General-purpose search and content extraction skill, providing two core capabilities:Search Mode(multi-source parallel search + result synthesis) andExtraction Mode(URL content extraction + structured analysis).
Tool Overview
This skill orchestrates the following categories of tools, combining them as needed:
| Category | Tool | Description |
|---|---|---|
| Built-in Search | WebSearch |
DuckDuckGo search, always available |
| MCP Search | MCP search tools available in the current environment | Chinese search, specialized search, etc., automatically selected based on availability |
| Technical Documentation | ctx7 CLI |
Official documentation lookup for programming libraries/frameworks |
| Browser | agent-browser CLI |
Access search engines, page interaction (native Rust, token-efficient) |
| Content Extraction | crwl CLI → agent-browser |
Fallback by priority |
For specific invocation syntax, see references/tool-catalog.md。
Intent Detection
This skill handles the identification and routing of all URLs and search intents. Automatically determines based on user input:
URL Detection
| Situation | Handling |
|---|---|
| MessageContains only a URL(no other text, or only URL + spaces/punctuation) | Go directly to Fetch Mode |
| Message contains URL + reading intent ("read this", "analyze", "take a look at this") | Go directly to Fetch Mode |
| Message contains URL + save intent ("save", "clip", "save", "archive") | Fetch Mode After completion, automatically trigger the knowledge base skill's Clipping Save |
| URL appears in a longer message, intent unclear | Ask: "Do you want me to extract the content for analysis, or just the link?" |
| URL is clearly utilitarian (localhost, intranet address, file path) | Do not trigger |
Search Detection
| Situation | Handling |
|---|---|
| User asks a question, research topic, needs to find information | Search Mode |
| Provide URL + "search for more similar ones" | Fetch first, then Search |
| Delegated invocation by other skills | Execute according to the mode specified by the delegating party |
Search Mode
Multi-source parallel search, synthesized and deduplicated, then output structured results.
Step 1 — Query Classification
Determine the type based on query content, select tool combination:
| Query Type | Basis for Judgment | Tool Combination |
|---|---|---|
| Technical Documentation | Involves programming libraries, APIs, framework usage | ctx7(library → docs)+ WebSearch |
| General Query | General questions | WebSearch + MCP Search in parallel (Chinese queries prioritize MCP Search) |
| Deep Research | User explicitly requests comprehensive/in-depth investigation | All tools run in parallel, including agent-browser accessing Google/Bing |
Step 2 — Parallel search execution
Based on the classification result, withinthe same round of responsecall multiple search tools in parallel. For tool invocation details, see references/tool-catalog.md。
Key principles:
- Independent tools are called in parallel without waiting
- context7 has sequential dependencies: first
ctx7 libraryresolve the library ID, thenctx7 docsquery the content (at most 3 calls per question) - For the same query, wording can be adjusted for different search engines (e.g., for Chinese queries, give MCP search the original text, and give WebSearch an English translation)
- During deep research, use
agent-browserto open the Google/Bing search page, and viasnapshotobtain element references and then extract results
Step 3 — Result synthesis
- Deduplication: deduplicate by URL, merging information from the same page across different sources
- Grouping: group by subtopic or relevance
- Sorting: sort by relevance and information quality
- Output format:
## 搜索结果:[查询主题]
### [子主题 1]
- 要点摘要(来源:[标题](url))
- 要点摘要(来源:[标题](url))
### [子主题 2]
- ...
---
**来源列表**
1. [标题](url) — 简述
2. [标题](url) — 简述
Step 4 — Follow-up actions
After search results are displayed, provide follow-up options:
- "Would you like me to read the full content of a certain link?" → switch to Fetch Mode
- "Would you like to save it to the knowledge base?" → guide the user to trigger the knowledge management skill
- The user can continue asking follow-up questions to narrow or broaden the search scope
Fetch Mode
Extract clean content from a URL and perform structured analysis.
Step 1 — Extract content
Try in order of priority; automatically fall back when the previous one fails (error, empty return, content < 100 characters):
1. crwl <url> -o md-fit ← 首选,支持 JS 渲染
2. agent-browser open <url> → snapshot → get text ← 最终手段
After successful extraction, record which tool was used, so the user can be informed transparently.
For tool invocation details, see references/tool-catalog.md。
Step 2 — Structured analysis
Selectively apply analysis dimensions based on the content type. For detailed templates, see references/analysis-template.md。
| Content type | Applicable dimensions |
|---|---|
| Articles, reviews, papers | All five dimensions (text deconstruction → concept extraction → critical examination → thought deepening → practical transformation) |
| News, product pages | Text deconstruction + concept extraction + practical transformation |
| Tutorials, technical documentation | Text deconstruction + concept extraction + practical transformation |
| Data pages, API documentation, reference tables | Skip analysis and directly present the extracted content |
Step 3 — Output
Two output modes:
Standalone mode(when used directly by the user):
## [页面标题](url)
**作者**: xxx | **日期**: xxx | **字数**: xxx
### 摘要
一段话总结。
### 分析
[按 analysis-template.md 输出适用维度]
### 关键要点
- 要点一
- 要点二
- 要点三
After the analysis output, proactively prompt: "Would you like to save it to the knowledge base?" If the user agrees, guide them to trigger the knowledge management skill's save process.
Integrated mode(when invoked by another skill on its behalf):
Output the following information block in natural language for the caller to consume directly (do not output JSON):
- Page titleandoriginal URL
- One-sentence summary
- Key points(list of key points)
- Five-dimension analysis(output according to applicable dimensions)
- Full main text(the full extracted clean markdown text)
- Extracted entities: mentioned tools/software, stocks/companies, people (name + role), books/papers/resources
Fallback strategy
How to handle when a certain tool is unavailable:
| Tool unavailable | Impact | Alternative |
|---|---|---|
| crwl | Unable to extract content | Fall back to agent-browser |
| MCP search | Fewer search sources | Use WebSearch for all queries |
| ctx7 | No library documentation search | Use WebSearch + site:xxx targeted search |
| agent-browser | Unable to handle JS-heavy pages | Inform the user, extraction failed |
| All search tools | Unable to search | Inform the user, suggest providing the URL directly |
Do not interrupt the process just because a certain tool fails—always try the next fallback option, and only inform the user after all have been exhausted.
Integration protocol
This skill can be delegated and invoked by other skills. The caller just needs to specify "delegate search-and-fetch" in its own SKILL.md.
Behavior when delegated
- Execute according to the mode specified by the caller (Search/Fetch)
- UseIntegration modeOutput format
- Do not directly interact with the user for follow-up questions (handled by the caller)
- If content extraction fails completely, return a failure flag and the reason, and let the caller decide how to inform the user