Philosophy: The README is just the facade; the real value lies in Issues, Commits, and community discussions.

• Use

  web_search

  to search

  site:github.com <project_name>

  to confirm the full org/repo path
• Use

  search-layer

  (Deep Mode + Intent Awareness) to supplement community links and non-GitHub resources:
  bash

  python3 skills/search-layer/scripts/search.py \
    --queries "<project_name> review" "<project_name> evaluation user experience" \
    --mode deep --intent exploratory --num 5

• Use

  web_fetch

  to crawl the repo homepage for basic information (README, Stars, Forks, License, latest updates)

web_fetch

• README:

  curl -s -H "Authorization: token {PAT}" -H "Accept: application/vnd.github.v3.raw" "https://api.github.com/repos/{owner}/{repo}/readme"

• Repo Metadata:

  curl -s -H "Authorization: token {PAT}" "https://api.github.com/repos/{owner}/{repo}"

• Issues:

  curl -s -H "Authorization: token {PAT}" "https://api.github.com/repos/{owner}/{repo}/issues?state=all&sort=comments&per_page=10"

• Commits:

  curl -s -H "Authorization: token {PAT}" "https://api.github.com/repos/{owner}/{repo}/commits?per_page=10"

• File Tree:

  curl -s -H "Authorization: token {PAT}" "https://api.github.com/repos/{owner}/{repo}/git/trees/{branch}?recursive=1"

github.com/{org}/{repo}

web_fetch

github.com/{org}/{repo}/issues?q=sort:comments

browser

content-extract

web_fetch

content-extract

search-layer

python3 skills/search-layer/scripts/search.py --queries "<project> review" "<project> evaluation" --mode deep --intent exploratory --num 5

python3 skills/search-layer/scripts/search.py "<project> latest release" --mode deep --intent status --freshness pw --num 5

python3 skills/search-layer/scripts/search.py --queries "<project> vs <competitor>" "<project> alternatives" --mode deep --intent comparison --num 5

python3 skills/search-layer/scripts/search.py "<project> official docs" --mode fast --intent resource --num 3

python3 skills/search-layer/scripts/search.py "<project> discussion experience" --mode deep --intent exploratory --domain-boost reddit.com,news.ycombinator.com --num 5

factual

status

comparison

tutorial

exploratory

news

resource

Without

--intent

, the behavior is exactly the same as v1 (no scoring, output in original order).

web_search

web_fetch

content-extract

1. Domain Restrictions:

   mp.weixin.qq.com

   ,

   zhihu.com

   ,

   xiaohongshu.com

   .
2. Complex Structure: Pages contain a large number of formulas (LaTeX), complex tables, or the Markdown returned by

   web_fetch

   is extremely messy.
3. Content Missing:

   web_fetch

   returns empty content or a Challenge page due to anti-crawling measures.

python3 skills/content-extract/scripts/content_extract.py --url <URL>

• First checks the domain whitelist (WeChat/Zhihu, etc.), uses MinerU directly if matched
• Otherwise, first uses

  web_fetch

  for probing, falls back to MinerU-HTML if failed
• Returns a unified JSON contract (including fields like

  ok

  ,

  markdown

  ,

  sources

  )

• Project Phase: Early Experimentation / Rapid Growth / Mature & Stable / Maintenance Mode / Stagnant (based on commit frequency and content)
• High-Quality Issue Criteria: High number of comments, maintainer participation, exposes architecture issues, or contains valuable technical discussions
• Competitor Identification: Extract from the "Comparison"/"Alternatives" section of the README, Issue discussions, and web searches

1. Title Must Link to GitHub Repository (Format:

   # [Project Name](https://github.com/org/repo)

   , ensure clickable jump)
2. Uniform Empty Lines Around Titles (End of previous section → empty line → title → empty line → content, ensure clear visual separation)
3. Telegram Empty Line Fix (Mandatory): Telegram will swallow empty lines after list items (starting with

   -

   ). Solution: Insert a line of braille space

   ⠀

   (U+2800) between the end of the list block and the next title, format as follows:

   - Last item in list
   
   ⠀
   **Next Title**

   This ensures the empty line before the title is not swallowed during Telegram rendering.
4. All Titles Bold (emoji + bold text)
5. Competitor Comparison Must Include Links (GitHub / official website / documentation, at least one)
6. Community Volume Must Be Specific: Quote specific post/tweet/discussion content summaries, attach original links. Do not use general descriptions like "highly praised" or "very popular"; instead, write "someone said something" or "a certain post discussed specific issues"
7. Information Traceability Principle: All quoted external information should include the original link so readers can trace the source

undefined

• Stars: {Number} | Forks: {Number} | License: {Type}
• Team/Author: {Background}
• Commit Trend: {Recent activity + project phase judgment}
• Latest Updates: {Overview of recent important commits}

• vs GraphRAG — Difference description
• vs RAGFlow — Difference description

• DeepWiki: {Link or "Not Included"}
• Zread.ai: {Link or "Not Included"}

• @Username: "Specific content..."
• Discussion Thread: Discussed specific issues... {Note "No relevant discussions found" if none are available}

• Zhihu: Post Title — Discussed content
• V2EX: Post Title — Discussed content {Note "No relevant discussions found" if none are available}

undefined

• Prioritize using

  web_search

  +

  web_fetch

  , with

  browser

  as a fallback
• Search Enhancement: For project research tasks, default to

  search-layer

  v2 Deep Mode +

  --intent exploratory

  (Brave + Exa + Tavily three-source parallel deduplication + intent-aware scoring), single-source failure does not block the main process
• Mandatory Crawling Degradation: When

  web_fetch

  fails/returns 403/anti-crawling page/too short content, or the source domain belongs to high-risk sites (such as WeChat/Zhihu/Xiaohongshu): switch to

  content-extract

  (which internally falls back to MinerU-HTML) to get cleaner Markdown + traceable sources
• Collect from different sources in parallel to improve efficiency
• All links must be valid and accessible, do not fabricate URLs
• Output in Chinese, retain English for technical terms

• Title Link: In the format

  # [Project Name](GitHub URL)

  , clickable for jump
• Empty Lines Around Titles: Each bold title (

  **🎯 ...**

  ) has one empty line before and after
• Telegram Empty Line: There is a braille space

  ⠀

  line between the end of each list block and the next title (prevents Telegram from swallowing empty lines)
• Issue Links: Each selected Issue is in the complete format

  [#Number Title](Full URL)

• Competitor Links: Each competitor is attached with

  [Name](GitHub/Official Website Link)

• Community Volume Links: Each quote is in the format

  [Source: Title](URL)

• No Vague Descriptions: No general descriptions like "highly praised" or "very popular" in the community volume section
• Information Traceability: All external quotes are attached with original links