When to Use
Baidu requests need routing before execution because the brand covers search, maps, knowledge, cloud, and AI surfaces that behave differently across mainland China and global contexts. Use this when the user needs help choosing, comparing, researching, implementing, or de-risking something inside the Baidu ecosystem.
Do not use it as a generic China-market skill or as a substitute for a narrower product skill when the task is already limited to one endpoint or one console workflow.
Architecture
Memory lives in ~/baidu/. If ~/baidu/ does not exist, run setup.md. See memory-template.md for structure.
CODEBLOCK0
Quick Reference
Use the smallest relevant file for the current task.
| Topic | File | Use it for |
|---|
| Setup flow and activation behavior | INLINECODE4 | Initialize local state and activation rules |
| Memory schema and status values |
memory-template.md | Create baseline local files |
| Baidu product-family routing |
ecosystem-map.md | Decide whether the task is about search, maps, knowledge, or Qianfan |
| Mainland versus global constraints |
mainland-vs-global.md | Check language, region, and rollout assumptions |
| Search, AI Search, Baike, and Wenku workflows |
search-knowledge.md | Handle research and source-comparison tasks |
| Qianfan and Baidu AI Cloud planning |
qianfan-cloud.md | Route model, platform, and implementation questions |
| Source verification ladder |
source-validation.md | Rank sources before trusting them |
| Final delivery checklist |
execution-checklist.md | Review risk, scope, and approval boundaries before answering |
Baidu Surfaces
Use this matrix before recommending tools or workflows. If two surfaces are active, split the answer and name the owner or approval boundary for each one.
| Surface | Typical user goal | Primary official homes | Route here when |
|---|
| Search and discovery | Find pages, trends, official statements, or current web information | INLINECODE12 | The task starts as web search or SERP interpretation |
| Knowledge and long-form content |
Retrieve encyclopedia-style context or long documents |
baike.baidu.com,
wenku.baidu.com | The user needs explanation, background, or document-style material |
| Maps and local data | Geocoding, routing, nearby search, or China-local navigation context |
map.baidu.com,
lbsyun.baidu.com | The task is location-aware or map-platform specific |
| AI Cloud and Qianfan | Model choice, AI platform design, agents, or cloud implementation |
cloud.baidu.com,
qianfan.cloud.baidu.com | The task is about LLMs, AI workflows, or Baidu cloud capabilities |
| Corporate and ecosystem research | Understand Baidu as a company, business unit, or partner |
baidu.com, investor and product pages | The task is strategic, vendor-related, or ecosystem-wide |
Requirements
- - No credentials required for planning, research, or source verification
- Account-specific execution may require user-approved access to Baidu Maps Open Platform, Baidu AI Cloud, Qianfan, or other Baidu consoles
- Never ask the user to paste passwords, SMS codes, cookies, refresh tokens, or private access keys into chat
Data Storage
Save only durable context that improves later Baidu work:
- - product surfaces the user actually uses
- approved regions, languages, and source preferences
- repeated compliance, localization, or rollout constraints
- account labels the user explicitly wants remembered
- final decisions with open risks and follow-up items
Do not store secrets, copied console tokens, billing exports, or raw customer data.
Core Rules
1. Lock the Baidu Surface First
- - Start by classifying the request into one primary surface: search, knowledge, maps, AI Cloud and Qianfan, or corporate research.
- If a request spans multiple surfaces, split the answer into separate workstreams instead of forcing one Baidu narrative over everything.
- Most bad Baidu guidance starts with the word "Baidu" staying vague for too long.
2. Make Mainland Versus Global Assumptions Explicit
- - Every non-trivial recommendation must state whether it targets mainland China, Chinese-language research, or a broader international workflow.
- Also state the assumed docs language and whether Chinese-first sources are acceptable.
- Product availability, search results, maps coverage, and documentation quality shift sharply by region and language.
3. Prefer Official Baidu Sources, Then Reconcile Translation Drift
- - Start with Baidu-owned docs or product pages for capabilities, limits, pricing, launch requirements, and platform behavior.
- If Chinese and English material diverges, record the mismatch and prefer the fresher or more specific source.
- Never present third-party English summaries as stronger evidence than current Baidu documentation.
4. Separate Public Research From Account Execution
- - Planning, source comparison, and workflow design are safe defaults.
- Any step that would log in, change cloud resources, submit map keys, or touch billing must wait for explicit user approval and the correct account context.
- Never assume search, maps, Qianfan, and cloud administration share the same owner or permission boundary.
5. Treat Qianfan and Baidu AI Cloud as Their Own Platform
- - Do not explain Qianfan by analogy alone to OpenAI, AWS, or another cloud vendor.
- Call out model access, region assumptions, account boundaries, and deployment tradeoffs before recommending architecture.
- If the user only wants consumer-facing Baidu search behavior, keep cloud guidance out of the answer.
6. Read Search and Knowledge Surfaces as Signals, Not Truth
- - Baidu Search, AI Search, Baike, and Wenku have different evidence quality and intent signals.
- Use search for discovery, Baike for encyclopedia-style orientation, Wenku for document-style leads, and official docs for final verification.
- Do not treat ranking order, reposted documents, or scraped mirrors as proof.
7. End With a Decision Record
- - Finish non-trivial tasks with a concise decision record: chosen surface, rejected paths, assumptions, hard blockers, and what still needs human confirmation.
- Save only that durable decision state under
~/baidu/. - This keeps future Baidu work consistent instead of repeating the same ambiguity.
Common Traps
- - Treating Baidu Search, Baike, Wenku, Maps, and Qianfan as one interchangeable product family -> wrong routing and wrong evidence
- Assuming English summaries mirror current Chinese documentation -> stale or simplified advice
- Recommending account-changing steps before clarifying the owning team -> avoidable permission and billing mistakes
- Using Baidu Search ranking as if it were proof -> discovery is not verification
- Assuming mainland and global use cases share the same defaults -> region-specific blockers appear late
- Mixing consumer-web research with AI Cloud implementation -> the answer loses the real job to be done
External Endpoints
| Endpoint | Data Sent | Purpose |
|---|
| https://www.baidu.com | Query text and page requests | Search, discovery, and product-overview verification |
| https://baike.baidu.com |
Query text and page requests | Encyclopedia-style topic orientation |
| https://wenku.baidu.com | Query text and page requests | Document-style reference discovery and long-form materials |
| https://map.baidu.com | Query text and page requests | Consumer map and local-search verification |
| https://lbsyun.baidu.com | Query text and page requests | Baidu Maps Open Platform documentation and API checks |
| https://cloud.baidu.com | Query text and page requests | Baidu AI Cloud documentation, pricing, and product verification |
| https://qianfan.cloud.baidu.com | Query text and page requests | Qianfan platform documentation and model workflow verification |
No other endpoints should be used unless the user explicitly approves additional sources or account-specific execution.
Security & Privacy
Data that may leave your machine:
- - search terms and page requests sent to Baidu-owned product or documentation pages
- optional comparison requests sent to approved supporting sources when official docs are incomplete
Data that stays local:
- - activation preferences and work history under INLINECODE21
- saved region defaults, trusted-source notes, and decision records in local markdown files
- account labels and approval boundaries only if the user explicitly wants them remembered
This skill does NOT:
- - store secrets in local markdown files
- treat search ranking as verified truth
- log into Baidu consoles without explicit user approval
- rewrite its own skill definition files
Trust
This skill depends on Baidu-owned websites and any approved supporting sources used for verification.
Only install and use it if you trust those services with your research queries and planning workflow.
Scope
This skill ONLY:
- - routes Baidu-related requests to the correct product family
- compares Baidu options using official-source verification
- plans Baidu Search, Maps, knowledge, and Qianfan workflows with region-aware assumptions
- keeps lightweight local memory in INLINECODE22
This skill NEVER:
- - invent product capabilities or region availability
- treat unofficial summaries as final evidence
- request passwords, QR-login tokens, cookies, or raw secrets
- execute account-changing actions without explicit user approval
Related Skills
Install with
clawhub install <slug> if user confirms:
- -
maps - Deeper routing and geospatial workflows beyond Baidu surface selection - INLINECODE25 - Competitive framing and ecosystem analysis outside one vendor
- INLINECODE26 - Add thresholds, status rules, and recurring checks
- INLINECODE27 - Compare another major China platform when vendor tradeoffs matter
- INLINECODE28 - Inspect specific pages after the Baidu routing pass
Feedback
- - If useful: INLINECODE29
- Stay updated: INLINECODE30
何时使用
百度请求在执行前需要进行路由,因为该品牌涵盖搜索、地图、知识、云和AI等多个领域,这些领域在中国大陆和全球环境中的表现各不相同。当用户需要在百度生态系统中进行选择、比较、研究、实施或降低风险时,请使用此技能。
当任务已局限于单一端点或单一控制台工作流时,请勿将其用作通用中国市场技能或替代更细分的产品技能。
架构
记忆文件存储在 ~/baidu/ 目录下。如果 ~/baidu/ 目录不存在,请运行 setup.md。结构请参见 memory-template.md。
text
~/baidu/
|- memory.md # 激活规则、默认界面和首选输出
|- accounts.md # 已知账户、项目和审批边界
|- regions.md # 中国大陆、跨境和语言默认设置
|- sources.md # 官方文档、可信翻译和弱来源说明
- - decisions.md # 最终建议、假设和未决风险
快速参考
为当前任务使用最小的相关文件。
| 主题 | 文件 | 用途 |
|---|
| 设置流程和激活行为 | setup.md | 初始化本地状态和激活规则 |
| 记忆模式和状态值 |
memory-template.md | 创建基线本地文件 |
| 百度产品系列路由 | ecosystem-map.md | 判断任务涉及搜索、地图、知识还是千帆 |
| 中国大陆与全球约束 | mainland-vs-global.md | 检查语言、区域和发布假设 |
| 搜索、AI搜索、百科和文库工作流 | search-knowledge.md | 处理研究和来源比较任务 |
| 千帆和百度AI云规划 | qianfan-cloud.md | 路由模型、平台和实施相关问题 |
| 来源验证阶梯 | source-validation.md | 在信任来源前对其进行排序 |
| 最终交付清单 | execution-checklist.md | 在回答前审查风险、范围和审批边界 |
百度界面
在推荐工具或工作流之前使用此矩阵。如果两个界面同时激活,请拆分答案并为每个界面指定负责人或审批边界。
| 界面 | 典型用户目标 | 主要官方主页 | 何时路由至此 |
|---|
| 搜索与发现 | 查找页面、趋势、官方声明或当前网络信息 | baidu.com | 任务以网络搜索或SERP解读开始 |
| 知识与长内容 |
检索百科风格上下文或长文档 | baike.baidu.com, wenku.baidu.com | 用户需要解释、背景或文档风格材料 |
| 地图与本地数据 | 地理编码、路线规划、附近搜索或中国本地导航上下文 | map.baidu.com, lbsyun.baidu.com | 任务与位置相关或针对地图平台 |
| AI云与千帆 | 模型选择、AI平台设计、智能体或云实施 | cloud.baidu.com, qianfan.cloud.baidu.com | 任务涉及大语言模型、AI工作流或百度云能力 |
| 企业与生态系统研究 | 了解百度作为公司、业务单元或合作伙伴 | baidu.com,投资者和产品页面 | 任务具有战略性、供应商相关或涉及整个生态系统 |
要求
- - 规划、研究或来源验证无需凭据
- 特定账户的执行可能需要用户批准访问百度地图开放平台、百度AI云、千帆或其他百度控制台
- 切勿要求用户在聊天中粘贴密码、短信验证码、Cookie、刷新令牌或私有访问密钥
数据存储
仅保存可改善后续百度工作的持久上下文:
- - 用户实际使用的产品界面
- 已批准的区域、语言和来源偏好
- 重复出现的合规、本地化或发布约束
- 用户明确希望记住的账户标签
- 包含未决风险和后续事项的最终决策
不存储机密信息、复制的控制台令牌、账单导出或原始客户数据。
核心规则
1. 首先锁定百度界面
- - 首先将请求分类到一个主要界面:搜索、知识、地图、AI云与千帆,或企业研究。
- 如果请求涉及多个界面,将答案拆分为独立的工作流,而不是强行用一个百度叙事覆盖所有内容。
- 大多数糟糕的百度指导都始于百度一词长时间保持模糊。
2. 明确说明中国大陆与全球假设
- - 每个重要的建议都必须说明其目标是中国大陆、中文研究,还是更广泛的国际工作流。
- 同时说明假定的文档语言以及是否接受中文优先的来源。
- 产品可用性、搜索结果、地图覆盖范围和文档质量因区域和语言而有显著差异。
3. 优先使用百度官方来源,然后调和翻译偏差
- - 对于功能、限制、定价、发布要求和平台行为,从百度自有文档或产品页面开始。
- 如果中英文材料存在差异,记录不匹配之处,并优先选择更新或更具体的来源。
- 切勿将第三方英文摘要作为比当前百度文档更强的证据。
4. 将公开研究与会话执行分开
- - 规划、来源比较和工作流设计是安全的默认操作。
- 任何涉及登录、更改云资源、提交地图密钥或触及计费的步骤,都必须等待用户明确批准和正确的账户上下文。
- 切勿假设搜索、地图、千帆和云管理共享相同的负责人或权限边界。
5. 将千帆和百度AI云视为独立平台
- - 不要仅通过类比OpenAI、AWS或其他云供应商来解释千帆。
- 在推荐架构之前,明确指出模型访问、区域假设、账户边界和部署权衡。
- 如果用户只需要面向消费者的百度搜索行为,请将云指导排除在答案之外。
6. 将搜索和知识界面视为信号,而非事实
- - 百度搜索、AI搜索、百科和文库具有不同的证据质量和意图信号。
- 使用搜索进行发现,使用百科进行百科风格定位,使用文库获取文档风格线索,使用官方文档进行最终验证。
- 不要将排名顺序、转载文档或抓取镜像视为证据。
7. 以决策记录结束
- - 对于非平凡任务,以简洁的决策记录结束:选择的界面、被拒绝的路径、假设、硬性障碍以及仍需人工确认的内容。
- 仅将该持久决策状态保存在 ~/baidu/ 下。
- 这使未来的百度工作保持一致,而不是重复相同的模糊性。
常见陷阱
- - 将百度搜索、百科、文库、地图和千帆视为一个可互换的产品系列 -> 错误的路由和错误的证据
- 假设英文摘要反映了当前的中文文档 -> 过时或简化的建议
- 在明确负责团队之前推荐更改账户的步骤 -> 可避免的权限和计费错误
- 将百度搜索排名视为证据 -> 发现不等于验证
- 假设中国大陆和全球用例共享相同的默认设置 -> 区域特定障碍在后期出现
- 将消费者网络研究与AI云实施混为一谈 -> 答案失去了真正的任务目标
外部端点
| 端点 | 发送的数据 | 用途 |
|---|
| https://www.baidu.com | 查询文本和页面请求 | 搜索、发现和产品概览验证 |
| https://baike.baidu.com |
查询文本和页面请求 | 百科风格主题定位 |
| https://wenku.baidu.com | 查询文本和页面请求 | 文档风格参考发现和长材料 |
| https://map.baidu.com | 查询文本和页面请求 | 消费者地图和本地搜索验证 |
| https://lbsyun.baidu.com | 查询文本和页面请求 | 百度地图开放平台文档和API检查 |
| https://cloud.baidu.com | 查询文本和页面请求 | 百度AI云文档、定价和产品验证 |
| https://qianfan.cloud.baidu.com | 查询文本和页面请求 | 千帆平台文档和模型工作流验证 |
除非用户明确批准其他来源或特定账户执行,否则不应使用其他端点。
安全与隐私
可能离开您机器的数据:
- - 发送到百度自有产品或文档页面的搜索词和页面请求
- 当官方文档不完整时,发送到已批准支持来源的可选比较请求
保留在本地数据:
- - ~/baidu/ 下的激活偏好和工作历史
- 本地Markdown文件中保存的区域默认设置、可信来源说明和决策记录
- 仅当用户明确希望记住时的账户标签和审批边界
此技能不会:
- - 在本地Markdown文件中存储机密信息
- 将搜索排名视为已验证的事实
- 未经用户明确批准登录百度控制台
- 重写自身的技能定义文件
信任
此技能依赖于百度拥有的网站以及用于验证的任何已批准支持来源。
仅当您信任这些服务处理您的研究查询和规划工作流时,才安装和使用它。
范围
此技能仅:
- - 将百度相关请求路由到正确的产品系列
- 使用官方来源验证比较百度选项
- 规划百度搜索、地图、知识和千帆工作流,并带有区域感知假设
- 在 ~/baidu/ 中维护轻量级本地记忆
此技能绝不