| name | web-search |
| description | Provides capabilities to search the web and read online articles. |
| description_cn | 提供搜索互联网内容和读取网络文章的能力。 |
| license | Apache-2.0 |
| type | bundled_plugin |
| source_type | bundled_plugin |
| default_enabled | false |
| metadata | {"author":"deepseek-cowork team","version":"2.0"} |
| security_level | medium |
| allowed-tools | search_web read_web_article |
Web Search Skill
This skill provides a structured free web search pipeline and a markdown-first article extraction pipeline.
Capabilities
- Search Web: Search using free public search providers with fixed fallback order: DuckDuckGo -> Bing.
- Structured Output: Both tools return JSON object strings with
ok, error, metadata, and normalized content.
- Read Web Article:
read_web_article extracts internet article text with fixed fallback order: markdown.new -> defuddle.md -> r.jina.ai -> Scrapling.
- Context Safety:
read_web_article returns cleaned text content only and does not return raw HTML by default.
Usage Guidelines
- Search Stability: The search tool retries requests, records fallback chains, de-duplicates URLs, and limits over-representation from the same domain.
- Domain Controls:
search_web supports allowed_domains and blocked_domains, but they are mutually exclusive.
- Article Extraction:
read_web_article never treats raw HTML as a successful default response.
- Response Contract: Callers should parse JSON and inspect
ok, error, provider_used / extractor_used, results, and content.
- Fallback Order:
markdown.new → defuddle.md → r.jina.ai → Scrapling.
Current Runtime Notes
- Web search and article reads are read-only and may be good candidates for
parallel_tools when querying several independent pages or sources.
- For volatile facts, prefer current web lookup and state the query/source timing in the answer.
- Respect domain filters when the user asks for official sources or excludes a site.