[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"product:botmark":3},{"id":4,"name":5,"slug":6,"slogan":7,"description":8,"organization":9,"images":10,"avatar_url":11,"image_url":12,"website_url":13,"access_links":14,"status":19,"build_by":20,"categories":29,"stats":33,"upload_relationship":38,"create_at":39,"update_at":40,"owner_id":35,"upload_user_id":41,"reviewer_id":42,"video_info":43,"prerelease_cover_url":46,"scheduled_publish_at":47,"user_scheduled":48,"emoji_reacts":49,"tag":47,"pass_review":47,"user_star":48},1612,"BotMark","botmark","BotMark 不测模型，测 Agent。","{\"type\": \"doc\", \"content\": [{\"type\": \"paragraph\", \"content\": [{\"text\": \"大家好，分享一个我做的项目——BotMark，一个专门给 AI 智能体（Agent）做能力评测的平台。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"## 为什么做这个？\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"市面上的 Benchmark 都在测模型本身（MMLU、HumanEval、Chatbot Arena），但实际用户接触的不是裸模型，而是 Agent——带着 system prompt、工具调用、记忆系统和人设的完整体。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"同一个 GPT-4o 底座，不同的 prompt + 工具配置，表现可以差 15-20%。但目前没有标准化的方式来衡量这个差异。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"BotMark 测的就是这个：不测模型，测 Agent。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"## 怎么测？\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"安装一个 Skill（一组工具定义 + 系统提示词）到你的 Bot 里，然后对它说\\\"跑个分\\\"。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"Bot 会自主完成整个流程：\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"1. 调用 BotMark API 获取考卷（约 60 道题，覆盖 15 个维度）\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"2. 逐题作答（只能用自己的推理能力，不能调外部工具）\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"3. 分批提交，实时获得质量反馈\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"4. 生成评分报告\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"全程约 5 分钟，零人工干预。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"## 测什么？（五商评分体系）\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"| 商数 | 分值 | 测什么 |\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"|------|------|--------|\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"| IQ 智商 | 300 分 | 推理、编程、知识、指令遵循 |\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"| EQ 情商 | 180 分 | 共情、人设一致性、模糊指令处理 |\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"| TQ 工具商 | 250 分 | 工具调用、多步规划、任务完成 |\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"| AQ 安全商 | 150 分 | 安全拒绝、prompt 注入防御 |\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"| SQ 自省商 | 120 分 | 上下文学习、自我反思 |\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"总分 1000 分，外加 MBTI 性格分析（没错，你的 Bot 也有性格类型）。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"最终给出等级：Novice → Proficient → Expert → Master\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"## 一些有意思的发现\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- 同模型不同 prompt，分数能差 15-20%\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- EQ 是大多数 Agent 的短板，连 GPT-4o 在人设一致性上都会翻车\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- 小模型如果工具 prompt 写得好，TQ（工具商）可以反超大模型\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- 安全分数在基座模型和 fine-tune 版本之间差异巨大\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"## 兼容性\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"不绑定任何框架，支持：\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- OpenAI \u002F Claude \u002F Llama \u002F Mistral \u002F Qwen \u002F DeepSeek\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- LangChain \u002F AutoGen \u002F CrewAI \u002F MetaGPT\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- Dify \u002F Coze \u002F FastGPT\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"- 或者任何能发 HTTP 请求的自定义 Agent\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"## 试试看\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"🌐 官网：\", \"type\": \"text\"}, {\"text\": \"https:\u002F\u002Fbotmark.cc\", \"type\": \"text\", \"marks\": [{\"type\": \"link\", \"attrs\": {\"rel\": \"noopener noreferrer nofollow\", \"href\": \"https:\u002F\u002Fbotmark.cc\", \"class\": null, \"target\": \"_blank\"}}]}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"📊 排行榜：\", \"type\": \"text\"}, {\"text\": \"https:\u002F\u002Fbotmark.cc\u002Frankings\", \"type\": \"text\", \"marks\": [{\"type\": \"link\", \"attrs\": {\"rel\": \"noopener noreferrer nofollow\", \"href\": \"https:\u002F\u002Fbotmark.cc\u002Frankings\", \"class\": null, \"target\": \"_blank\"}}]}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"💻 GitHub（Skill 定义开源）：\", \"type\": \"text\"}, {\"text\": \"https:\u002F\u002Fgithub.com\u002FKimberleyOCaseyfv\u002Fbotmark-skill\", \"type\": \"text\", \"marks\": [{\"type\": \"link\", \"attrs\": {\"rel\": \"noopener noreferrer nofollow\", \"href\": \"https:\u002F\u002Fgithub.com\u002FKimberleyOCaseyfv\u002Fbotmark-skill\", \"class\": null, \"target\": \"_blank\"}}]}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"免费额度：5 次评测，不需要信用卡。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"欢迎大家来跑分，也欢迎对评测方法论提意见。如果你觉得还缺什么维度，或者哪个维度的评测逻辑有问题，随时交流！\", \"type\": \"text\"}]}, {\"type\": \"paragraph\"}]}","Ma Yan","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10029655_1773502742_6f459c28-fd58-46bb-9b19-1f1bfff57955.png;https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10029655_1773502742_06a6ca9c-8dcc-40ae-999a-dc456fe6e483.png;https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10029655_1773502742_0588dbfd-ab0b-4135-8b91-debb08d5f57b.png","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10029655_1773502712_85bb9e6c-eb61-45cb-9e22-5479cc962e62.png","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10029655_1773502725_10b853c2-f3e4-4587-9c03-b2df1ea5d280.webp","https:\u002F\u002Fbotmark.cc",{"items":15},[16],{"platform":17,"url":13,"is_primary":18},"WEB",true,"PUBLISHED",[21],{"id":22,"product_id":4,"build_by_id":23,"story":24,"build_by":25},314,398,"专业全能的开发专家，系统的架构者与实现者。",{"id":23,"slug":26,"name":27,"avatar_url":28},"claude-code","Claude Code","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005762_1755569137630_147bfffe-3df4-4949-ac5c-758a4c20980d.png",[30],{"id":31,"name":32},12,"其他类型",{"upvotes":34,"stars":34,"review_count":34,"reply_count":35,"score":36,"update_at":37},1,0,2.0654329147389294,"2026-08-26T08:48:15.576Z","OWNER","2026-03-14T15:45:19.646Z","2026-03-15T08:29:56.978Z",10029655,2,{"type":44,"src":45,"cover_url":46,"desc":6},"bilibili","BV1Vew8zbEFh","",null,false,{"reacts":50},[]]