[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"pack-detail-ai-hr-recruiting-stack-zh":3,"seo:pack:ai-hr-recruiting-stack:zh":106},{"code":4,"message":5,"data":6},200,"操作成功",{"pack":7},{"slug":8,"icon":9,"tone":10,"status":11,"status_label":12,"title":13,"description":14,"items":15,"install_cmd":105},"ai-hr-recruiting-stack","🧑‍💼","#0891B2","stable","稳定","AI 招聘 + HR 工具包","十件给招聘官、HR 负责人把 AI 真的接到流程里的工具：找候选人、解析 + 筛简历、生成面试题、录音转写+总结、起草 offer letter、入职 — 任何决策落到候选人之前先跑一遍偏见审计。ATS 对接走 MCP，不是聊天框。",[16,28,38,48,56,65,73,80,87,97],{"id":17,"uuid":18,"slug":19,"title":20,"description":21,"author_name":22,"view_count":23,"vote_count":24,"lang_type":25,"type":26,"type_label":27},2818,"b2f4accc-45ba-44f3-9a1e-aea8c3096e0d","tavily-search-search-api-built-for-ai-agents","Tavily Search — Search API Built for AI Agents","Tavily Search returns LLM-ready answers from the web — not link lists. One call gets snippets, citations, optional generated answer. Free tier 1K\u002Fmo.","Tavily",309,0,"en","agent","Agent",{"id":29,"uuid":30,"slug":31,"title":32,"description":33,"author_name":34,"view_count":35,"vote_count":24,"lang_type":25,"type":36,"type_label":37},3607,"6c45e855-cfad-5617-b5fb-3660436a827a","apify-mcp-server-8-000-web-scrapers-for-agents","Apify MCP Server — 8,000+ Web Scrapers for Agents","Apify MCP Server connects agents to Apify Actors via a hosted endpoint (mcp.apify.com) or local run, turning thousands of web scrapers into callable tools.","MCP Hub",288,"mcp","MCP",{"id":39,"uuid":40,"slug":41,"title":42,"description":43,"author_name":44,"view_count":45,"vote_count":24,"lang_type":25,"type":46,"type_label":47},226,"a9cbbc61-0159-41a5-82a0-f44c24da8b55","jina-reader-convert-any-url-llm-ready-text-a9cbbc61","Jina Reader — Convert Any URL to LLM-Ready Text","Convert any URL to clean, LLM-friendly markdown with a simple prefix. Just prepend r.jina.ai\u002F to any URL. Handles JS-rendered pages, PDFs, and images. 10K+ stars.","Script Depot",7363,"skill","Skill",{"id":49,"uuid":50,"slug":51,"title":52,"description":53,"author_name":54,"view_count":55,"vote_count":24,"lang_type":25,"type":46,"type_label":47},866,"0d39058c-33e3-11f1-9bc6-00163e2b0d79","reactive-resume-ai-powered-open-source-resume-builder-0d39058c","Reactive Resume — AI-Powered Open-Source Resume Builder","Free open-source resume builder with AI integration. Supports Claude, GPT, Gemini for content generation. Drag-and-drop, PDF export, self-hostable, privacy-first. MIT, 36,000+ stars.","AI Open Source",493,{"id":57,"uuid":58,"slug":59,"title":60,"description":61,"author_name":44,"view_count":62,"vote_count":24,"lang_type":25,"type":63,"type_label":64},173,"443e86c2-3811-496e-8e4d-6eef742ab219","docling-document-parsing-ai-443e86c2","Docling — Document Parsing for AI","IBM document parsing library. Converts PDFs, DOCX, PPTX, images, and HTML into structured markdown or JSON. Built for RAG pipelines and LLM ingestion.",315,"script","Script",{"id":66,"uuid":67,"slug":68,"title":69,"description":70,"author_name":71,"view_count":72,"vote_count":24,"lang_type":25,"type":46,"type_label":47},2842,"91b1b2a3-8be3-42c3-9366-c71fe29ed30d","phoenix-evals-llm-as-judge-library-with-built-in-templates","Phoenix Evals — LLM-as-Judge Library with Built-in Templates","Phoenix Evals runs LLM-as-judge on traces or datasets. Pre-built templates: hallucination, relevance, toxicity, QA. Outputs scored DataFrames.","Arize AI",244,{"id":74,"uuid":75,"slug":76,"title":77,"description":78,"author_name":54,"view_count":79,"vote_count":24,"lang_type":25,"type":46,"type_label":47},2634,"0e1eab94-47b5-11f1-9bc6-00163e2b0d79","anarlog-open-source-ai-meeting-notes-stay-your-machine-0e1eab94","Anarlog — Open-Source AI Meeting Notes That Stay on Your Machine","A privacy-first, local-first meeting note application built with Rust and Tauri that transcribes, summarizes, and organizes your meetings without sending data to the cloud.",282,{"id":81,"uuid":82,"slug":83,"title":84,"description":85,"author_name":44,"view_count":86,"vote_count":24,"lang_type":25,"type":46,"type_label":47},270,"24576b2c-a9d1-4f7a-9696-b1e5c50a17f3","faster-whisper-4x-faster-speech-text-24576b2c","Faster Whisper — 4x Faster Speech-to-Text","Faster Whisper is a reimplementation of OpenAI Whisper using CTranslate2, up to 4x faster with less memory. 21.8K+ GitHub stars. GPU\u002FCPU, 8-bit quantization, word timestamps, VAD. MIT licensed.",374,{"id":88,"uuid":89,"slug":90,"title":91,"description":92,"author_name":93,"view_count":94,"vote_count":24,"lang_type":25,"type":95,"type_label":96},627,"11680977-685a-479a-acce-d2ecc762fe8f","prompt-perfect-system-prompt-engineering-templates-11680977","Prompt Perfect — System Prompt Engineering Templates","Battle-tested system prompt templates for building LLM personas, agents, and workflows. Structured formats for role definition, constraints, and output control. 4,000+ GitHub stars.","Prompt Lab",289,"prompt","Prompt",{"id":98,"uuid":99,"slug":100,"title":101,"description":102,"author_name":103,"view_count":104,"vote_count":24,"lang_type":25,"type":46,"type_label":47},4284,"9b79e28a-1720-4be1-9b10-4c6cedfadf68","claude-code-agent-ai-ethics-advisor-9b79e28a","Claude Code Agent: AI Ethics Advisor","AI ethics and responsible AI development specialist. Use when reviewing an AI system for bias, fairness violations, or regulatory compliance gaps; when generating a model card,...","TokRepo精选",138,"tokrepo install pack\u002Fai-hr-recruiting-stack",{"pageType":107,"pageKey":8,"locale":108,"title":109,"metaDescription":110,"h1":111,"tldr":112,"bodyMarkdown":113,"faq":114,"schema":130,"internalLinks":136,"citations":149,"wordCount":162,"generatedAt":163},"pack","zh","AI 招聘 + HR 工具包 — 10 件工具把 AI 接到招聘漏斗里","Tavily \u002F Apify \u002F Jina Reader \u002F Reactive Resume \u002F Docling \u002F Phoenix Evals \u002F Anarlog \u002F Faster Whisper \u002F Prompt Perfect \u002F AI Ethics Advisor — 给招聘官的一套 agent 链：找人、筛简历、面试准备、录音转写、起 offer、入职，外加偏见审计闸门。","AI 招聘 + HR 工具包 — 招聘官的 agent 漏斗","十个工具按招聘漏斗顺序：找人 → 筛简历 → 面试 → 发 offer → 入职。每一步在结果落到候选人之前都先跑一遍偏见审计。ATS 对接走 MCP，不是聊天框替代品。","## 这个 pack 包含什么\n\n这是招聘官或 HR 负责人真的会拿来跑一次完整招聘的工具集 — 不是 50 家 vendor demo day 的清单。每个工具只干漏斗里的一件事：找候选人、解析简历、按 JD 筛选、面试准备、录音、起 offer、入职。还有一个工具横跨每一步：在结果给到 hiring manager 之前先跑一遍偏见审计。\n\n整套**故意 agent 驱动**。招聘官花一个上午把流程搭起来；之后 agent 干粗活 — boolean 搜索、简历格式化、筛选总结、转写笔记 — 招聘官只在需要判断的地方介入（电面本身、谈判、close）。**关键铁律：不允许自动 reject**。每一步筛选都产出**带原因**的排序，而不是悄悄把候选人筛掉、人都看不到。\n\n## 推荐安装顺序\n\n1. **Tavily Search** — 给 sourcing agent 用的搜索引擎。可以查「2026 年发过 Server Components 博客的资深 React 工程师」「柏林 staff PM 的薪资基准」「Acme 上个月为啥裁了 growth team」。免费档每月 1000 次查询，小团队够用。\n2. **Apify MCP Server** — 8,000+ 现成 scraper（LinkedIn 风格 profile、招聘网站、GitHub、Stack Overflow Careers）以 MCP 工具暴露。Tavily 的文本片段不够、需要**结构化**候选人列表去重排序的时候用它。\n3. **Jina Reader** — `https:\u002F\u002Fr.jina.ai\u002F\u003Curl>` 把任何网页转成干净 markdown。最朴实的主力：贴一个候选人个人站、竞品的招聘页、40 页的 benefits PDF，回来的就是 LLM 能消化的纯文本。\n4. **Reactive Resume** — 开源简历构造器，同时也是**解析器**。候选人提交原始简历后，先过 Reactive 导成 JSON Resume 标准 schema，再跑 LLM 筛选。每次都是同样的字段 = 可比的筛选信号。\n5. **Docling** — IBM 工业级文档解析器，吃 PDF \u002F DOCX \u002F 扫描简历。处理 Reactive 搞不定的真实烂数据（1990 年代扫描 CV、欧式双栏排版、只有图的 PDF）。输出是结构化 markdown，下游筛选 agent 能直接用。\n6. **Phoenix Evals** — LLM-as-judge 库，内置模板。**真正的筛选发生在这里**：定义评分表（相关经验年数、领域匹配度、沟通清晰度），Phoenix 用同一套 prompt 跑每个候选人，返回带理由的数字得分。可审计、可复现。\n7. **Anarlog** — 开源本地 AI 会议记录。在**招聘官自己的电脑上**录音 + 转写电面 — 候选人音频不出本机。输出是总结 + action items，可以直接塞进 ATS，不用上传给第三方 SaaS。\n8. **Faster Whisper** — 比 OpenAI Whisper 快 4 倍，本地跑。Anarlog 和你的批量面试流水线底层就是它。一周 20 个电面、需要分钟级而不是小时级出转写的时候上它。\n9. **Prompt Perfect** — 系统 prompt 模板库。用来把 offer letter prompt、拒信 prompt、reference check prompt 纳入版本管理。「友好通用语气 + 不带具体薪资 + 提示下一步」周一发的信和周五发的信应该是同一封。\n10. **Claude Code Agent: AI Ethics Advisor** — 闸门。任何 shortlist 给到 hiring manager 之前，Ethics Advisor 复查筛选评分表 + 排序结果，找受保护类别的代理变量（邮编、毕业学校、毕业年份、照片）。flag 退回招聘官，**绝不自动应用**。这是整条链上**唯一允许 block 下游动作**的环节。\n\n## 它们怎么协同\n\n```\n            ┌─ Tavily ─── Apify MCP ───┐\n            │ (搜索)     (爬虫)         │   找人\n            └─────────┬─────────────────┘\n                      ▼\n              Jina Reader (URL → 文本)\n                      │\n                      ▼\n         Reactive Resume ── Docling           筛简历\n         (JSON schema)    (烂格式 PDF)\n                      │\n                      ▼\n              Phoenix Evals\n           (LLM 评分表)\n                      │\n                      ▼ ────────────────────────┐\n              AI Ethics Advisor                 │\n             (偏见审计闸门)                     │\n                      │                         │\n                      ▼                         │\n         Anarlog + Faster Whisper        面试\n         (本地录音 + 转写)                       │\n                      │                         │\n                      ▼                         │\n              Prompt Perfect             Offer\n         (offer 信 \u002F 拒信 \u002F             + 入职\n          推荐人调用模板)\n```\n\n非显然的连接是 **Phoenix Evals → Ethics Advisor**：Phoenix 给你一份可辩护、可复现的评分表；Ethics Advisor 在**排序给到任何人之前**先检查这份评分表有没有代理变量。没有这道闸门，LLM-as-judge 流水线会悄悄把训练数据里的所有偏见再编码一遍。加上闸门，就有了纸面凭证。\n\n## 你会遇到的取舍\n\n- **Reactive Resume vs Docling** — Reactive 是可选的（候选人愿意用 builder 才有）；Docling 是必装的（你得解析任何收上来的东西）。两个都装：Reactive 用于候选人配合的干净 schema，Docling 用于那 40% 一上来就是 2014 年扫描 PDF 的烂格式。\n- **Anarlog（本地）vs 云端会议机器人** — Anarlog 让音频留在招聘官的笔记本上。云端机器人（Fireflies \u002F Otter）配置更快，但候选人音频会落到美国 vendor 的日志里，对欧洲候选人能不能过 GDPR 是模糊的。EU 候选人默认走本地。\n- **Phoenix Evals vs 人工读简历** — Phoenix 可复现、快；招聘官亲自读简历能捕捉评分表覆盖不了的漏斗顶部信号。正确的搭配是 Phoenix 跑第一轮（200 → 30），人读第二轮（30 → 8）。\n- **Ethics Advisor 自动应用 flag** — 别。Ethics Advisor 是**审查者**，不是执行者。模型 flag 了一个代理变量就自动拒掉候选人，恰恰就是你想避免的失败模式。flag 给招聘官，招聘官决定。\n\n## 常见踩坑\n\n- **评分表只在某个人脑子里** — Phoenix 要的是写下来的评分表。如果你说不出「相关领域经验 3 分、相邻领域 2 分、可迁移 1 分、不相关 0 分」，筛选就不可复现，偏见审计也抓不到任何东西。先把评分表写下来，再去搭 agent。\n- **候选人 PII 走第三方 LLM** — SaaS Claude \u002F OpenAI 接口会记日志。简历内容含姓名 \u002F 邮箱 \u002F 地址 \u002F 学校，**筛选这一步默认走本地模型**（Ollama + 12B Llama 变体），只把**评分结果**上行给云端。云端调用留给 offer letter，不留给筛选。\n- **ATS 自带「AI 集成」** — 大部分 ATS vendor 卖的是套了 UI 的 GPT 调用。本 pack 的意义在于 prompt \u002F 评分表 \u002F 审计链路在你自己手上，而不是外包到 vendor 锁死的界面里。用 ATS 的 MCP \u002F webhook 层，跳过捆绑的「AI 筛选」。\n- **拒信走 human-in-the-loop** — 评分表再完美，自动发拒信都是最快通向歧视投诉的路。每封拒信在离开公司之前都得过一个人。\n- **忘了到期清候选人数据** — 多数司法区对申请人数据的保留期是有上限的。给流水线加 cron，到点自动清简历 + 转写。别相信「我们手动会清」。",[115,118,121,124,127],{"q":116,"a":117},"这套在欧盟 \u002F 纽约市 \u002F 加州做招聘决策合法吗？","工具本身是中性的 — 合不合规取决于你怎么用。欧盟 AI Act 把招聘算法定为高风险：你得有文档、有人类监督、能解释单个决策。纽约市 Local Law 144 要求每年一次偏见审计 + 自动招聘决策工具使用时要告知候选人。加州的草案也朝这个方向走。Ethics Advisor + Phoenix Evals 的组合能产出这些法规要的审计链路，但前提是你真跑了 + 留了日志。上线前找你们的劳动法律顾问聊一下。",{"q":119,"a":120},"这套怎么对接现有 ATS（Greenhouse \u002F Lever \u002F Workable \u002F Moka \u002F 北森）？","走 MCP 或 webhook，不走聊天框。多数现代 ATS 都有 candidate \u002F application \u002F interview 的 REST API；把它包成一个 MCP server（或者用社区现成的），agent 调用方式跟调 Tavily、Apify 一样。别装 ATS vendor 捆绑的「AI 助手」 — 一装你就丢失了 prompt 和数据链路的控制权。ATS 当 system of record，agent 作为读 \u002F 排序 \u002F 写结构化 note 的层。",{"q":122,"a":123},"偏见审计具体在审什么？","AI Ethics Advisor 复查筛选评分表里有没有「跟受保护类别相关但不衡量真实岗位要求」的变量 — 常见代理变量是邮编（种族）、学校声望（阶级）、毕业年份（年龄）、就业 gap（家庭照护）、照片有无（什么都暴露）。还会把排序结果对照输入池查 disparate impact：申请池 40% 是女性、Top-20 里只有 10%，那就是一个值得人去复查评分表的 flag。它本身**不做决策**，也**不应该做**。",{"q":125,"a":126},"这周能跑起来的最小版本是哪几件？","三件：Docling（解析任何掉进收件箱的 PDF）、Phoenix Evals（一份写下来的评分表 + 每份简历一次 LLM 调用）、AI Ethics Advisor（在结果给到任何人之前先复查评分表和排序）。配齐大约一天，就是一个站得住脚的 AI 辅助筛选。一周 10+ 电面再加 Anarlog + Faster Whisper。Tavily + Apify 只有 sourcing 量真上来才装 — 小招聘团队大多数时候用不上 sourcing agent。",{"q":128,"a":129},"整套跑下来一个招聘团队一个月大概多少钱？","现实预算：30-100 美元 \u002F 月（小内招团队）。Tavily 免费档 1000 次查询；Apify 按量付费、典型爬虫量 5-30 美元 \u002F 月；Jina Reader 免费档够慷慨；Anarlog \u002F Faster Whisper \u002F Reactive Resume \u002F Docling 都是自托管 + 开源 = 0；Phoenix Evals 本身开源，但它发出去的 LLM 调用走你的 Claude \u002F OpenAI 账户（每周几百份简历预算 20-50 美元 \u002F 月）。Ethics Advisor 是 Claude Code subagent — 已经在用 Claude Code 就包含了。隐藏成本是写 + 版本化评分表和 prompt 模板的时间，每个岗位族系预算半天。",{"@context":131,"@type":132,"name":13,"description":133,"numberOfItems":134,"inLanguage":135},"https:\u002F\u002Fschema.org","ItemList","十个 AI 工具组成招聘漏斗：找人、解析简历、LLM-as-judge 筛选、面试录音转写、起 offer letter、入职，结果给候选人之前过一道偏见审计闸门。",10,"zh-CN",[137,141,145],{"url":138,"anchor":139,"reason":140},"\u002Fzh\u002Ftopics","浏览其他角色 pack","还有财务 \u002F 法务 \u002F PM \u002F 设计师等多个角色的 AI 工具包",{"url":142,"anchor":143,"reason":144},"\u002Fzh\u002Fai-tools-for\u002Fautomation","AI Agent 自动化工具集","ATS 的 MCP 对接器在更大的自动化目录里",{"url":146,"anchor":147,"reason":148},"\u002Fzh\u002Ffeatured","TokRepo 精选资产","这十件属于精选 agent-ready 资产目录",[150,154,158],{"claim":151,"source_name":152,"source_url":153},"Reactive Resume 是开源的 AI 友好简历构造器","Reactive Resume GitHub","https:\u002F\u002Fgithub.com\u002FAmruthPillai\u002FReactive-Resume",{"claim":155,"source_name":156,"source_url":157},"Docling 是开源的 PDF \u002F Office 文档解析器","Docling GitHub","https:\u002F\u002Fgithub.com\u002FDS4SD\u002Fdocling",{"claim":159,"source_name":160,"source_url":161},"Phoenix Evals 提供可复现的 LLM-as-judge 评分表模板","Phoenix Evals 文档","https:\u002F\u002Fdocs.arize.com\u002Fphoenix\u002Fevaluation\u002Fllm-evals",905,"2026-05-22T14:30:00Z"]