WebWalkerQA
Fits agents that must navigate inside one site (a company portal, a university site, a conference site) rather than search the open web. Borrow the root-site plus question design and the action-count efficiency metric.
What it asks the agent to do
- start at a given root website and click through subpages to find an answer
- find information that search engines do not surface
- answer in Chinese or English depending on the site
What to borrow from it
case templatebaseline
Caveats
Pages change, so answers can rot; only four domains, 60% Chinese; the linked WebAgent repo is a general family repo and does not clearly host the dataset or its eval script; license for the data not stated.
Details
| Workloads | web browsing, knowledge qa |
|---|---|
| Interaction | autonomous long horizon |
| Environment | browser |
| End state checked | No |
| Tools | page navigation |
| Grading | llm judge |
| Size | 680 |
| Data | real scraped |
| Personal data | none |
| License | unknown |
| Harness | none |
| Status | active |