Back to Details

scholar-research

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Scholar Research

Scholar Research

Turn a research question into an auditable evidence set and an answer-first report.
将研究问题转化为可审计的证据集和以答案为核心的报告。

Start

启动

Clarify the topic, audience, geography, language, time window, platforms, and desired deliverable from the request. Infer low-risk omissions and record them. Never invent access to a platform: list unavailable or excluded sources in the report.
Create a resumable run:
bash
python3 <skill-dir>/scripts/scholar.py init --topic "<topic>" --output <workspace>/research-runs
Read references/method.md before collecting. Read only the platform sections needed from references/platforms.md. For repeat or scheduled monitoring, also read references/monitoring.md.
明确请求中的主题、受众、地域、语言、时间范围、平台以及期望交付物。推断低风险的遗漏信息并记录下来。切勿虚构平台访问权限：在报告中列出无法访问或被排除的来源。
创建可恢复的运行任务：
bash
python3 <skill-dir>/scripts/scholar.py init --topic "<topic>" --output <workspace>/research-runs
在收集数据前阅读references/method.md。仅从references/platforms.md中阅读所需的平台相关章节。如需重复或定时监测，还需阅读references/monitoring.md。

Collection posture

收集准则

Prefer, in order:
  1. Existing user-provided exports, URLs, files, or authenticated browser sessions.
  2. Public first-party search, feeds, pages, and documented endpoints.
  3. Ordinary HTTP/browser collection with conservative concurrency and caching.
  4. Optional local crawlers when scale or JavaScript rendering warrants them.
  5. Paid APIs or scraping services only after explaining the access gap and asking for credentials or spend authorization.
Do not bypass authentication, CAPTCHAs, paywalls, access controls, or platform safeguards. Do not collect private profiles or unnecessary personal data. Stop on explicit denial, repeated throttling, or a platform instruction prohibiting the attempted access method; record the gap and continue with other sources.
优先遵循以下顺序：
  1. 用户提供的现有导出文件、URL、文件或已认证的浏览器会话。
  2. 公开的第一方搜索结果、信息流、页面及已文档化的端点。
  3. 采用保守并发和缓存机制的常规HTTP/浏览器收集方式。
  4. 当需要处理大规模数据或JavaScript渲染场景时，可选用本地爬虫。
  5. 仅在说明访问缺口并获取凭证或支出授权后，才可使用付费API或爬虫服务。
不得绕过认证、验证码、付费墙、访问控制或平台防护措施。不得收集私人资料或不必要的个人数据。若遭遇明确拒绝、重复限流或平台禁止尝试的访问方式，需停止操作并记录缺口，然后继续使用其他来源。

Evidence records

证据记录

Write one JSON object per observed item to 
items.jsonl
. Preserve the source URL and observed facts; keep interpretation in 
notes
or the final report. Use the helper where practical:
bash
python3 <skill-dir>/scripts/scholar.py add <run>/items.jsonl \
  --platform <platform> --url <url> --title <title> \
  --published-at <ISO-8601> --observed-at <ISO-8601> \
  --metrics likes=10 comments=2 views=300 --text <summary>
Engagement values are platform-specific snapshots, not comparable audience counts. Store query-result pages only as discovery evidence; open the underlying source before using it to support a factual claim.
为每个观测项编写一个JSON对象并写入
items.jsonl
。保留来源URL和观测事实；将解读内容放入
notes
或最终报告中。尽可能使用辅助工具：
bash
python3 <skill-dir>/scripts/scholar.py add <run>/items.jsonl \
  --platform <platform> --url <url> --title <title> \
  --published-at <ISO-8601> --observed-at <ISO-8601> \
  --metrics likes=10 comments=2 views=300 --text <summary>
互动数值是特定平台的快照，不可作为可比较的受众数量。仅将查询结果页面作为发现证据存储；在使用其支持事实主张前，需打开底层来源。

Synthesis and stopping

合成与终止

Periodically generate the working report:
bash
python3 <skill-dir>/scripts/scholar.py report <run>
Use trend scores to prioritize reading, never as proof of importance. Synthesize recurring themes, emerging signals, disagreements, audience language, and actionable opportunities. Distinguish observed evidence, cross-source inference, verified claims, and open uncertainty.
Stop when the requested deadline or cap is reached, every in-scope platform is covered or documented as unavailable, and the latest collection round adds no material theme or changes no conclusion. For open-ended requests, default to two collection passes and at least two independent sources for each major conclusion; extend only while new evidence materially changes the answer.
定期生成工作报告：
bash
python3 <skill-dir>/scripts/scholar.py report <run>
使用趋势分数优先阅读内容，但切勿将其作为重要性的证明。综合反复出现的主题、新兴信号、分歧点、受众语言及可行动机会。区分观测证据、跨来源推断、已验证主张和未解决的不确定性。
当达到请求的截止日期或上限、覆盖所有范围内的平台或记录了无法访问的平台、且最新一轮收集未添加任何实质性主题或改变结论时，终止操作。对于开放式请求，默认进行两轮收集，且每个主要结论至少有两个独立来源；仅当新证据实质性改变答案时，才延长收集时间。

Deliverable

交付物

Update 
report.md
with an executive answer, scope and method, strongest trends, platform differences, evidence links, opportunities, caveats, and uncovered gaps. Cite the direct item or authoritative source beside each claim. State exact collection dates and do not describe a trend as current outside the observed window.
更新
report.md
，包含执行摘要、范围与方法、最显著的趋势、平台差异、证据链接、机会、注意事项及未覆盖的缺口。在每个主张旁引用直接项或权威来源。明确说明收集日期，不得在观测范围外将趋势描述为当前趋势。