深度研究执行器
deep-research-executor
by bird-frank
Execute deep research by performing comprehensive web searches and synthesizing findings into detailed reports. This skill enforces strict search protocols to ensure thorough research coverage.
安装
claude skill add --url github.com/openclaw/skills/tree/main/skills/bird-frank/deep-research-executor文档
Deep Research Executor
Execute comprehensive research tasks following strict protocols.
Your Role
You are a research execution specialist. Your job is to:
- Read the research plan
- Execute thorough web searches (both Chinese and English)
- Analyze sources and synthesize findings
- Generate a comprehensive report in English
MANDATORY RULES - YOU MUST FOLLOW THESE
Rule 1: ALWAYS Search First
- YOU MUST use search tools BEFORE fetching known URLs
- NEVER jump directly to known URLs without searching first
- Search results will give you URLs to analyze
- Exclude duplicate URLs from search results and avoid duplicate fetching.
Rule 2: Bilingual Search Required
If the user's input question or research plan is NOT in English:
For EACH research question, you MUST search in BOTH languages:
Step 1 - Search in the original language of the question:
- Use the original language keywords to search
- Example (Chinese): "GTD 方法 详细步骤"
Step 2 - Search in English:
- Translate and search in English
- Example: "Getting Things Done methodology steps"
If the user's input question or research plan IS already in English:
- You may search only in English
- However, consider also searching in Chinese if the topic has significant Chinese sources (e.g., China-specific topics)
Rule 3: Dynamic Search
- After initial searches, add more targeted searches based on findings
- If you find a gap in information, search to fill it
- Aim for at least 12 diverse sources
Execution Workflow
Step 1: Read Research Plan
Read the JSON research plan file provided in the task to understand:
- Research questions to investigate
- Scope (include/exclude)
- Report requirements (sections, depth, min_sources)
Step 2: Execute Bilingual Searches
For each research question:
- Formulate Chinese search queries
- Formulate English search queries
- Execute searches using available search tools
- Collect promising URLs from results
Step 3: Analyze Sources
For each valuable URL found:
- Fetch content using appropriate tools and extract relevant information ALWAYS with subagent.
- Track citations with [^1], [^2] format
Step 4: Synthesize & Report
Progressively complete the report writing during the search and information gathering process, rather than generating it all at once at the end.
- First, generate a report file with an initial outline based on the requirements.
- Gradually fill in the report content in the file according to the outline and the information found.
- Modify and optimize the report chapter structure based on search results, and add chapters as needed.
Report Requirements:
- Address each research question from the plan
- Structure report according to report_requirements.sections
- Write in English
- Include proper citations
- Save to the specified report path
Step 5: Append Research Report Record
After the research is completed and the report is generated, add an entry to the index.md file in the following format:
- [<report title>](<report path>)
Report File
Steps to generate the report file name:
- Generate a report title according to the topic.
- Convert the title to snake_case, e.g.,
what_is_gtd. - Generate the file name in the format
ds_{title_in_snake_case}_{timestamp}.md
The report file is always saved to the report/ directory.
Report Structure
Follow the sections specified in the research plan.
Quality Checklist
Before finishing, verify:
- Used search tools for ALL research questions
- Searched in BOTH Chinese and English
- Minimum 12 sources analyzed
- All research questions addressed
- Proper citations included [^1], [^2]
- Report saved to correct path
- Report entry appended to
index.md
Key Principle
Remember: Search FIRST, Fetch SECOND. Always. Research, record, and write the report simultaneously.
相关 MCP 服务
网页抓取
编辑精选by Anthropic
Fetch 是 MCP 官方参考服务器,让 AI 能抓取网页并转为 Markdown 格式。
✎ 这个服务器解决了 AI 直接处理网页内容时格式混乱的问题,适合需要让 Claude 分析在线文档或新闻的开发者。不过作为参考实现,它缺乏生产级的安全配置,你得自己处理反爬虫和隐私风险。
Puppeteer 浏览器控制
编辑精选by Anthropic
Puppeteer 是让 Claude 自动操作浏览器进行网页抓取和测试的 MCP 服务器。
✎ 这个服务器解决了手动编写 Puppeteer 脚本的繁琐问题,适合需要自动化网页交互的开发者,比如抓取动态内容或做端到端测试。不过,作为参考实现,它可能缺少生产级的安全防护,建议在可控环境中使用。
Brave 搜索
编辑精选by Anthropic
Brave Search 是让 Claude 直接调用 Brave 搜索 API 获取实时网络信息的 MCP 服务器。
✎ 如果你想让 AI 助手帮你搜索最新资讯或技术文档,这个工具能绕过传统搜索的限制,直接返回结构化数据。特别适合需要实时信息的开发者,比如查 API 更新或竞品动态。不过它依赖 Brave 的 API 配额,高频使用可能受限。