| name | web_research |
| description | Research a topic in a live browser using PageState-based navigation, extraction, and evidence synthesis. |
| official | true |
| version | 0.1.0 |
Web Research Skill
Use this skill when the user wants live information gathered from websites instead of a static codebase or document search.
Goal
Produce a short, evidence-backed answer by navigating pages, reading structured browser_get_page output, and extracting only the content needed to answer the user's question.
Preferred Tool Sequence
- Use
browser_navigate to open the destination or a search engine.
- Use
browser_get_page after every meaningful navigation to inspect the current PageState.
- Use
browser_click and browser_type with backend_node_id when available; otherwise use element_id.
- Use
browser_press_key after typing into search boxes or forms.
- Use
browser_wait when the page is still loading or a result list is expected.
- Use
browser_extract_content or browser_get_text only after landing on the right page.
Operating Rules
- Prefer pages that already show the answer directly; do not open unnecessary tabs.
- Re-read
browser_get_page after clicks, form submits, or infinite-scroll changes.
- If the page state warns about login, captcha, or blocked flows, stop and tell the user exactly what is needed.
- Summaries must distinguish observed facts from inference.
- Cite the page title and URL in the final natural-language answer.
Minimal Demo Flow
browser_navigate to the target page or search engine.browser_get_page to find the primary result.browser_click the result using backend_node_id.browser_extract_content to gather the answer.- Return a concise answer with the source URL.
