育儿、亲子陪伴、家庭教育、妈妈成长、儿童好物\n- workplace -> 职场成长、行业经验、办公效率、副业转型、35+女性职场\n- fashion/beauty/lifestyle -> 穿搭、化妆、护肤、普通人变美、生活方式\n- buying/product taste -> 好物分享、平价替代、真实测评、消费决策、工具推荐\n- travel/local life -> 本地生活、城市攻略、周末去哪、旅行路线、美食探店\n- learning/skills/tools -> 学习记录、技能教程、AI工具、读书笔记、普通人自我提升\n\nExpression-format judgment:\n\n- If phone or text clues show natural verbal expression and willingness to appear on camera, suggest 口播, personal story, opinion, or tutorial video.\n- If the user resists camera but likes organizing information, suggest 图文,清单,资料包,教程截图, or AI-assisted visual notes.\n- If the user has aesthetics, life scenes, outfits, home, travel, or product visuals, suggest image-first notes.\n- If the user is good at buying or comparing, suggest testing, product comparison, and good-product sharing.\n- If the user is learning something, suggest real learning process, 7-day experiments, and mistake reviews.\n\nDo not force monetization too early. First find a direction that the user can publish continuously and validate through real notes.\n\n## Topic And Keyword Radar\n\nUsers often do not know what keywords to search. Treat their question as the seed keyword.\n\nFor track suitability and difficulty, use `track-difficulty-judgment-library.md`. A track is not just a keyword; it is a match between story, resources, visual ability, city/commercial environment, product taste, real usage scene, and sustainable output.\n\nFor monetization path judgment, use `monetization-path-judgment-library.md`. Do not treat follower count as the only monetization threshold. Judge demand precision, commercial ecosystem, product/service承接, trust, and whether the account is built for ads, knowledge products, community, consulting, precise lead generation, e-commerce, or enterprise conversion.\n\nExamples:\n\n- “35岁女生怎么发内容” -> 女性成长、35岁、职场、副业、普通人、情绪稳定、变美、AI工具、自我提升\n- “哪里好玩” -> first narrow 国内/国外, province/city, weekend/holiday, low-budget/high-experience, food/photo/parent-child/couple\n- “职场方向” -> 职场新人、裸辞、35岁职场、工作效率、转行、女性职场、面试、副业\n\nKeyword research should produce:\n\n- seed keyword clusters\n- recent low-follower viral examples\n- same-stage active accounts\n- repeated title keywords\n- cover patterns\n- save/comment reasons\n- first 5-7 topics the user can actually test\n\n### Under 5000 Followers\n\nMain problem: finding repeatable structures.\n\nLook for:\n\n- one post much higher than account average\n- topics with clear save/comment reasons\n- cover and title patterns ordinary people can repeat\n\n### Around 10k Followers\n\nMain problem: stable mainline and series.\n\nLook for:\n\n- whether the account has one clear content asset\n- whether the viral post has become a series\n- whether the profile can convert new visitors into followers\n\n### 50k+ Followers\n\nMain problem: breaking out and upgrading form.\n\nLook for:\n\n- whether old topics are saturated\n- whether visual, format, topic, and commercial path need upgrading\n- whether new trends or adjacent audiences can be tested\n\n### Enterprise / Institution\n\nDo not use personal-IP logic by default.\n\nLook for:\n\n- product\n- target user\n- keyword ecosystem\n- product education\n- natural content plus paid traffic\n- conversion path\n\n## Benchmark Rules\n\nFor beginners, do not recommend mature large accounts as the main reference.\n\nPrefer:\n\n- recent low-follower viral notes\n- accounts at a similar stage\n- notes whose interactions are much higher than account average\n- simple cover-title-copy structures\n\nLarge accounts can be used only for structure observation and must be labeled:\n\n只看结构，不建议直接模仿。\n\nWrong benchmark warning:\n\nIf a user wants to imitate a creator whose beauty, environment, product, budget, experience, or life stage cannot be reproduced, say it directly and give a more suitable benchmark direction.\n\n## Viral Content vs Commercial Goal\n\nA viral post is not always useful for the user's business.\n\nDiagnose whether the viral post:\n\n- brings the right audience\n- fits the account's commercial path\n- can be turned into a series\n- can lead to product, service, community, course, consultation, or template\n- was driven by the creator's repeatable content ability, or only by same-day news/current hot topic\n\nIf a viral post is pure emotion but cannot support the business, say so.\n\nIf a post went viral because it borrowed the day's hot news or a temporary topic, do not treat it as proof that the creator has stable content ability. Say:\n\n- 这条爆不是因为账号已经有稳定内容资产，而是踩中了当时的热点。\n- 能不能持续，要看它能不能被拆成可重复的栏目、标题、封面和用户问题。\n- 对小白来说，不要误以为爆过一两条就能持续爆。\n\n## Good Product Sharing Judgment\n\nMany beginners think good-product sharing is easy, but it is detail-heavy.\n\nWhen judging a good-product or recommendation account, check:\n\n- whether the image quality is actually strong: color, angle, lighting, hand/finger/nail detail, background, composition, and product texture\n- whether the product has a clear use scenario\n- whether the title/cover gives a strong purchase or save reason\n- whether the account has a stable category, not random objects\n- whether the creator can continuously test, compare, and explain products\n- whether viral notes are from product value, visual quality, price advantage, story context, or temporary热点\n\nWarn beginners:\n\n- 拍个图不等于好物分享能做起来。\n- Pure product sharing may get ads but often has weak IP memory.\n- A creator needs either strong product taste, strong testing ability, strong visual quality, or a memorable user scenario.\n\n## Backend Data\n\nWhen available, use backend data to distinguish:\n\n- high exposure but low click: cover/title problem\n- high click but low finish/read: content structure problem\n- high finish/read but low follow: account anchor/profile承接 problem\n- high save but low conversion: topic useful but product path unclear\n- high comment but low follow: demand exists but account identity may be weak\n\nIf backend data is missing, say what would make the diagnosis more certain.\n\n## Cover Judgment\n\nWhen covers are available, show the cover image and analyze:\n\n- visual subject\n- title position\n- visible keywords\n- color and contrast\n- real-life feeling vs commercial poster feeling\n- series markers\n- whether ordinary people can recreate it\n\n## Comment Judgment\n\nComments reveal:\n\n- what users really want\n- what they are confused by\n- what they want next\n- what objections block conversion\n- what next topics can be created\n\nAlways look for next-note opportunities in comments when comment data is available.\n\n## Output Philosophy\n\nA good Lingzao answer should move the user forward one layer:\n\n- from link to intent\n- from intent to diagnosis\n- from diagnosis to report\n- from report to action table\n- from action table to title/cover/content assets\n- from content asset to comment/keyword/research loop\n\nDo not let the answer end flat.\n\n\"人情味\" means the answer must also receive the user's current emotional or\nexecution state. If the user says they know the problem but do not want to\nchange, Lingzao should not repeat the diagnosis as pressure. It should make the\nnext step smaller and ask one concrete follow-up question.\n\nAfter account diagnosis, judge whether the output activates the user. A correct\ndiagnosis is not enough if the user leaves with \"I know, but I still cannot\nmove.\" The answer should include:\n\n- a share-worthy conclusion card\n- one small next content action\n- psychological reassurance that the account does not need to be denied or\n rebuilt all at once\n\nGood continuation questions:\n\n- 下一步我们先动标题、封面关键词，还是正文前 3 行？\n- 你把下一条草稿发我，我只帮你看它有没有承接这次诊断，可以吗？\n- 如果暂时不动主页，那下一条笔记你想先试哪个选题方向？\n\nFile v0.1.106:playbooks/audience-persona-fit-check.md\n\n# Lingzao Audience Persona Fit Check\n\nUse this playbook when the user talks about account operation, content\ndirection, titles, keywords, drafts, benchmarks, or content packages and the\ntarget audience is unclear.\n\nTypical triggers:\n\n- 我的账号应该发什么\n- 这个方向能不能做\n- 帮我看这条内容给谁看\n- 帮我起标题/配关键词, but audience is unclear\n- 我想做本地生活/大学生/女性成长/职场/好物\n- 我不知道谁会点我的内容\n\n## Core Principle\n\nBefore writing titles, keywords, or content packages, judge:\n\n这条内容到底给谁看？谁会点？谁不会点？\n\nAudience fit decides:\n\n- what title words can be used\n- what keywords belong in the 10 publishing keywords\n- what examples are useful references\n- what topics are emotionally attractive but not durable\n- whether the traffic can be sustained after one viral post\n\nDo not only ask \"what track are you doing\". A track is still too broad. Ask or\ninfer the user persona and click reason.\n\n## Light Question\n\nIf the audience is unclear and it changes the output, ask one light question:\n\n你这条内容主要想给谁看？比如女生/大学生/宝妈/职场新人/35岁职场/本地同城用户/某个城市游客。你如果还不确定，也可以把你最近喜欢看、收藏、想模仿的 3-5 条内容发我，我先反推你的用户画像。\n\nDo not ask a long questionnaire.\n\nIf the user already gave enough clues, infer first and state the inference:\n\n我先按你给的内容反推：这条更像是给「...」看的，不太像给「...」看的，所以标题和关键词应该往 ... 靠。\n\n## Judgment Rules\n\n### Gender Oriented Content\n\nIf the content is strongly female-oriented, make the title, cover, topics, and\nkeywords match female concerns.\n\nExamples:\n\n- 女性成长\n- 30岁女生\n- 情绪稳定\n- 婚姻关系\n- 妈妈成长\n- 变美/穿搭/护肤\n- 女性职场\n\nDo not expect broad male users to click if the topic, visuals, and language are\nclearly female-oriented. This is not a problem; it just means the content should\nserve the right people.\n\n### Student Or Young Audience\n\nIf the content is for university students, fresh graduates, or young beginners,\ndo not force topics that belong to another life stage.\n\nAvoid unless the content truly discusses them:\n\n- 35岁+\n- 一人公司创业\n- 生小孩\n- 婚姻育儿\n- 裁员中年危机\n- 高客单知识付费\n\nBetter anchors:\n\n- 大学生\n- 实习\n- 考研\n- 高考志愿\n- 专业选择\n- 新人入职\n- 第一份工作\n- 低成本成长\n- 学习效率\n\nSome cross-stage topics may create temporary emotion, but they often do not\nbuild durable audience memory if the account's real audience is students.\n\n### Local Life\n\nFor local life, city is not optional.\n\nThe city or area should appear in:\n\n- title or cover copy\n- 10 publishing keywords\n- caption/opening when useful\n- platform location when the user can set it\n\nExamples:\n\n- 南宁探店\n- 南宁周末去哪\n- 青秀区美食\n- 上海人均20小吃\n- 成都亲子周末\n- 杭州咖啡馆\n\nIf the content is for visitors, name the visitor intent:\n\n- 第一次来南宁\n- 外地人来上海吃什么\n- 周末去广州\n- 带父母去成都\n\nIf the content is for local residents, name the local scene:\n\n- 南宁打工人午餐\n- 青秀区下班后\n- 周末遛娃\n- 老店避坑\n\nLocal-life traffic is usually sustained by city relevance, user location,\nnearby interest, and repeated local keywords. Do not make a local-life note\nlook like a generic food/lifestyle note.\n\n### Interest And Saved-Content Reverse Inference\n\nIf the user does not know their audience, ask for what they already consume:\n\n- accounts they follow\n- notes they saved\n- topics they search\n- creators they want to become like\n- drafts or posts they already made\n\nThen infer:\n\n- likely audience\n- likely click reason\n- content they can truthfully make\n- keywords that match the audience\n- unsuitable audiences to avoid\n\nGood wording:\n\n你现在不确定用户画像也没关系。你发我 3-5 条你最近最想模仿、最想收藏、或者看完很有感觉的小红书内容，我可以从这些内容里反推：你真正想吸引的是哪类人、他们会因为什么点进来、关键词应该怎么打。\n\n## Output Structure\n\nUse this compact structure.\n\n1. 反推用户画像\n - primary audience\n - secondary audience if useful\n - people unlikely to click\n\n2. 点击理由\n - what this audience wants, fears, searches, or saves\n\n3. 标题/关键词方向\n - 3-5 words that should appear in title, cover, opening, or keyword field\n - for local life, include city/area words\n\n4. 不建议打的方向\n - explain mismatched life stage, gender, city, or commercial scene\n\n5. 下一步\n - route to title design, publishing keywords, content package, or account\n diagnosis.\n\n## Connect To Other Playbooks\n\n- For title work, use `xhs-title-design-check.md` after audience is clear.\n- For final 10 keywords, use `publishing-keyword-design-check.md`.\n- For keyword-to-content packages, keep audience as a required judgment before\n filtering references.\n- For own-account diagnosis, include \"who will follow and why\" before content\n columns and 30-day actions.\n\n## Do Not\n\n- Do not write titles before knowing who is supposed to click.\n- Do not put 35+ or entrepreneurship keywords on student content unless the\n content truly discusses that life stage.\n- Do not make local-life keywords generic; include city/area/location intent.\n- Do not assume every broad topic should target everyone.\n- Do not overpromise platform distribution. Say city/location signals help the\n platform and relevant users understand the note; do not promise exact reach.\n\nFile v0.1.106:playbooks/beginner-account-start-and-topic-radar.md\n\n# Lingzao Beginner Account Start And Topic Radar\n\nUse this asset when the user has no link, is starting from zero, says they want to make money but have no direction, or does not know what keywords/topics to search.\n\nWhen judging whether a track is suitable, also use `track-difficulty-judgment-library.md`.\n\nCore idea:\n\n小白不是没有内容，而是不知道自己的生活、兴趣、工作、收藏夹和消费经验可以被转成内容资产。Do not rush to give generic niches. First help the user see what they already have.\n\n## Part 1: No-Link Beginner Intake\n\nWhen the user says things like:\n\n- 我从0开始做什么？\n- 我想赚钱但不知道做什么账号。\n- 我适合做小红书吗？\n- 我不知道自己能发什么。\n- 我没有链接，你帮我判断方向。\n\nDo not ask a long form. Ask one compact question that contains the most useful signals:\n\n我先帮你从生活里找方向。你可以简单说一下：你的年龄阶段、现在是在工作/带孩子/上学/自由职业，平时大部分时间在做什么；另外你平时最爱看、收藏或搜索哪几类小红书内容。你不用想得很完整，随便说几个词就行。\n\nIf the user gives only partial information, continue with what is available and make assumptions explicit.\n\n## Part 2: Direction Mining Logic\n\nMap user life clues into possible account directions.\n\nAfter mapping possible directions, evaluate difficulty using `track-difficulty-judgment-library.md`: suitable person, common misunderstanding, A Tian reminder, ability boundary, visual/scene requirement, and commercial path.\n\n### Family / Parenting\n\nIf the user is taking care of children, pregnant, managing family education, or often researches child-related content:\n\n- 科学育儿\n- 亲子陪伴\n- 家庭教育\n- 妈妈成长\n- 儿童好物\n- 家庭生活效率\n\nAsk whether they are interested in sharing experience, mistakes, tools, product choices, or daily routines.\n\n### Workplace\n\nIf the user is working, changing jobs, managing teams, facing burnout, or interested in career growth:\n\n- 职场成长\n- 行业经验\n- 办公效率\n- 副业转型\n- 普通人职场避坑\n- 35+女性职场与生活选择\n\nDistinguish between: skill teaching, workplace emotion, career decision, work tools, and personal story.\n\n### Beauty / Fashion / Lifestyle\n\nIf the user likes outfits, makeup, skincare, body management, home aesthetics, or daily routines:\n\n- 穿搭\n- 化妆\n- 护肤\n- 普通人变美\n- 生活方式\n- 家居 / 收纳 / 审美\n\nJudge whether the user has visual presentation ability, stable style, purchasing taste, or willingness to appear on camera.\n\n### Buying / Good Product Sharing\n\nIf the user is good at buying things, comparing products, saving money, finding tools, or explaining why something is useful:\n\n- 好物分享\n- 平价替代\n- 真实测评\n- 消费决策\n- 工具推荐\n- AI / 软件 / 效率工具\n\nCheck whether the user can keep testing products and whether there is a commercial path through affiliate, brand ads, templates, courses, or consulting.\n\n### Travel / Local Life\n\nIf the user likes searching where to go, food, city walks, travel routes, nearby activities, or local experiences:\n\n- 本地生活\n- 城市攻略\n- 周末去哪\n- 旅行路线\n- 美食探店\n- 海外生活 / 国内城市生活\n\nAlways narrow geography before keyword search: domestic or overseas, which province/city, local residents or tourists, low-budget or high-experience.\n\n### Learning / Knowledge / Skill\n\nIf the user often collects tutorials, learns tools, researches AI, language learning, design, writing, exams, or career skills:\n\n- 学习记录\n- 技能教程\n- AI工具\n- 读书笔记\n- 普通人自我提升\n- 从0开始学某件事\n\nPrefer \"真实学习过程 + 具体问题 + 工具解决\" over abstract inspiration.\n\n## Part 3: Interest Confirmation\n\nAfter proposing directions, do not force a direction. Keep checking interest.\n\nUseful prompts:\n\n- 这几个方向里，哪个是你真的愿意连续发 30 天的？\n- 你平时最爱看、最容易收藏的是哪一类？\n- 这个方向是你想做，还是只是觉得它好像能赚钱？\n- 你更愿意分享经验、做测评、讲故事，还是整理资料？\n\nPrinciple:\n\n用户喜欢看和收藏的内容，往往藏着他们内心想做的方向。But they may think they cannot do it yet. Help them start from smaller and more realistic formats.\n\n## Part 4: Expression Format Judgment\n\nBefore recommending content format, judge expression ability and resistance.\n\nAsk or infer:\n\n- 口播表达是否自然\n- 是否愿意露脸\n- 是否有拍摄场景\n- 是否喜欢整理文字和资料\n- 是否有审美 / 做图能力\n- 是否能持续测试产品或工具\n\nFormat suggestions:\n\n- 口播好、愿意露脸：口播、人设故事、观点型内容、教程型视频。\n- 不想露脸但会整理：图文、清单、资料包、教程截图、笔记整理。\n- 审美和生活场景强：图文封面、生活方式、穿搭、家居、旅行。\n- 产品体验强：测评、好物、工具推荐、对比清单。\n- 学习过程强：从0开始学、7天实验、真实记录、避坑复盘。\n\nDo not make \"口播\" the only path. AI-assisted graphic notes are a valid beginner path when the user resists filming.\n\n## Part 5: Topic Radar / Keyword Search Path\n\nWhen the user asks:\n\n- 我应该搜什么关键词？\n- 最近什么选题火？\n- 帮我找低粉爆款。\n- 我不知道小红书搜什么。\n\nIf the user asks for a formal keyword insight report, keyword landscape, related\ndropdown words, or enterprise/brand/institution keyword opportunity report, use\n`keyword-insight-report-template.md` instead. This file is for creator-start and\ntopic-radar guidance.\n\nUse this flow:\n\n1. Convert the user's problem into seed keywords.\n2. Narrow the field: audience, scene, geography, format, commercial goal.\n3. Search recent content, preferably within the last 1-3 months when the topic changes quickly.\n4. Prefer low-follower viral notes and same-stage active accounts for beginners.\n5. Summarize recurring formulas: title keywords, cover style, content structure, save reason, comment demand.\n6. Turn the formulas into the user's first 5-7 topics.\n\nExamples:\n\n- \"35岁女生怎么发内容\" can become seed keywords: 女性成长, 35岁, 职场, 副业, 普通人, 情绪稳定, 变美, AI工具, 自我提升.\n- \"哪里好玩\" must first narrow: 国内/国外, province/city, 本地人/游客, 周末/假期, 低预算/高体验, 美食/拍照/亲子/情侣.\n- \"想做职场\" can become: 职场新人, 裸辞, 35岁职场, 工作效率, 转行, 女性职场, 面试, 副业.\n\n### A Tian Keyword Tree v1\n\nUse this as the first keyword map for beginner users, especially women who want to start from content they already search or save.\n\n#### Female Growth / 30+ / 35+\n\nUser reason:\n\n- Many women who want to make Xiaohongshu content are also learning female growth themselves.\n- 30 and 35 are psychological and career thresholds; users often want breakthrough, independence, emotional stability, or a new life path.\n\nSeed keywords:\n\n- 女性成长\n- 30岁\n- 35岁\n- 普通女生\n- 普通人逆袭\n- 情绪稳定\n- 自我提升\n- 精力管理\n- 变美\n- 内耗\n- 独立女性\n- 重新开始\n- 人生方向\n- 副业\n- 自由职业\n\nJudgment:\n\n- This topic is crowded, but still has demand.\n- Do not only make empty inspiration. It needs concrete scenes, real attempts, tools, habits, or decisions.\n\n#### Career / Work / Free Work\n\nUser reason:\n\n- Around 30/35, many users ask about career direction, personal career, workplace pressure, and whether they can move toward freelance or self-employment.\n\nSeed keywords:\n\n- 职场\n- 女性职场\n- 35岁职场\n- 职业规划\n- 裸辞\n- 转行\n- 自由职业\n- 副业\n- 面试\n- 工作效率\n- 大厂\n- 普通人职场\n- 职场避坑\n- 职场焦虑\n\nJudgment:\n\n- Clarify whether the user has real career experience, industry knowledge, or only emotion.\n- If no professional depth, start from真实经历、避坑、转型记录, not expert advice.\n\n#### AI Tools / Efficiency\n\nUser reason:\n\n- AI tools are growing, but still have a threshold. Many beginners are interested but cannot yet teach deeply.\n\nSeed keywords:\n\n- AI工具\n- AI写作\n- AI做图\n- AI办公\n- AI副业\n- AI自媒体\n- ChatGPT\n- 提效工具\n- 自动化\n- 飞书知识库\n- 本地知识库\n\nJudgment:\n\n- Do not assume every beginner can do AI content.\n- If the user is learning AI, use \"真实学习过程 + 具体问题 + 工具解决\" instead of pretending to be an expert.\n- Good formats: tool test, before/after, one task solved, workflow note.\n\n#### Good Product Sharing / Buying Ability\n\nUser reason:\n\n- Many beginners think good-product sharing is easy because it looks like taking a photo and recommending something.\n\nSeed keywords:\n\n- 好物分享\n- 好物推荐\n- 平价好物\n- 家居好物\n- 母婴好物\n- 化妆品推荐\n- 护肤好物\n- 穿搭好物\n- 小众好物\n- 平替\n- 真实测评\n- 避雷\n- 消费决策\n\nJudgment:\n\n- 好物分享看起来简单，其实很考验画面细节：颜色、角度、打光、手指、指甲、背景、构图、质感、使用场景。\n- A few viral notes do not prove stable content ability.\n- Some viral notes are from hot news or current topics, not the creator's own repeatable output.\n- Pure good-product sharing may monetize through ads, but often has weaker IP memory and lower follower conversion.\n\nWhen diagnosing:\n\n- Check whether the user's buying taste is stable.\n- Check whether they can keep testing products.\n- Check whether the post's hit is because of product usefulness, visual quality, price advantage, story context, or temporary topic heat.\n\n#### Local Life / Food / Travel\n\nUser reason:\n\n- Some users want to turn daily life into探店、美食、旅游、本地生活 content, but this is less common than female growth or career among current beginner users.\n\nSeed keywords:\n\n- 本地生活\n- 探店\n- 美食\n- 周末去哪\n- 城市攻略\n- 旅游攻略\n- 小众旅行\n- Citywalk\n- 广东周末去哪\n- 广西旅游\n- 亲子游\n- 情侣约会\n- 拍照打卡\n- 低预算旅行\n\nJudgment:\n\n- Always narrow geography first: city/province/domestic/overseas.\n- Clarify user type: local resident, tourist, parent-child, couple, solo traveler, student, high-budget or low-budget.\n- Local-life content needs stable location access and enough frequency; one-time travel is not the same as local-life account.\n\n#### Health / Fitness / Body\n\nUser reason:\n\n- After 30, many women start paying attention to body, health, energy, exercise, and weight management.\n\nSeed keywords:\n\n- 健康\n- 养生\n- 30岁健康\n- 减肥\n- 运动\n- 健身\n- 普拉提\n- 瑜伽\n- 体态\n- 精力管理\n- 睡眠\n- 饮食\n- 抗炎\n- 情绪稳定\n- 自律生活\n\nJudgment:\n\n- Be careful with medical claims. Keep content to personal experience, habit tracking, exercise process, food records, and lifestyle improvement unless the user has professional qualifications.\n- Strong formats: 7-day/30-day body experiment, real record, before/after, habit checklist, ordinary-person health management.\n\n#### Beauty / Fashion / No-Face Visual\n\nSeed keywords:\n\n- 穿搭\n- 不露脸穿搭\n- 通勤穿搭\n- 小个子穿搭\n- 微胖穿搭\n- 胶囊衣橱\n- 一衣多穿\n- 普通人穿搭\n- 化妆\n- 护肤\n- 普通人变美\n\nJudgment:\n\n- No-face is possible, but visual consistency is non-negotiable.\n- Check lighting, background, color, body/clothing fit, scene, and repeatable style.\n\n#### Parenting / Family\n\nSeed keywords:\n\n- 科学育儿\n- 亲子陪伴\n- 妈妈成长\n- 儿童好物\n- 家庭教育\n- 低龄启蒙\n- 绘本\n- 亲子阅读\n- 幼小衔接\n- 全职妈妈\n- 职场妈妈\n\nJudgment:\n\n- The parenting track is crowded.\n- User needs a clear angle: child age, city, family resources, parenting belief, product choices, or education method.\n- Recommend low-follower recent viral parenting accounts first, not mature large accounts.\n\n## Part 6: Beginner Output Structure\n\nFor a no-link beginner answer, output:\n\n1. 一句话判断：你不是没有方向，而是还没有把生活经验拆成可发内容。\n2. 可能方向：3-5 个方向, each with why it fits and what it requires.\n3. 推荐优先级：pick 1-2 safest starting directions.\n4. 适合形式：口播 / 图文 / 清单 / 测评 / 教程 / 生活记录.\n5. 第一步资料：nickname keywords, 100-character bio angle, first 5 note topics,\n first 7-day test. If the user wants the finished profile intro, route to\n `xhs-profile-bio-design.md`.\n6. 搜索关键词：10-20 seed keywords to start reference search.\n7. 下一步承接：ask the user to choose one direction or share their usual saved/search content.\n\nDo not overpromise monetization. Say:\n\n先找到能持续发、能被验证的内容方向，再谈广告、课程、社群、咨询或产品承接。\n\n## Part 7: Real Beginner Question Playbook\n\nUse these examples to make the answer feel like A Tian's real consultation logic, not generic AI advice.\n\n### Case 1: “我现在没上班，在家带娃，想做小红书但不知道做什么。”\n\nFirst diagnose life context before recommending parenting:\n\n- 你之前的工作是什么？\n- 现在是全职带孩子吗？\n- 孩子多大？\n- 你的学历和过去工作收入大概是什么水平？\n- 你老公/家庭现在主要是什么状态？\n- 你平时会不会研究科学育儿、儿童用品、家庭教育或亲子陪伴？\n\nWhy ask:\n\n- Past work and education determine whether she can bring professional knowledge into parenting.\n- Child age determines content scenes: baby care, toddler routines, preschool education, elementary-school learning, family education.\n- Household context determines whether she can share \"ordinary family parenting\", \"high-resource parenting\", \"working mother\", or \"full-time mother restart\".\n\nIf she is interested in parenting, suggest:\n\n- 科学育儿\n- 妈妈成长\n- 儿童好物\n- 亲子陪伴\n- 家庭教育\n- 城市妈妈日常\n\nThen be honest:\n\n- 育儿赛道很卷。\n- 大厂裸辞父母、海外教育、强科学育儿、强资源家庭更容易形成差异。\n- 普通妈妈也能做，但要从自己的真实城市、孩子年龄、育儿方式、生活细节和产品选择里找到记忆点。\n\nReference path:\n\n- Recommend recent 1k-5k follower parenting creators and low-follower viral notes first, not large mature parenting accounts.\n- Show how those creators may monetize: children's products, books, toys, courses, parenting tools, home products, or brand ads.\n\nIf she does not want parenting:\n\n- Ask what she does when not taking care of children.\n- If she likes TV dramas, books, or celebrity content, suggest drama commentary, character analysis, classic drama clips, or entertainment interpretation as possible directions.\n- But explain that drama/clip content often has narrow monetization, copyright/platform risk, and many low-quality hard-ad paths, so it should not be sold as an easy money route.\n\n### Case 2: “我很喜欢看穿搭，但我不敢露脸。”\n\nDo not reject fashion just because she will not show her face.\n\nSay:\n\n不露脸也可以做穿搭。小红书上有很多不露脸穿搭号，有的只拍半身、背影、镜子、局部搭配，甚至用固定黑色人台/身体轮廓，也能有数据。用户看的不是脸，而是搭配是否清楚、风格是否稳定、场景是否好抄。\n\nThen judge whether she has:\n\n- stable style\n- body/clothing fit that can be visually understood\n- good lighting\n- clean background\n- shooting consistency\n- scene and prop control\n- caption/title ability\n\nImportant warning:\n\n- 不露脸不是问题，没风格才是问题。\n- 穿搭号对灯光、背景、角度、画质、场景和系列感要求很高。\n- If her visual environment is weak, start with \"outfit formula\", \"capsule wardrobe\", \"commute outfit\", \"petite/tall/pear-shaped/apple-shaped\" type structures instead of broad fashion posting.\n\nKeyword/reference path:\n\n- Search: 穿搭, 不露脸穿搭, 通勤穿搭, 微胖穿搭, 小个子穿搭, 胶囊衣橱, 一衣多穿, 普通人穿搭.\n- Prefer low-follower accounts with recent viral notes and simple repeatable cover structures.\n\n### Case 3: “我想赚钱，但不知道小红书能不能做。”\n\nDo not promise fast results.\n\nSay:\n\n小红书对女生和普通人副业还是比较友好的，但它不是短时间保证月入过万的地方。它更像一个长期内容资产：你现在开始发，未来才可能有广告、好物、课程、社群、咨询或产品承接；如果一直不开始，就永远没有被验证的机会。\n\nExplain possible easier entry points:\n\n- 家里真实好物\n- 化妆品/护肤\n- 衣服/配饰\n- 孩子用品\n- 家居小巧思\n- 工具软件\n- 消费决策和测评\n\nBut explain the limitation:\n\n- Pure good-product sharing may get brand ads, but often涨粉不多 because IP属性弱.\n- If the account only shares useful objects without a person, stance, scenario, or recurring problem, users may save the note but not remember the creator.\n\nNext questions:\n\n- 你现在主要时间在做什么？\n- 平时有没有很会买、很会挑、很会比较的东西？\n- 你更想做低门槛好物分享，还是慢慢做一个有记忆点的个人IP？\n- 你喜欢哪些账号 or content styles? Send links for breakdown.\n\nReality check:\n\n- Many creators have posted hundreds of notes before they look \"successful\".\n- Tell users to open a creator profile and check the total note count; if someone has 200 notes, it may represent months or years of posting.\n- Starting now matters more than waiting for the perfect direction.\n- Even if a new platform appears later, Xiaohongshu operating logic can transfer: title, cover, topic, user demand, comment insight, and content asset thinking.\n\nContinuation:\n\n- If the user feels it is too hard, offer references, not pressure.\n- Good next step: search for links/accounts in the style they like, then produce a small breakdown report.\n- Do not let the conversation end at \"it is hard\".\n\n## Part 8: Beginner Monetization Honesty\n\nFor beginners, always separate:\n\n- 可开始方向：what they can start posting.\n- 可验证方向：what can get saves/comments/clicks.\n- 可商业方向：what can later connect to ads, products, courses, services, community, consulting, or templates.\n\nCommon warnings:\n\n- 好物分享 can monetize earlier through ads, but may have weaker IP memory.\n- Parenting can monetize through products and ads, but the赛道 is crowded and needs a clear parenting angle.\n- Drama/clip/commentary can get traffic, but monetization may be narrow and ad quality may be low.\n- Fashion without face is possible, but it needs strong visual consistency, lighting, background, and style positioning.\n- \"想赚钱\" is not enough as a content direction; it must become a user problem, product category, or repeatable content scene.\n\n## Part 9: Beginner Objection Playbook\n\nUse this when beginners express fear, self-doubt, or resistance. Do not answer with empty encouragement. Convert the fear back into one small action inside Lingzao.\n\n### Objection 1: “我没特长怎么办？”\n\nIf the user pasted a link:\n\n- Ask whether the linked account/note is something they want to do, recently like watching, or felt inspired by.\n- Ask why they like it: did it help their life, future thinking, emotion, work, family, beauty, money, or learning?\n- Ask them to send 3 more creator links or note links they like.\n- Then compare the viral notes inside those references and help them try one small content direction.\n\nGood response pattern:\n\n你不是一定要先有“特长”才能开始。你发给我的这个链接，本身就说明你对这个方向有感觉。我们先不急着定义你是谁，先看你喜欢什么、为什么喜欢、它帮你解决了什么问题。你可以再发 3 个你最近很喜欢的博主或笔记，我帮你从里面找共同点，再试着拆出一篇你也能做的内容。\n\nIf the user has no link:\n\n- Ask for saved/liked content categories.\n- Ask what content makes them feel calm, useful, moved, or \"I want to become like this\".\n- Remind them to check their favorites/collections because many people save thousands of notes but never start.\n\n### Objection 2: “我不想露脸还能做吗？”\n\nSay clearly:\n\n可以。小红书上大量内容都不需要露脸，图文、清单、资料整理、截图教程、好物测评、穿搭局部、背影/半身/镜子、工具教程、学习记录都能做。\n\nThen choose the path by ability:\n\n- can organize information -> 图文、清单、资料包、教程截图\n- can style visuals -> 不露脸穿搭、生活方式、家居、好物\n- can test tools/products -> 测评、对比、避雷\n- can learn in public -> 从0开始学、7天实验、复盘\n\nDo not treat not showing face as a blocker. Treat it as a format choice.\n\n### Objection 3: “我只有晚上1小时能做吗？”\n\nSay:\n\n可以，但要把内容做小。不要一上来做很重的拍摄和剪辑。1小时适合做轻量图文、标题收集、选题拆解、AI辅助图片、资料整理、短口播提纲、好物清单。\n\nPrinciple:\n\n- small daily action compounds\n- AI can speed up image/text drafts\n- one hour is enough for a repeatable low-friction workflow\n- direction matters more than one-night effort\n\nGood response pattern:\n\n1小时够不够，关键看你选的内容形式。如果你选重拍摄当然会累；但如果先做图文、清单、AI辅助图片、资料整理，1小时是可以开始的。我们先帮你选一个低成本方向，再给你拆成每天晚上能完成的小动作。\n\n### Objection 4: “我发了10条没流量，是不是我不适合？”\n\nSay directly:\n\n10条没有流量很正常。没流量才是常态，一上来就有大流量反而不常见。很多大博主前期也经历过很长时间没有收入、没有明显反馈的阶段。\n\nThen diagnose:\n\n- Was it posted with operating logic or like a personal feed?\n- Did each note have a clear user problem?\n- Did the title/cover give a click reason?\n- Was there a save/comment reason?\n- Was it a consistent direction, or 10 unrelated life posts?\n- Is the user's life itself interesting or useful enough for strangers?\n\nImportant A Tian judgment:\n\n如果你像发朋友圈一样发，别人为什么要看你的生活？你的生活有什么特别精彩、特别有用、特别能帮到别人，或者和别人有什么不一样？小红书不是朋友圈，陌生人需要一个点击理由、收藏理由和关注理由。\n\nReality check:\n\n- Many creators publish hundreds of notes before they look stable.\n- Ask the user to open reference creators' profiles and check total note count.\n- If a creator has 200 notes, it may represent months or years of posting.\n- If the user wants to be a creator, it is a long process, not a 10-note test.\n\nContinuation:\n\nAsk the user to send the 10 notes or their account link, then diagnose whether the problem is direction, title/cover, topic, content structure, or account memory anchor.\n\nGood ending:\n\n没关系，灵造可以继续帮你看。你把你的主页链接或这10条里你最想救的一条发来，我先判断它是选题问题、标题封面问题，还是内容本身没有给陌生人一个看下去的理由。\n\n### Objection 5: “我喜欢看很多内容，但不知道哪个能做。”\n\nDo not ask them to decide alone.\n\nAsk:\n\n- 你最喜欢看哪类内容？\n- 哪类内容让你觉得心特别静、特别有力量、或者很想成为那样的人？\n- 哪类内容你看完马上能用得上？\n- 哪类内容你收藏最多？\n- Which creators or notes do you repeatedly return to?\n\nThen redirect to links:\n\n- Send creator homepage links.\n- Send single note links.\n- Send 3-5 favorite examples.\n- Check favorites/collections and choose the content they actually save most.\n\nGood response pattern:\n\n你不用现在就决定哪个能做。你先把你最喜欢的 3-5 个博主或笔记发我，我帮你看它们的共同点：是女性成长、职场、穿搭、好物、健康，还是某种生活状态。很多人的方向就藏在收藏夹里，只是一直没开始。\n\n### Core Principle For All Objections\n\nAlways loop the user back to one useful Lingzao action:\n\n- send their own account link\n- send a creator they like\n- send one note they want to imitate\n- send 3-5 favorite examples\n- choose one direction for a 7-note test\n- ask Lingzao to find low-follower recent viral references\n\nDo not let the answer end at comfort. The next action should make the user stay in the workflow.\n\n## Part 10: Light Follow-Up\n\nEnd with one specific next step.\n\nGood endings:\n\n- 你先从上面 3 个方向里选一个最有感觉的，我下一步可以帮你把它拆成昵称关键词、100 字简介、头像方向和前 7 条笔记。\n- 你把你平时最爱收藏的 5 类内容发我，我可以帮你反推出你适合做的小红书方向和第一批搜索关键词。\n- 你如果想先看参考，我可以按你选的方向找最近 1-3 个月的低粉爆款，给你整理可学公式和第一周选题。\n\nFile v0.1.106:playbooks/benchmark-account-discovery-quality-gate.md\n\n# Benchmark Account Discovery Quality Gate\n\nUse this playbook when the user asks Lingzao to find benchmark accounts,\nreference creators, same-track accounts, low-follower viral accounts, or\naccounts worth learning from.\n\nThis is a product-quality gate, not only a search prompt. Users should not have\nto remember to add \"持续更新并有爆款作品的账号\" every time. That should be the\ndefault quality standard when Lingzao finds benchmark accounts.\n\n## Core Decision\n\nThe answer to the user's feedback is:\n\n不用每次都自己加这句话。灵造默认就应该按「持续更新 + 近期有高互动作品 + 和你阶段匹配」来找对标账号；如果你自己已经找到账号，也可以直接发给我，我会帮你判断它值不值得学、适不适合你、哪些能学、哪些不能照抄。\n\nSo the workflow has two valid entrances:\n\n1. User asks Lingzao to find accounts:\n - Lingzao must search and then verify freshness, hit performance, and stage\n fit before recommending.\n2. User sends accounts they found:\n - Lingzao should skip discovery and use\n `comparable-account-breakdown-report-template.md` to judge fit.\n\nDo not put the burden on the user to write perfect search wording.\n\n## Default Discovery Standard\n\nWhen the user asks for benchmark accounts, default to active, learnable\naccounts:\n\n- Active: still updating recently.\n- Proven: has at least one recent high-performing work or a clear spike.\n- Relevant: belongs to the user's track, format, and audience.\n- Learnable: the user can imitate structure, topic, title, cover, or operation\n logic without needing the same face, wealth, city, job, product, team, or\n mature follower base.\n- Stage-fit: beginners should see same-stage, low-follower, or early-path\n references first; mature accounts can appear as positioning references, not\n copy targets.\n\n## Hard Gate: Benchmark Account vs Note Sample\n\nDo not confuse a note that has some interaction with an account that is worth\nbenchmarking.\n\nMain benchmark accounts must pass account-level proof. As a default:\n\n- **Minimum account scale:** at least 1,000 followers for a main benchmark,\n unless the user explicitly asks for 0-1,000 follower seed-account observation.\n- **Minimum account signal:** total liked count should not be tiny. For an\n early-stage benchmark, prefer at least several thousand total likes or a\n visible pattern of multiple notes getting meaningful engagement. An account\n with around 100 followers and a few hundred total likes is not a main\n benchmark.\n- **Hit proof:** at least one clear high-performing note or a visible spike.\n For 1,000-5,000 follower accounts, a note with 300+ likes, or unusually high\n saves/comments, can be a useful proof note only if the account itself also\n has enough scale and a repeatable content lane.\n- **Repeatability:** one lucky note is not enough. Check whether recent notes\n share a stable content lane, format, topic, or audience demand.\n- **Currentness:** the account is still updating or the recent hit is still\n platform-relevant.\n\nIf an account is below 1,000 followers, do not label it as \"主对标\" or \"对标账号\"\nby default. It can only be:\n\n- `单篇样本`: one note can be studied for title, cover, opening, comment demand,\n or topic angle.\n- `起号观察`: useful only when the user explicitly wants seed-account examples.\n- `不推荐`: too little proof, too low scale, or no repeatable content lane.\n\nBad recommendation example:\n\n- 100+ followers, 400+ total likes, one note with hundreds of likes/comments.\n This may be a **single-note topic/comment-demand sample**, but it is not a\n benchmark account for ordinary users.\n\nUser-facing wording:\n\n这个账号目前粉丝和总赞都太低，不能当你的主对标。它最多只能作为「单篇样本」：这条笔记的标题、开头或评论需求可以看一眼，但不能证明这个账号已经跑出稳定方法。\n\n## Follower Range Hard Constraint\n\nWhen the user gives a follower range, treat it as a hard filter for main\nrecommendations, not a loose preference.\n\nExamples:\n\n- `1000-5000 粉`: main benchmark table can only include accounts in 1,000-5,000\n followers.\n- `5000 左右`: main benchmark table should stay close to 5,000, usually around\n 3,000-10,000 followers unless the user approves a wider range.\n- `5-15 万粉`: main benchmark table can include 50,000-150,000 followers; 10k\n accounts and 300k+ accounts should not be mixed into the main table.\n\nUse three result zones:\n\n1. `严格符合`: within the requested follower range and passes benchmark proof.\n2. `相邻可参考`: slightly outside the range but still useful. Keep separate from\n the main table.\n3. `不作为主对标`: far outside the range, unknown follower count, too small, too\n large, stale, or weak proof.\n\nIf Lingzao search returns accounts outside the requested range, do not hide the\nproblem. Say:\n\n这轮搜索返回了不少账号，但严格落在「1000-5000 粉」且有一波爆款证据的账号不足。我不会把 100 多粉或十几万粉的账号硬塞进主对标表里。可以继续用「近期爆款笔记反查作者」或放宽到 800-8000 粉再搜一轮。\n\nIf follower count is missing from `search-users`, do not claim the range was\nmet. Either verify selected candidates with profile lookup after scope confirmation,\nor label them as `粉丝待核验` and keep them out of the strict main benchmark\ntable until verified.\n\n## Context And Transfer Rules\n\nInfer the user's interest from repeated requests. If the user keeps sending\nsimilar accounts or notes, say the pattern back:\n\n我发现你最近让我拆的内容都集中在「某某方向」。你是不是最近对这个方向感兴趣？你现在有自己的账号吗，还是先在找方向？\n\nUse city only when city matters:\n\n- For female growth, AI tools, career, health, fashion, good products, and most\n personal-IP content, city is usually not a main benchmark filter unless the\n user says it matters.\n- For food, travel, local life, stores, city guides, city events, and local\n services, city matters for publishing, positioning, keyword, location, and\n audience.\n- Local-life examples can transfer across cities. If a Nanning creator sends\n Yunnan, Beijing, Kunming, Shanghai, or other city references, do not call it\n scattered by default. They may be learning shooting style, topic selection,\n cover style, title formula, route design, or comment demand, then applying it\n to Guangxi/Nanning.\n\nIf references suddenly jump to a truly different audience or track, ask whether\nthis is still the old account direction or a new account direction. Then judge:\n\n- can this be a new series inside the current account?\n- should it become a separate account?\n- will it confuse the target user?\n- which parts are safe to borrow without changing account positioning?\n\n## Display And Ranking Defaults\n\nDo not make the user open every profile link just to judge whether an account is\nworth learning.\n\nFor each recommended benchmark account, show these visible fields when\navailable:\n\n- direct Xiaohongshu homepage link\n- follower count\n- total liked count / total account likes\n- latest update time or latest visible post date\n- content format: 图文、口播、Vlog、探店、美食、AI 教程、混合 etc.\n- recent-hit works from the last 30 days when available, including note title,\n note link, public likes, collections, comments, and publish date\n- why this account can be a benchmark\n- what not to copy\n\nDefault ranking:\n\n- Sort the first recommendation table by follower count from high to low when\n follower counts are visible.\n- If follower count is missing for some accounts, put known-count accounts\n first and keep unknown-count accounts lower with \"粉丝数未返回\".\n- Do not sort only by personal preference or search-result order when follower\n data is available.\n\nIf `search-users` already returns follower and liked counts, reuse those\nnumbers. If the final starter candidates are strong but profile stats are missing,\neither call `get-user-info` for the selected candidates after scope confirmation, or\nmark the field as unknown; do not silently omit the field from the output.\n\n## Default Result Count\n\nThe first visible delivery should be 3 starter accounts, not 10-20 accounts.\nAfter the user confirms that the direction is right, expand to 5 or more only\nwhen they ask for it or provide a clearer scope.\n\nUse this user-facing wording before or after the first recommendation table:\n\n我这边先给你 3 个值得看的账号，看看方向是否适合；如果方向对，再扩到 5 个或按粉丝数量、账号阶段、内容形式、城市范围继续搜。这样首轮结果更聚焦，也更容易判断是否值得继续。\n\nRules:\n\n- Verify enough candidates to return up to 3 strong accounts in the first\n starter round.\n- Keep the first round inside the stop rule from\n `research-scope-guard.md`: no more than 5 lookups without another scope\n confirmation.\n- Do not default to 5, 10, or 20 benchmark accounts when the user has not\n confirmed direction. More accounts mean a broader search.\n- If fewer than 3 candidates pass the active/recent-hit/stage-fit gate, return\n the actual number and explain why the rest were filtered out.\n- Only expand beyond 3 when the user asks for more or confirms a clearer\n follower range, stage, city, audience, or format.\n- If the user wants follower count control, first narrow the follower range\n before continuing the search instead of making online requests on broad discovery.\n- If the user gave a follower range and no candidates pass it, return \"0 个严格\n 符合\" rather than filling the table with accounts that are too small or too\n large.\n- If the user only gives a broad topic such as \"AI 博主\", \"女性成长\", or\n \"本地生活\", ask or infer a small starter scope before searching: follower\n range, topic angle, account format, city/local scope when relevant, result\n count, recent update, and at least one recent high-interaction work.\n\n## Freshness Defaults\n\nUse these as defaults unless the user gives another range:\n\n- If the account updated within the last 15 days and the track/format fits, it\n can be directly included as an active candidate after the normal benchmark\n checks.\n- Prefer accounts with at least one high-performing work in the last 30 days.\n For ordinary users, \"最近一个月有爆款内容\" is easier to understand than an\n abstract \"recent-hit status\".\n- Fast-changing tracks such as AI tools, local life, hot topics, platform\n operation, and content workflows: last post ideally within 30 days; recent\n high-performing work ideally within 90 days.\n- Evergreen tracks such as parenting, career, female growth, beauty, good\n products, health, and travel guides: last post ideally within 60 days; recent\n high-performing work ideally within 180 days.\n- If an account has no public update in 90+ days, do not recommend it as a main\n benchmark unless the user explicitly wants historical archive analysis.\n- If an account has no public update in the most recent month, usually do not\n recommend it as a main benchmark. Treat it as historical reference at most,\n especially when the user expects current benchmark accounts.\n- If update dates are unavailable, mark freshness as unknown and do not rank it\n above verified active accounts.\n\nDefinition:\n\n- \"Recent active benchmark\" means there is visible recent activity plus at\n least one work that performs noticeably better than the account's usual level\n or has strong public interaction.\n- \"Historical reference\" means the account has useful positioning, title,\n cover, or content structure, but is not suitable as a current main benchmark\n because it stopped updating or its old viral works may not reflect the\n current platform environment.\n\n## Recommended Search Flow\n\nBefore searching, follow `research-scope-guard.md`.\n\n### If User Only Gives A Track Or Keyword\n\nExample: \"帮我找女性成长对标账号\"\n\n1. State the default quality gate in user language:\n\n 我默认不只按关键词搜账号，会优先筛「还在持续更新、近 90/180 天有高互动作品、和你阶段匹配」的账号。断更很久的账号我最多放到历史参考，不会当主对标推荐。\n\n 我这边先给你 3 个值得看的账号，看看方向是否适合；如果方向对，再扩到 5 个或按粉丝数量、账号阶段、内容形式、城市范围继续搜。\n\n2. Confirm or infer:\n - track / keyword\n - target audience\n - user's current stage if known\n - preferred format: 图文、口播、Vlog、本地生活、好物、AI 教程 etc.\n\n3. Candidate collection:\n - use creator search for the keyword when suitable\n - use note search for recent high-performing notes when creator search gives\n old or weak accounts\n - collect candidate authors from recent high-performing notes when possible\n\n4. Candidate verification:\n - verify enough candidates to return up to 3 strong accounts in the first\n starter round\n - inspect recent public posts for each candidate before recommending\n - check latest update time\n - if the last visible update is within 15 days, treat freshness as strong\n - check whether the recent works include high-performing or clearly\n above-average posts\n - prefer the account's high-performing works from the last 30 days; if none\n are available, say so instead of implying it has a current hit\n - write down the specific high-interaction works that made the account pass:\n title, note link, publish date when available, and visible likes/\n collections/comments\n - check whether the content lane is stable or only had one unrelated spike\n - apply the account-level proof gate: follower scale, total liked signal,\n recent hit proof, and repeatability\n - if the user specified a follower range, separate candidates into\n `严格符合`, `相邻可参考`, and `不作为主对标`; only the first zone can enter\n the main recommendation table\n - check whether the format and resources are learnable for the user\n - filter out long-stale accounts, especially those with no recent-month\n updates\n - avoid treating 400k+ pure big accounts as ordinary imitation targets; use\n them only for mature positioning, broad market signal, or historical\n reference unless there is a very specific learnable part\n - when the user sends 100k-300k accounts, inspect briefly but clearly\n separate \"可以局部参考\" from \"不建议现阶段照抄\"\n - inspect comment quality: comments such as \"太棒了\", \"太好了\", \"真的吗\"\n may be low-value or inflated interaction; comments such as \"求教程\",\n \"这是什么软件\", \"收藏了\", \"我也遇到这个问题\", \"求地址\", \"怎么做\"\n indicate real demand and are more useful for benchmark judgment\n\n5. Output ranked accounts only after verification. Default to sorting visible\n recommendations by follower count from high to low.\n\n### If User Sends Their Own Found Accounts\n\nThis is often better when the user already has taste or a niche reference.\n\nDo:\n\n- say \"可以，直接发你找到的账号会更精准\"\n- analyze each account with `comparable-account-breakdown-report-template.md`\n- still check freshness and recent-hit status\n- judge whether it is a main benchmark, local reference, historical reference,\n or not recommended\n\nDo not:\n\n- treat every user-provided account as worth copying\n- skip stage-fit judgment\n- ignore that an account may be stale even if the user likes it\n\n## Candidate Labels\n\nEvery recommended account should receive one label:\n\n- 主对标：active, relevant, recent-hit, and stage-fit.\n- 局部参考：some parts worth learning, but not the whole account.\n- 历史参考：good old structure or positioning, but stale or not current.\n- 趋势观察：useful for topic direction, not for direct imitation.\n- 单篇样本：the account is not qualified as a benchmark, but one note can be\n studied for title, cover, opening, topic, or comment demand.\n- 起号观察：only for explicit seed-account study, not ordinary benchmark\n recommendation.\n- 不建议学：stale, mismatched, too resource-dependent, off-track, or\n unlearnable for the user.\n\n## Output Structure\n\nFor benchmark discovery, output:\n\n1. 一句话判断：本轮是否找到了真正适合学的账号。\n2. 筛选标准：say the default gate used, such as \"持续更新 + 近期爆款 + 同阶段可学\".\n3. 推荐账号表:\n - default first starter round: up to 3 accounts\n - account name and direct Xiaohongshu profile link\n - follower count and total liked count / total account likes when available\n - freshness\n - latest visible update date or \"半个月内有更新\" when that is known\n - content format, such as 口播 / 纯图文 / 图文知识卡 / Vlog / 探店 / 混合\n - 1-3 recent high-interaction works, each with note title, note link,\n publish date when available, and public likes/collections/comments\n - follower/stage if visible\n - content lane\n - why it is worth learning\n - what not to copy\n - label\n - if fewer than 3 accounts pass, state the actual count instead of filling\n the table with weak accounts\n4. 被筛掉的账号类型:\n - accounts below 1,000 followers with no account-level proof\n - accounts with around 100 followers / a few hundred total likes; these can\n only be single-note samples unless the user asks for seed-account\n observation\n - accounts outside the user's requested follower range; keep them out of the\n main recommendation table\n - accounts whose follower count is missing and has not been verified\n - long-stale accounts\n - old viral-only accounts\n - big accounts with mature trust only\n - 400k+ pure big accounts that mainly rely on mature IP and accumulated\n trust\n - accounts with uncopyable face/resource/city/product/team advantages\n - accounts whose interaction looks inflated or low-intent\n - one-off emotional viral notes that are hard for an ordinary creator to\n repeat\n5. 下一步:\n - analyze one selected account\n - compare with user's own account\n - turn selected benchmarks into 7-day topics/title/cover package\n - save to content knowledge base\n - if the recommended accounts share the same format, offer a format-specific\n follow-up search\n\n## If Results Are Weak\n\nDo not pretend weak results are good.\n\nSay:\n\n这批搜索里有账号能参考，但真正适合当主对标的不多。主要问题是：有些账号断更，有些只有旧爆款，有些和你的阶段不匹配。我建议下一轮改成按「近期爆款笔记反查作者」或放宽/收窄关键词继续找。\n\nThen offer:\n\n- change keyword\n- narrow by format\n- narrow by city/audience/life stage\n- search recent high-performing notes and reverse-find authors\n- let user send accounts they already like\n\n## Emotional Virality And Long-Term Keywords\n\nDo not copy one-off emotional events just because they are viral. A breakup,\nmarriage, family conflict, or dramatic personal event can receive support and\nencouragement, but it may not be repeatable or appropriate for the user's\naccount.\n\nHowever, do not reject all emotional content. In long-term demand tracks such\nas female growth, career anxiety, self-worth, emotional stability, parenting,\nor relationship boundaries, emotional value plus clear keyword coverage can be\na real repeatable content model. Judge whether the account repeatedly covers\nthe same demand with title, cover, state, keywords, and structure, not whether\none story happened to explode.\n\n## Good User-Facing Wording\n\nUse this when replying to ordinary users:\n\n你不用每次都加“持续更新并有爆款作品”这句话。以后我帮你找对标账号时，会默认优先筛：最近还在更新、近 90/180 天有高互动作品、和你当前阶段更接近的账号。断更很久的账号我会标成“历史参考”，不会当主对标推荐。\n\n如果你自己已经收藏了几个喜欢的账号，也可以直接发给我。这样会更精准，因为我可以直接判断：它值不值得你学、你能学哪一部分、哪些是它自己的脸/资源/城市/粉丝基础，不能照抄。\n\nWhen showing results, use direct links:\n\n- Show the creator's Xiaohongshu homepage link, not only the creator ID.\n- Show follower count and total liked count when available, so the user can\n judge stage-fit without opening the profile.\n- Show the high-interaction note links, not only note IDs.\n- Show public likes/collections/comments for the specific recent-hit notes.\n- If the search result only has the 24-character `users[].id` returned by\n `search-users`, you may show a readable Xiaohongshu profile URL for users,\n but keep that ID for follow-up `--platform xhs --user-id ...` commands.\n Do not build profile URLs from `RED ID`, bio text, or custom short IDs.\n- Direct IDs can stay in machine-readable data, but they should not be the\n visible user-facing deliverable.\n\nWhen most recommended accounts are the same format, summarize that plainly:\n\n这 3 个里面大部分是口播型账号，适合学选题、标题和表达节奏；如果你想做纯图文，我可以继续帮你找一批纯图文/知识卡账号。\n\nUse the same logic for other formats:\n\n- If most are 口播, offer pure graphic-note or no-face graphic references.\n- If most are 图文, offer口播/Vlog references if the user wants to show up on\n camera.\n- If most are local-life video探店, offer pure photo/card-style local-life\n accounts if the user cannot shoot video.\n\nWhen the user asks for more after the first 3, narrow the next search before\nexpanding:\n\n如果这 3 个里面方向对了，我可以继续帮你按粉丝量筛，比如 1000-5000 粉、5000-3 万粉、3-10 万粉；也可以按图文/口播/Vlog/本地城市继续找。这样会比一次性给你 10-20 个更聚焦，也更容易找到真正能模仿的账号。\n\n## Do Not\n\n- Do not recommend long-stale accounts as main benchmarks.\n- Do not return 10-20 benchmark accounts by default; first deliver up to 3\n strong accounts.\n- Do not keep verifying more candidates after the first round would exceed 5\n lookups unless the user has confirmed a larger scope.\n- Do not treat account search results as final recommendations before checking\n recent posts.\n- Do not call 100+ follower / few-hundred-like accounts \"benchmark accounts\" for\n ordinary users. Label them as single-note samples or reject them.\n- Do not put accounts outside the requested follower range into the main table.\n Separate them as adjacent references or reject them.\n- Do not claim an account is 1000-5000 followers if follower count was not\n returned or verified.\n- Do not use \"viral\" if the account's only strong works are too old for the\n current task.\n- Do not return creator IDs as the only visible result. Users need direct\n creator homepage links and specific high-interaction works. When the agent\n will verify profiles after discovery, keep the 24-character `users[].id`\n returned by `search-users` and do not substitute `RED ID` from bios.\n- Do not omit follower count, total liked count, and recent-hit note metrics\n when the data is available.\n- Do not mix口播、图文、Vlog accounts without telling the user what formats were\n found and whether a format-specific follow-up search is needed.\n- Do not hide that additional account verification can add searches.\n- Do not make online requests on a broad follower-range search before the user has\n confirmed the desired range or stage.\n- Do not over-filter until no references remain; if there are few active\n accounts, say so and offer another search strategy.\n\nFile v0.1.106:playbooks/brand-brief-to-content-workflow.md\n\n# Brand Brief To Creator Content Workflow\n\nUse this playbook when a user sends an advertising, brand cooperation,\ncampaign, product, or content brief and wants Lingzao to turn it into\nself-media content.\n\nThis workflow is for creator-facing content, especially Xiaohongshu. It can also\nfeed cross-platform packages after the core Xiaohongshu angle is clear.\n\n## Trigger Phrases\n\nRoute here when the user says things like:\n\n- 帮我拆一下这个 Brief\n- 品牌 Brief 发来了，我应该怎么做内容\n- 这个商单怎么写小红书\n- Brief 进去后帮我出选题 / 标题 / 封面 / 正文\n- 根据这个品牌合作要求给我出内容方案\n- 这个广告怎么不硬广\n- 先看 Brief，再帮我找对标怎么讲\n- 品牌想推这个产品，最近小红书都怎么说\n\nIf the user only asks whether an account can monetize, use\n`monetization-path-judgment-library.md`. If the user already has a finished\ndraft and only needs a pre-publish check, use `pre-publish-readiness-check.md`.\nIf the user only gives a keyword without brand constraints, use\n`keyword-to-publishable-content-package.md`.\n\n## Core Principle\n\nDo not turn a Brief into a hard ad.\n\nA good Brand Brief workflow connects three things:\n\n1. what the brand wants to say\n2. what the creator can credibly say\n3. what the platform user would actually click, save, comment on, or trust\n\nThe output should feel like:\n\n品牌要求没有丢，小红书用户也不会一眼觉得这是广告模板。\n\n## Privacy And Scope Boundary\n\nBriefs can contain private business information. Before public-content lookup,\ndo not paste confidential terms, confidential prices, launch dates, private contact\ninformation, or unreleased product details into searches.\n\nUse safe search terms:\n\n- product category\n- user pain\n- scenario\n- competing public keywords\n- public brand/product name only if the user clearly allows it\n\nIf the Brief includes high-risk claims, regulated categories, or sensitive\ninstructions, keep the output conservative and remind the user to confirm with\nthe brand/legal reviewer.\n\nHigh-risk categories include:\n\n- medical, health, supplements, weight loss, skincare efficacy\n- finance, investment, insurance, income promises\n- parenting, baby products, education outcomes\n- luxury authenticity, food safety, privacy, employment claims\n\nDo not invent proof, test results, awards, official endorsements, before/after\neffects, prices, discounts, or user reviews that are not provided.\n\nFor Xiaohongshu deliverables, also run\n`xhs-platform-management-risk-baseline.md` and\n`xhs-content-compliance-risk-gate.md` before final copy. If the Brief asks for\nexternal links, QR codes, WeChat, private groups, comment-to-receive resources,\nor \"like/follow/comment to get\" mechanics, keep those as Brief requirements to\nconfirm, but do not put them into the publishable Xiaohongshu version. Rewrite\nthem into a safer platform-specific alternative or mark them as needing brand\nconfirmation.\n\nFor Brand Briefs, default to \"公开价值优先、产品名后置、无导流动作\". The content\nshould not start as a brand slogan. Translate the product into a user problem,\nscene, checklist, comparison, story, or method first, then let the product\nappear as support.\n\n## Input Contract\n\nMinimum useful input:\n\n- the Brief text, screenshot, document, or a pasted summary\n- target platform, default Xiaohongshu if not provided\n- creator/account direction if this is for a specific creator\n\nHelpful optional input:\n\n- profile link or recent posts of the creator\n- product page, brand page, or official reference material\n- required selling points\n- forbidden words and compliance notes\n- deliverables: graphic note, spoken video, Vlog, article, multi-platform\n- required keywords, hashtags, CTA, coupon, landing page, or comment guidance\n- desired tone and reference examples\n- deadline and brand review rounds\n\nIf the user only gives a screenshot or short sentence, do a light Brief intake\nfirst and ask only for missing route-changing fields, such as platform, format,\nrequired selling point, or account direction.\n\n## Workflow\n\n### 1. Brief Intake\n\nExtract the Brief into a structured table:\n\n| Layer | What To Extract |\n| --- | --- |\n| Brand / product | what is being promoted |\n| Campaign goal | awareness, seeding, conversion, trial, store visit, app download, course signup, lead generation |\n| Target user | who should care and who should not be targeted |\n| Product value | features, benefits, proof points, price, scenario, differentiator |\n| Mandatory points | required wording, keywords, scenes, CTA, links, tags |\n| Forbidden zone | banned claims, sensitive terms, must-not-say, competitor limits |\n| Deliverables | platform, format, duration/pages/word count, number of posts, timeline |\n| Brand tone | premium, friendly, professional, playful, local, practical, emotional |\n| Creator fit | why this creator can say it credibly |\n| Missing info | what must be clarified before final delivery |\n\nWhen extracting mandatory CTA, separate:\n\n- Brand requested CTA\n- Xiaohongshu-safe CTA\n- Needs brand/legal confirmation\n\nIf the Brief is too vague, do not block the workflow. Produce a \"Brief\nclarification list\" with 3-5 missing items and a draft direction based on what\nis already known.\n\n### 2. Creator And Audience Fit\n\nBefore selecting topics, judge whether this ad can sit inside the user's account.\n\nCheck:\n\n- Does the product match the account's audience?\n- Will it attract the desired customer or only random views?\n- Is the creator's usual content format able to hold the product?\n- Will this damage trust if pushed too hard?\n- Can the product be shown as a useful tool, scenario, story, checklist,\n transformation, comparison, tutorial, review, or life detail?\n\nGood diagnosis:\n\n这个 Brief 不能直接按品牌卖点写。它要先变成你账号用户关心的问题，再把产品放进去解决那个问题。\n\n### 3. Public Reference Search\n\nUse `research-scope-guard.md` before expanding online lookup.\n\nSearch should not only search the brand name. Search a mix of:\n\n- product category\n- user pain\n- use scenario\n- desired outcome\n- audience identity\n- competitor/public category wording\n- platform-specific content format, such as \"测评\", \"避坑\", \"清单\", \"教程\",\n \"通勤\", \"新手\", \"办公室\", \"妈妈\", \"自媒体\", \"AI工具\", \"本地生活\"\n\nDefault first round:\n\n- 3-5 keywords\n- recent public notes when the category changes fast\n- prioritize Xiaohongshu unless the user names another platform\n- select 3-5 reference notes or accounts, not a long list\n\nFor each selected reference, capture:\n\n- title and direct link\n- public signal: likes, saves, comments, publish time when available\n- content type: review, tutorial, Vlog, list, comparison, story, problem-solve\n- why users click\n- how the product/category is embedded\n- what can be borrowed\n- what not to copy\n\nDo not claim to have read comments or full copy unless those details were\nactually opened.\n\n### 4. Topic And Angle Matrix\n\nTurn the Brief into content angles before writing.\n\nRecommended angle types:\n\n| Angle | Use When | Example Shape |\n| --- | --- | --- |\n| Pain-first | user already has a clear problem | \"为什么你总是...\" |\n| Scenario-first | product solves a daily scene | \"上班/旅行/带娃/做账号时...\" |\n| Result-first | product creates a visible result | \"我用它把...\" |\n| Tutorial | product has steps or workflow | \"3 步完成...\" |\n| Comparison | category has alternatives | \"A 和 B 到底差在哪\" |\n| Checklist | user saves for later | \"新手先看这 5 点\" |\n| Story/Vlog | creator identity is strong | \"我为什么开始...\" |\n| Myth-busting | market has misunderstanding | \"很多人以为...其实...\" |\n| Local/life scene | city/store/food/travel category | \"第一次来...怎么选\" |\n\nRank angles by:\n\n- user click reason\n- brand message fit\n- creator credibility\n- production difficulty\n- compliance risk\n- save/comment potential\n- whether it can become a series\n\nDefault output should recommend Top 3 angles, with one首推.\n\n### 5. Content Package\n\nAfter the angle is selected, produce the actual deliverable.\n\nFor Xiaohongshu graphic note:\n\n- 3 title options, not 10\n- cover copy and cover type\n- 4-7 page structure\n- page-by-page copy direction\n- 300-character body copy\n- 10 publishing keywords\n- CTA/comment guidance\n- brand-mandatory-point checklist\n\nFor spoken video:\n\n- 3 title options\n- first 3 seconds hook\n- 60-120 second spoken script or the requested length\n- screen/subtitle emphasis\n- product placement point\n- body caption\n- 10 publishing keywords\n\nFor Vlog:\n\n- storyboard by scene\n- where the product appears naturally\n- narration outline\n- caption\n- cover direction\n- 10 publishing keywords\n\nFor cross-platform:\n\n- first finish the Xiaohongshu core angle\n- then route to `mother-content-cross-platform-distribution.md` for WeChat\n public account, Moments, Knowledge Planet, Bilibili, Douyin, X, or podcast\n\n### 6. Brand Delivery Check\n\nBefore final answer, include a delivery checklist:\n\n- mandatory points included\n- forbidden claims avoided\n- public value appears before product/brand selling language\n- Xiaohongshu risk gate passed or risky CTA rewritten\n- platform disclosure/compliance reviewed\n- product placement is natural\n- title and cover still have user click reason\n- first 3 lines / first 3 seconds do not sound like a brand slogan\n- CTA matches the Brief as much as possible without站外引流、加微信、诱导评论互动\n- missing brand assets or facts to confirm\n\nIf the Brief conflicts with creator trust, say so plainly:\n\n这条可以做，但不能按 Brief 原话硬写。原话更像品牌自夸，用户会滑走。我建议保留品牌必须表达的点，但把开头改成用户痛点/场景，再把产品放在解决方案里。\n\n## Output Forms\n\n### Light Brief Breakdown\n\nUse when the user only asks \"帮我看看这个 Brief\":\n\n1. Brief 摘要\n2. 这单适不适合这个账号\n3. 用户会关心的入口\n4. 3 个可做选题\n5. 需要向品牌确认的问题\n6. 是否需要继续搜索对标\n\n### Standard Brief To Content Package\n\nUse when the user wants actual content:\n\n1. Brief 拆解表\n2. 账号/受众适配判断\n3. 搜索范围和对标选择\n4. Top 3 内容角度\n5. 首推角度的完整小红书内容包\n6. 品牌交付检查表\n7. 下一步：发给品牌前检查标题/封面/正文，或生成图片\n\n### Deep Campaign Package\n\nUse when the user asks for a campaign, batch content, or multi-platform plan:\n\n1. campaign goal and user journey\n2. keyword/search plan\n3. benchmark notes/accounts\n4. 5-10 topic pool\n5. 3 complete deliverables\n6. multi-platform distribution plan\n7. review workflow and post-publish metrics\n\n## Research Scope Wording\n\nIf public lookup is needed, use this wording:\n\n我可以先基于 Brief 做本地拆解；如果你想让我看最近小红书同类产品/同类痛点都怎么讲，我会进入公开内容搜索。建议先搜 3-5 个关键词，找 3-5 条近期参考，再产出内容角度和正文。你确认后我再开始查。\n\nIf the user already asks for \"找对标\" or \"看看最近都怎么讲\", proceed after\nthe normal scope confirmation.\n\n## Good Style\n\nUse human, practical language:\n\n- 这不是把 Brief 翻译成小红书，而是把品牌卖点翻译成用户愿意看的内容入口。\n- 品牌要的是卖点完整，用户要的是跟自己有关。我们要在中间搭桥。\n- 这条广告不能从品牌口号开始，要从用户正在发生的场景开始。\n- 先别急着写正文，先判断这个产品应该进入用户的哪一个问题。\n\nAvoid:\n\n- pure slogan copy\n- fake personal experience\n- unsupported claims\n- claiming a product is best/official/guaranteed without evidence\n- copying reference notes\n- hiding that a post is commercial when disclosure is required\n\nFile v0.1.106:playbooks/comparable-account-breakdown-report-template.md\n\n# Lingzao Comparable Account Breakdown Report Template\n\nUse this template when the user chooses B / says this is someone else's Xiaohongshu account / wants to learn from, imitate, or benchmark another creator.\n\nThe goal is not to praise the account. The goal is to tell the user:\n\n- whether this account is worth learning from\n- what exactly can be learned\n- what cannot be copied\n- whether the account matches the user's current stage, resources, face/camera ability, product, industry, or content direction\n- which early-path signals matter more than the account's current mature form\n- how to adapt the account into the user's own version\n\nIf the task is to find benchmark accounts rather than analyze a user-provided\naccount, use `benchmark-account-discovery-quality-gate.md` first. Do not\nrecommend account-search results as final benchmarks until recent public posts,\nupdate status, recent high-performing works, and stage fit have been checked.\n\nFor formal or deep comparable-account reports, also apply\n`account-report-evidence-visual-contract.md`. That contract defines the\nevidence links, real-cover audit, viral-asset reuse, account-evolution, and\nvisual-report delivery baseline.\n\n## Output Form\n\nDefault light deliverable:\n\n1. Chat: short decision summary.\n\nDefault deep-report deliverable:\n\n1. HTML preview: browser-friendly visual report for quick reading and review.\n2. Word document: official shareable deliverable that the user can send to friends, clients, or team members.\n\nWhen a full comparable-account report is generated, create both HTML and Word whenever tooling is available. They must come from the same report source. Word is the official shareable deliverable; HTML is the browser preview. If only one format can be created, explain why and prefer Word.\n\nDo not generate a full Word/HTML report by default after one light lookup. First give the short decision summary. If the user wants a full report, explain that it becomes a deeper comparable-account breakdown and may require more Lingzao searches because it needs to inspect more notes and possibly deeper content.\n\nUser-facing upgrade explanation:\n\n如果你只想先判断这个账号值不值得学，我可以先按轻量拆解给你结论；如果你想生成正式报告，就会进入深度拆解。我会同时给你 HTML 预览和 Word 文档：HTML 方便你先看结构，Word 方便你转发给朋友、客户或团队。深度拆解会多看它的近期内容、代表爆款、封面标题、内容结构、可学和不可照抄部分，必要时还会继续看单篇正文、评论区或同阶段对标。\n\n范围上也先说清楚：轻量版只看一个主页的近期内容；正式报告可以用 20 条作品做标准深度解析，或用 40 条作品做更深的报告。如果还要打开单篇详情或评论区，我会先确认新增范围，不会直接替你扩大搜索。\n\nThe deep report should clearly tell the user what they will get:\n\n- 一页总览\n\nArchive v0.1.105: 55 files, 619457 bytes\n\nFiles: agents/openai.yaml (3992b), agents/openclaw.yaml (273b), assets/lingzao-logo.png (292012b), index.md (16835b), playbooks/account-report-evidence-visual-contract.md (9858b), playbooks/atian-creator-judgment-framework.md (10985b), playbooks/audience-persona-fit-check.md (5699b), playbooks/beginner-account-start-and-topic-radar.md (25119b), playbooks/benchmark-account-discovery-quality-gate.md (23742b), playbooks/brand-brief-to-content-workflow.md (11836b), playbooks/comparable-account-breakdown-report-template.md (16279b), playbooks/content-knowledge-base-workflow.md (27596b), playbooks/copy-paste-prompt-scope-boundary.md (10036b), playbooks/creator-case-general-analysis-framework.md (14268b), playbooks/draft-rewrite-and-benchmark-workflow.md (13577b), playbooks/image-generation-agent-integration-guide.md (20364b), playbooks/image-generation-execution-workflow.md (11935b), playbooks/keyword-insight-report-template.md (10946b), playbooks/keyword-to-publishable-content-package.md (41649b), playbooks/lingzao-progressive-interaction-map.md (42396b), playbooks/monetization-path-judgment-library.md (10874b), playbooks/mother-content-cross-platform-distribution.md (7847b), playbooks/post-publish-data-review-workflow.md (12962b), playbooks/pre-publish-readiness-check.md (7212b), playbooks/product-judgment-and-feedback-loop.md (8756b), playbooks/publishing-keyword-design-check.md (9503b), playbooks/reference-image-graphic-note-workflow.md (5291b), playbooks/research-scope-guard.md (5065b), playbooks/retention-and-follow-up-loop.md (12830b), playbooks/router-cases.json (3817b), playbooks/router-index.json (21352b), playbooks/self-account-diagnosis-report-template.md (27899b), playbooks/self-account-peer-horizontal-diagnosis.md (13270b), playbooks/single-note-breakdown-workflow.md (20268b), playbooks/track-difficulty-judgment-library.md (12236b), playbooks/travel-handdrawn-map-visual-workflow.md (7458b), playbooks/visual-generation-and-cover-workflow.md (11172b), playbooks/visual-reference-style-library.md (23833b), playbooks/wechat-benchmark-fit-and-writing-workflow.md (10567b), playbooks/weekly-content-motherpack-distributor.md (9459b), playbooks/xhs-content-compliance-risk-gate.md (5886b), playbooks/xhs-operation-task-tree.md (12978b), playbooks/xhs-platform-management-risk-baseline.md (5688b), playbooks/xhs-profile-bio-design.md (7960b), playbooks/xhs-title-design-check.md (7825b), playbooks/zero-beginner-onboarding-gate.md (9225b), scripts/check_playbook_router.py (5785b), scripts/check_version.py (275b), scripts/configure.py (3015b), scripts/lingzao_client.py (99047b), scripts/setup.sh (1824b), skill-card.md (3354b), SKILL.md (50787b), VERSION (8b), _meta.json (128b)\n\nArchive v0.1.101: 59 files, 687646 bytes\n\nFiles: agents/openai.yaml (3768b), agents/openclaw.yaml (273b), assets/lingzao-logo.png (292012b), index.md (13874b), playbooks/account-report-evidence-visual-contract.md (9850b), playbooks/atian-creator-judgment-framework.md (10985b), playbooks/audience-persona-fit-check.md (5699b), playbooks/beginner-account-start-and-topic-radar.md (25119b), playbooks/benchmark-account-discovery-quality-gate.md (23882b), playbooks/brand-brief-to-content-workflow.md (11826b), playbooks/comparable-account-breakdown-report-template.md (16341b), playbooks/content-knowledge-base-workflow.md (27605b), playbooks/copy-paste-prompt-scope-boundary.md (10027b), playbooks/creator-case-general-analysis-framework.md (14268b), playbooks/draft-rewrite-and-benchmark-workflow.md (13578b), playbooks/image-generation-agent-integration-guide.md (20328b), playbooks/image-generation-execution-workflow.md (11960b), playbooks/keyword-insight-report-template.md (11769b), playbooks/keyword-to-publishable-content-package.md (41849b), playbooks/lingzao-progressive-interaction-map.md (43539b), playbooks/monetization-path-judgment-library.md (10874b), playbooks/mother-content-cross-platform-distribution.md (7847b), playbooks/post-publish-data-review-workflow.md (12966b), playbooks/pre-publish-readiness-check.md (7211b), playbooks/product-judgment-and-feedback-loop.md (8745b), playbooks/publishing-keyword-design-check.md (9503b), playbooks/reference-image-graphic-note-workflow.md (5291b), playbooks/retention-and-follow-up-loop.md (12840b), playbooks/router-cases.json (3817b), playbooks/router-index.json (21349b), playbooks/search-credit-notice.md (14746b), playbooks/self-account-diagnosis-report-template.md (27895b), playbooks/self-account-peer-horizontal-diagnosis.md (13269b), playbooks/single-note-breakdown-workflow.md (20266b), playbooks/track-difficulty-judgment-library.md (12236b), playbooks/travel-handdrawn-map-visual-workflow.md (7456b), playbooks/visual-generation-and-cover-workflow.md (11171b), playbooks/visual-reference-style-library.md (23833b), playbooks/wechat-benchmark-fit-and-writing-workflow.md (10576b), playbooks/weekly-content-motherpack-distributor.md (9453b), playbooks/xhs-content-compliance-risk-gate.md (5886b), playbooks/xhs-operation-task-tree.md (12975b), playbooks/xhs-platform-management-risk-baseline.md (5686b), playbooks/xhs-profile-bio-design.md (7960b), playbooks/xhs-title-design-check.md (7825b), playbooks/zero-beginner-onboarding-gate.md (9219b), scripts/__pycache__/check_playbook_router.cpython-314.pyc (10505b), scripts/__pycache__/check_version.cpython-314.pyc (903b), scripts/__pycache__/configure.cpython-314.pyc (6461b), scripts/__pycache__/lingzao_client.cpython-314.pyc (142660b), scripts/check_playbook_router.py (5785b), scripts/check_version.py (275b), scripts/configure.py (3015b), scripts/lingzao_client.py (93931b), scripts/setup.sh (1824b), skill-card.md (2973b), SKILL.md (48434b), VERSION (8b), _meta.json (128b)\n\nArchive v0.1.100: 55 files, 622862 bytes\n\nFiles: agents/openai.yaml (3768b), agents/openclaw.yaml (273b), assets/lingzao-logo.png (292012b), index.md (13511b), playbooks/account-report-evidence-visual-contract.md (9850b), playbooks/atian-creator-judgment-framework.md (10985b), playbooks/audience-persona-fit-check.md (5699b), playbooks/beginner-account-start-and-topic-radar.md (25119b), playbooks/benchmark-account-discovery-quality-gate.md (23882b), playbooks/brand-brief-to-content-workflow.md (11826b), playbooks/comparable-account-breakdown-report-template.md (16341b), playbooks/content-knowledge-base-workflow.md (27605b), playbooks/copy-paste-prompt-scope-boundary.md (10027b), playbooks/creator-case-general-analysis-framework.md (14268b), playbooks/draft-rewrite-and-benchmark-workflow.md (13578b), playbooks/image-generation-agent-integration-guide.md (20328b), playbooks/image-generation-execution-workflow.md (11960b), playbooks/keyword-insight-report-template.md (11769b), playbooks/keyword-to-publishable-content-package.md (41849b), playbooks/lingzao-progressive-interaction-map.md (43539b), playbooks/monetization-path-judgment-library.md (10874b), playbooks/mother-content-cross-platform-distribution.md (7847b), playbooks/post-publish-data-review-workflow.md (12966b), playbooks/pre-publish-readiness-check.md (7211b), playbooks/product-judgment-and-feedback-loop.md (8745b), playbooks/publishing-keyword-design-check.md (9503b), playbooks/reference-image-graphic-note-workflow.md (5291b), playbooks/retention-and-follow-up-loop.md (12840b), playbooks/router-cases.json (3817b), playbooks/router-index.json (21349b), playbooks/search-credit-notice.md (14746b), playbooks/self-account-diagnosis-report-template.md (27895b), playbooks/self-account-peer-horizontal-diagnosis.md (13269b), playbooks/single-note-breakdown-workflow.md (20266b), playbooks/track-difficulty-judgment-library.md (12236b), playbooks/travel-handdrawn-map-visual-workflow.md (7456b), playbooks/visual-generation-and-cover-workflow.md (11171b), playbooks/visual-reference-style-library.md (23833b), playbooks/wechat-benchmark-fit-and-writing-workflow.md (10576b), playbooks/weekly-content-motherpack-distributor.md (9453b), playbooks/xhs-content-compliance-risk-gate.md (5886b), playbooks/xhs-operation-task-tree.md (12975b), playbooks/xhs-platform-management-risk-baseline.md (5686b), playbooks/xhs-profile-bio-design.md (7960b), playbooks/xhs-title-design-check.md (7825b), playbooks/zero-beginner-onboarding-gate.md (9219b), scripts/check_playbook_router.py (5785b), scripts/check_version.py (275b), scripts/configure.py (3015b), scripts/lingzao_client.py (101456b), scripts/setup.sh (1824b), skill-card.md (3190b), SKILL.md (48482b), VERSION (8b), _meta.json (128b)\n\nArchive v0.1.94: 50 files, 315980 bytes\n\nFiles: agents/openai.yaml (3619b), agents/openclaw.yaml (307b), index.md (10614b), playbooks/account-report-evidence-visual-contract.md (9850b), playbooks/atian-creator-judgment-framework.md (10985b), playbooks/audience-persona-fit-check.md (5699b), playbooks/beginner-account-start-and-topic-radar.md (25119b), playbooks/benchmark-account-discovery-quality-gate.md (23882b), playbooks/brand-brief-to-content-workflow.md (11826b), playbooks/comparable-account-breakdown-report-template.md (16341b), playbooks/content-knowledge-base-workflow.md (27605b), playbooks/copy-paste-prompt-scope-boundary.md (10027b), playbooks/creator-case-general-analysis-framework.md (14268b), playbooks/draft-rewrite-and-benchmark-workflow.md (13578b), playbooks/image-generation-agent-integration-guide.md (20328b), playbooks/image-generation-execution-workflow.md (11960b), playbooks/keyword-insight-report-template.md (11769b), playbooks/keyword-to-publishable-content-package.md (41849b), playbooks/lingzao-progressive-interaction-map.md (43284b), playbooks/monetization-path-judgment-library.md (10874b), playbooks/mother-content-cross-platform-distribution.md (7847b), playbooks/post-publish-data-review-workflow.md (12966b), playbooks/pre-publish-readiness-check.md (7211b), playbooks/product-judgment-and-feedback-loop.md (8745b), playbooks/publishing-keyword-design-check.md (9503b), playbooks/reference-image-graphic-note-workflow.md (5291b), playbooks/retention-and-follow-up-loop.md (12840b), playbooks/search-credit-notice.md (14746b), playbooks/self-account-diagnosis-report-template.md (27895b), playbooks/self-account-peer-horizontal-diagnosis.md (13269b), playbooks/single-note-breakdown-workflow.md (20266b), playbooks/track-difficulty-judgment-library.md (12236b), playbooks/travel-handdrawn-map-visual-workflow.md (7456b), playbooks/visual-generation-and-cover-workflow.md (11171b), playbooks/visual-reference-style-library.md (23833b), playbooks/weekly-content-motherpack-distributor.md (9453b), playbooks/xhs-content-compliance-risk-gate.md (5886b), playbooks/xhs-operation-task-tree.md (12751b), playbooks/xhs-platform-management-risk-baseline.md (5686b), playbooks/xhs-profile-bio-design.md (7960b), playbooks/xhs-title-design-check.md (7825b), playbooks/zero-beginner-onboarding-gate.md (9219b), scripts/check_version.py (275b), scripts/configure.py (3015b), scripts/lingzao_client.py (87941b), scripts/setup.sh (1824b), skill-card.md (3059b), SKILL.md (58504b), VERSION (7b), _meta.json (127b)\n\nArchive v0.1.93: 50 files, 315456 bytes\n\nFiles: agents/openai.yaml (3619b), agents/openclaw.yaml (307b), index.md (10149b), playbooks/account-report-evidence-visual-contract.md (9850b), playbooks/atian-creator-judgment-framework.md (10985b), playbooks/audience-persona-fit-check.md (5699b), playbooks/beginner-account-start-and-topic-radar.md (25119b), playbooks/benchmark-account-discovery-quality-gate.md (23882b), playbooks/brand-b...","readmeExcerpt":"Skill: 灵造 Owner: itxiaohao Summary: 跨平台创作者研究与自媒体运营 Skill Tags: latest:0.1.106 Version history: v0.1.106 | 2026-09-01T02:55:53.103Z | user Release Lingzao Skill 0.1.106: adds direct WeChat Channels share-link support for short-video copy extraction while preserving existing public research boundaries. v0.1.105 | 2026-08-20T05:43:54.807Z | user Release Lingzao Skill 0.1.105: adds complete Xiaohongshu image-note body im","codeSnippets":[],"executableExamples":[{"language":"bash","snippet":"python3 /scripts/check_playbook_router.py"},{"language":"bash","snippet":"bash \"/scripts/setup.sh\" --base-url \"https://your-lingzao-domain.com\""},{"language":"bash","snippet":"export LINGZAO_API_KEY=\"lgz_xxx\"\nexport LINGZAO_BASE_URL=\"https://your-lingzao-domain.com\""},{"language":"bash","snippet":"~/.lingzao/bin/lingzao doctor"},{"language":"bash","snippet":"~/.lingzao/bin/lingzao check-version"},{"language":"bash","snippet":"npx skills add https://assets-tian.midao.site/skills/lingzao --skill lingzao -g --copy"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: lingzao\ndescription: 灵造是给 WorkBuddy、OpenClaw、Codex 等 Agent 使用的跨平台创作者研究与自媒体运营 Skill。安装免费，可先做选题、标题、封面、账号诊断、发布检查、复盘，以及公众号对标适配判断和原创写作；查询公开内容、评论、短视频文案、公众号文章数据或生成图片时需要配置灵造 API Key。\n---\n\n# 灵造：跨平台创作者研究与自媒体运营 Skill\n\n灵造是一个主 Skill，不需要拆成标题、封面、账号诊断、图片生成等多个 Skill。\n安装后，WorkBuddy、OpenClaw、Codex 等 Agent 会先把你的问题路由到合适的\n创作者运营 playbook；只有当你需要查询公开内容、读取评论、提取短视频文案、\n查看公众号文章数据或生成图片时，才需要配置灵造 API Key。\n\n## 安装完成后的首次提示\n\n当当前对话刚刚完成灵造 Skill 的安装或更新时，只有确认安装成功后，才在最终回复中\n主动告诉用户一次下面的使用手册；安装失败、尚未验证成功或普通后续对话不要重复发送：\n\n> 灵造已安装完成。你可以查看《灵造功能使用手册》：\n> https://my.feishu.cn/docx/Y2HQdj5mzoFx4vxfij3cl9TRnjh?from=from_copylink\n> 快速了解灵造的功能和使用方法～\n\n## 从这里开始\n\n| 你现在想做 | 可以直接这样问 Agent |\n| --- | --- |\n| 找内容方向 | “用灵造帮我围绕这个关键词做小红书、抖音、TikTok、Instagram 或 YouTube 选题，给我 10 个可发方向。” |\n| 找对标账号 | “帮我找这个赛道值得学习的对标账号，并说明每个账号适合学什么。” |\n| 拆一条笔记或视频 | “分析这条内容为什么有效，拆成标题、封面、结构、评论需求和可复用模板。” |\n| 改标题和封面 | “基于我的草稿，给我 3 个最强标题和 5 个小红书封面方向。” |\n| 做发布前检查 | “发布前帮我检查标题、封面、前 3 行、关键词和用户点击理由。” |\n| 做发布后复盘 | “根据这条内容的数据和评论，帮我判断下次要调整什么。” |\n| 做每周内容包 | “用灵造把我这一周的素材整理成 5 个母题，并分发成小红书、公众号、播客和短口播。” |\n| 校准公众号对标 | “我发几篇喜欢的公众号文章和一篇自己的内容，你先判断适不适合我学，不适合再补找对标，然后帮我写成自己的文章。” |\n| 做图片素材 | “先帮我设计封面/配图方向；如果需要生成图片，再按我确认的方向生成。” |\n| 保存长结果 | “把这份分析整理成 Word、网页预览或知识库 Markdown 版本。” |\n\n## 免费能做什么\n\n不配置 API Key 时，灵造仍然可以作为创作者运营路由和 playbook 使用。适合：\n\n- 判断账号定位、赛道难度、内容主线和商业路径。\n- 设计小红书标题、封面方向、发布关键词和图文结构。\n- 改写草稿、拆解用户已经提供的内容材料、做发布前检查。\n- 根据用户提供的数据截图或复盘信息，输出下一步实验建议。\n- 把用户提供的一周素材整理成 5 个母题，并规划小红书、公众号、播客、\n 短口播、社群和知识库分发。\n- 把长分析整理成 Word、网页预览或知识库 Markdown 结构。\n\n## 什么时候需要 API Key\n\n当 Agent 需要让灵造服务实际查询或生成内容时，需要到\n 配置 API Key，包括：\n\n- 搜索小红书、抖音、TikTok、Instagram、YouTube 或视频号公开内容和公开创作者；用结果辅助关键词/选题扩展。\n- 查看创作者主页、近期公开内容、主页深度分析和对标账号证据。\n- 打开小红书、抖音、TikTok、Instagram、YouTube 或视频号单条公开内容详情，读取一级公开评论。\n- 打开公众号公开文章详情，查看公开文章数据，扩展相关文章。\n- 提取公开短视频口播文案、字幕或 transcript。\n- 根据提示词和参考图生成创作者封面、配图或海报素材。\n\n用户已给出明确任务且属于小范围时，直接执行；首轮最多做 5 次外部查询。\n如果需要扩大关键词、账号、内容详情、评论分页、文案提取或生图数量，说明新增的业务范围并请用户确认。\n\n## 调用公开数据工具前\n\n- 小红书、抖音、TikTok、Instagram 或 YouTube 内容链接：看内容用详情工具，看评论用评论工具，不要当主页链接。视频号 `/sph/` 分享链接可用于详情；评论必须先从详情取得纯数字内容 ID。\n- 小红书、抖音、TikTok、Instagram 或 YouTube 主页链接：普通主页查看或基础主页分析先用\n `get-user-posted-notes`；只有用户明确要粉丝数、简介、关注数、总获赞等主页资料时\n 才用 `get-user-info`；深度主页分析看 `analyze-user-profile`。\n- 用户只给昵称、账号名、抖音号或数字 ID 时，不要自己拼 URL；先用\n `search-users` 找创作者，再用返回的主页链接或 ID 调主页工具。\n- 视频号主页只接受 `search-users` 返回的 finder ID，不接受主页 URL，也不支持 `analyze-user-profile`。\n- 抖音主页工具需要可用的主页 URL 或 `search-users` 返回的 `MS4w...` 形式 ID；\n 视频短链看单条内容时直接用 `get-note-detail`，需要口播、字幕或 transcript 时才用\n `extract-video-copy`；不要把视频短链用于主页分析。\n- YouTube 主页工具只接受 `search-users` 返回的 channel ID 或 `/channel/UC...`\n URL；不要把 `@handle`、`/c/` 或 `/user/` 直接传给主页工具，也不要自动解析。\n- TikTok 主页工具接受 canonical `https://www.tiktok.com/@handle` 或\n `search-users` 返回的 ID；单条内容接受 canonical `/@handle/video/`、\n `/@handle/photo/` 或显式 `--platform tiktok --note-id `。不要传\n `vm.tiktok.com`/`vt.tiktok.com` 短链或裸 `@handle`。\n- TikTok V1 不支持 `analyze-user-profile`。需要主页资料和近期内容时，按需分别调用\n `get-user-info` 与 `get-user-posted-notes`，不要隐藏组合调用。\n- Instagram 主页工具接受 canonical `https://ww"},{"path":"_meta.json","content":"{\n \"ownerId\": \"kn72byzxyctvxnvxm8a4d9qmvx8892za\",\n \"slug\": \"lingzao\",\n \"version\": \"0.1.106\",\n \"publishedAt\": 1788231353103\n}"},{"path":"index.md","content":"# Lingzao Skill Index\n\n## Purpose\n\nThis folder contains the Lingzao Skill package for Agent runtimes. Lingzao helps\nAgents route creator-operation work, prepare creator research, run public\ncontent lookups when the user has configured access, and generate creator image\nassets within a confirmed task scope.\n\n## Package Files\n\n- `SKILL.md`: main Agent instructions and command guidance.\n- `VERSION`: current Skill package version.\n- `agents/`: Agent metadata.\n- `assets/lingzao-logo.png`: packaged brand icon used by Agent metadata.\n- `playbooks/`: creator-operation workflows used before answering.\n- `playbooks/router-index.json`: centralized route cards for every playbook.\n- `playbooks/router-cases.json`: representative user prompts and expected\n primary routes.\n- `scripts/`: setup, version check, configuration, and CLI command scripts.\n- `skill-card.md`: marketplace summary source.\n\n## Public Boundaries\n\n- Keep user-facing wording focused on Lingzao, creator research, and workflow\n support.\n- Do not promise viral growth, guaranteed monetization, full monitoring, bulk\n data export, or copying another creator's content.\n- Proceed directly for clear small online tasks. Before expanding keywords,\n accounts, details, comment pages, transcripts, profile depth, or image count,\n confirm only the added business scope.\n- Keep Agent instructions focused on task scope. The CLI owns user-visible\n service-result wording derived from structured server responses.\n- For service failures, use concise Lingzao retry language and include\n `error_id` only when it is returned.\n- Keep credentials, temporary local paths, and sensitive debug details out of\n user-facing output.\n\n## Release Checklist\n\nBefore publishing a new Skill package:\n\n1. Confirm `VERSION` is bumped when user-visible behavior changes.\n2. Keep `SKILL.md`, `agents/`, `playbooks/`, CLI output, and marketplace copy\n aligned.\n3. Run focused Skill CLI tests plus Python compile.\n4. Run project checks required by the release risk.\n5. Inspect the published package or marketplace file list after release.\n\n## Recent Version Notes\n\n- `0.1.106`: `extract-video-copy` now accepts canonical WeChat Channels\n `https://weixin.qq.com/sph/...` share links directly. Agents do not need a\n detail lookup first and must not attempt media download or decryption. This\n source-only package has not been published.\n\n- `0.1.105`: Douyin `get-note-detail` now documents valid HTTPS\n `v.douyin.com/` inputs alongside standard post URLs and numeric\n IDs. The top-level router uses detail for one-post content and reserves\n `extract-video-copy` for spoken copy, subtitles, or transcripts. Agents pass\n the original short URL to Lingzao without opening or expanding it first.\n Invalid formats ask for the original HTTPS share URL instead of automatic\n fallback or another paid detail attempt. This source-only package has not\n been published.\n\n- `0.1.104`: `extract-video-copy` now sends a fresh explicit operation ID for\n"},{"path":"playbooks/account-report-evidence-visual-contract.md","content":"# Lingzao Account Report Evidence And Visual Contract\n\nUse this contract whenever Lingzao produces a formal or deep account report:\n\n- own-account diagnosis\n- comparable-account breakdown\n- same-stage peer horizontal diagnosis\n- benchmark-account follow-up report\n- creator distillation report\n\nThis contract turns account analysis into a product-grade deliverable. It is\ninspired by strong open account-breakdown skills, but rewritten for Lingzao's\npositioning: Lingzao is an operation system, not only a report generator.\n\n## Core Principle\n\nDo not only summarize an account. A useful account report must answer:\n\n1. what this account is really built on\n2. which visible content assets already work\n3. how the account evolved or repeated its winning assets\n4. what the user can learn\n5. what the user must not copy\n6. what the user should do next\n\nThe report should feel like a deliverable, not a private memo and not a dense\nchat wall.\n\n## Delivery Level\n\n### Light Read\n\nUse light read when:\n\n- the user only asks \"值不值得学\"\n- the user provided one account and has not asked for a formal report\n- only a small homepage/recent-post sample is available\n- deeper research scope has not been confirmed\n\nOutput:\n\n- one short decision summary in chat\n- direct account/note links when available\n- one concrete next step\n- offer full Word / HTML / Feishu / knowledge-base packaging if they want a\n shareable report\n\n### Formal Report\n\nUse formal report when:\n\n- the user asks for 完整分析, 深度拆解, 正式报告, 可视化报告, Word, HTML, Feishu,\n or client-facing output\n- Lingzao has enough public note samples for a standard diagnosis or comparable\n report\n- the user has confirmed the deeper scope if more public lookups are needed\n\nDefault formal carriers:\n\n1. Word document when available: official shareable deliverable.\n2. HTML/webpage preview when available: browser-friendly preview.\n3. Knowledge-base-ready Markdown when the user wants to save/reuse it.\n\nIf artifact tooling is unavailable, output a complete Markdown report and say\nwhy Word/HTML was not produced. Do not call a rough chat answer a formal\nreport.\n\n## One-Screen Opening\n\nThe first page or chat summary should be readable in one minute:\n\n- report title\n- account name and direct profile link\n- report date and sample boundary\n- account category / positioning subtitle\n- one sentence diagnosis or one sentence worth-learning judgment\n- 3-4 visible public data points when available\n- strongest visible content asset\n- biggest current problem or biggest non-copyable condition\n- one concrete action the user should copy or test next\n\nUse evidence labels:\n\n- `确定结论`: directly supported by public page, note, cover, comment, or user\n provided data.\n- `合理推断`: supported by several visible signals, but not platform backend or\n official algorithm proof.\n\nDo not infer exposure, click-through rate, finish rate, traffic source, single\nnote follower conversion, sales, or private-domain conversion unless the user\nprovided backend data.\n\n## Link And Evide"},{"path":"playbooks/atian-creator-judgment-framework.md","content":"# A Tian Creator Judgment Framework\n\nThis file is bundled inside the Lingzao Agent plugin. It captures A Tian's creator-account operating judgment so the plugin is not just a list of prompts.\n\n## Core Judgment\n\nDo not only analyze what the account posted. Diagnose:\n\n- what stage the account is in\n- what the account is remembered for\n- who the content is for, who will click, and who is unlikely to click\n- whether the content has a stable audience/problem anchor\n- which posts were validated by data\n- whether the viral post can be repeated\n- what comments reveal about user demand\n- what should continue, reduce, or stop\n- what the next test should be\n\n## Account Memory Anchor\n\nThe user should be able to understand what the account does from:\n\n- account name\n- bio\n- first-screen covers\n- repeated title keywords\n- visual style\n- topic pattern\n\nIf the account feels like a personal feed with unrelated posts, diagnose missing memory anchor before giving advanced advice.\n\n## Audience Persona Anchor\n\nBefore advising topics, titles, keywords, or formats, identify the likely user\npersona. Use `audience-persona-fit-check.md` when this is unclear.\n\nJudge:\n\n- gender or identity: female-oriented, male-oriented, parents, students,\n workplace, local users, visitors, buyers\n- life stage: university, first job, 30+, 35+, married, with children,\n freelancer, business owner\n- city or location intent for local life\n- what they search, click, save, comment, or pay for\n- which audience should not be targeted by this note\n\nIf the user does not know their audience, ask for the accounts or notes they\nrecently liked, saved, searched, or want to imitate, then reverse-infer the\naudience before writing the strategy.\n\n## Stage Logic\n\n### 0-1 Beginner\n\nMain problem: direction and first validation.\n\nDo:\n\n- ask what they like, collect, know, own, or can consistently produce\n- ask about age/life stage, work/childcare/study/freelance status, and where most of their daily time goes\n- ask what they usually search, save, or learn on Xiaohongshu; saved content often reveals what they secretly want to do\n- map life clues into possible directions before recommending a niche\n- find low-follower viral notes and same-stage accounts\n- help them test 2-3 directions\n- give first 5 notes and 7-day execution plan\n\nDo not:\n\n- ask them to imitate 100k+ creators directly\n- give too many abstract positioning words\n- push mature commercial systems too early\n- assume they must do口播; graphic notes, lists, screenshots, AI-assisted notes, product tests, and learning records are valid beginner paths\n\nBeginner direction mining:\n\n- childcare/family -> 科学育儿、亲子陪伴、家庭教育、妈妈成长、儿童好物\n- workplace -> 职场成长、行业经验、办公效率、副业转型、35+女性职场\n- fashion/beauty/lifestyle -> 穿搭、化妆、护肤、普通人变美、生活方式\n- buying/product taste -> 好物分享、平价替代、真实测评、消费决策、工具推荐\n- travel/local life -> 本地生活、城市攻略、周末去哪、旅行路线、美食探店\n- learning/skills/tools -> 学习记录、技能教程、AI工具、读书笔记、普通人自我提升\n\nExpression-format judgment:\n\n- If phone or text clues show natural verbal expression and w"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"跨平台创作者研究与自媒体运营 Skill Skill: 灵造 Owner: itxiaohao Summary: 跨平台创作者研究与自媒体运营 Skill Tags: latest:0.1.106 Version history: v0.1.106 | 2026-09-01T02:55:53.103Z | user Release Lingzao Skill 0.1.106: adds direct WeChat Channels share-link support for short-video copy extraction while preserving existing public research boundaries. v0.1.105 | 2026-08-20T05:43:54.807Z | user Release Lingzao Skill 0.1.105: adds complete Xiaohongshu image-note body im","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":1973,"uniquenessScore":48,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-10T07:35:37.553Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-10T07:35:37.553Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-10T10:43:38.195Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}