exa-unified-research

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Exa Unified Research

统一Exa研究

Progressive Disclosure Router

渐进式披露路由

Read only the minimum required reference files:
  • Code examples / API docs / debugging:
    references/code-search.md
  • Company and market research:
    references/company-research.md
  • People and profile research:
    references/people-research.md
  • SEC/earnings/filings:
    references/financial-report-search.md
  • Academic papers:
    references/research-paper-search.md
  • Blogs/portfolios:
    references/personal-site-search.md
  • X/Twitter discussion and sentiment:
    references/tweet-search.md
For mixed requests, load only the relevant subset and merge results at the end.
仅读取必要的参考文件:
  • 代码示例/API文档/调试:
    references/code-search.md
  • 企业与市场调研:
    references/company-research.md
  • 人物与档案调研:
    references/people-research.md
  • SEC/收益/备案文件:
    references/financial-report-search.md
  • 学术论文:
    references/research-paper-search.md
  • 博客/作品集:
    references/personal-site-search.md
  • X/Twitter讨论与舆情:
    references/tweet-search.md
对于混合请求,仅加载相关子集并在最后合并结果。

Global Tool Restriction

全局工具限制

Only use:
  • get_code_context_exa
  • web_search_advanced_exa
Do not use
web_search_exa
or other Exa tools unless the user explicitly overrides this skill policy.
仅使用:
  • get_code_context_exa
  • web_search_advanced_exa
除非用户明确覆盖本Skill策略,否则不得使用
web_search_exa
或其他Exa工具。

Global Token Isolation

全局Token隔离

Avoid large raw Exa outputs in main context. Use task/sub-agent delegation when available:
  • Run Exa calls inside sub-task
  • Deduplicate and normalize before returning
  • Return only distilled output to main context
If sub-agent is unavailable, run narrow iterative queries and summarize aggressively.
避免在主上下文中出现大量原始Exa输出。 在可用时使用任务/子Agent委托:
  • 在子任务中运行Exa调用
  • 返回前进行去重和标准化处理
  • 仅向主上下文返回提炼后的输出
如果子Agent不可用,则运行精准的迭代查询并进行深度总结。

Global Query Strategy

全局查询策略

  • Include explicit domain terms, entities, versions, and constraints in query
  • Use 2-3 query variations for recall-sensitive tasks
  • Merge and deduplicate by canonical URL/entity key
  • Tune
    numResults
    by user intent instead of hardcoding:
    • "a few": 10-20
    • "comprehensive": 50-100
    • explicit number: follow user request
  • 在查询中包含明确的领域术语、实体、版本和约束条件
  • 对召回敏感的任务使用2-3种查询变体
  • 按标准URL/实体键合并并去重
  • 根据用户意图调整
    numResults
    参数,而非硬编码:
    • "少量":10-20
    • "全面":50-100
    • 明确数字:遵循用户要求

Default Output Contract

默认输出约定

Return:
  1. Structured results (table or compact JSON)
  2. Source URLs with one-line relevance
  3. Notes on uncertainty/conflicts and coverage gaps
返回内容:
  1. 结构化结果(表格或紧凑JSON格式)
  2. 附带一行相关性说明的源URL
  3. 关于不确定性/冲突以及覆盖缺口的说明