| name | browser-task |
| description | Orchestrates browser tasks requiring multi-tool coordination: search-then-navigate, login-then-download, multi-page interactions. Use when the agent needs to: (1) Search for a website then interact with it, (2) Login or fill forms, (3) Navigate multiple pages to find a target, (4) Extract download links from dynamic pages. For simple cases: use webfetch for static pages, curl for known file URLs. |
BrowserTask
Complete complex browser tasks requiring multi-step, multi-tool coordination.
Available Tools
| Tool | Capability | Use When |
|---|
| Search tools | Find trusted URLs | Target website unknown |
| webfetch | Fetch page content/HTML | Static pages, parse links |
| agent-browser | Browser interaction | Click, fill forms, login, dynamic content |
| curl | Download files | File URL is known |
Core Principles
- Search first, never guess URLs - Use search tools to get trusted URLs
- Right tool for the job - agent-browser for interaction, curl for download
- Extract link then download - agent-browser cannot download files directly; extract href then use curl
Standard Download Pattern
Search/known URL -> agent-browser navigate to download page -> get download button href -> curl download
Key commands:
agent-browser get attr @download-button href
webfetch <url> --format html
agent-browser Key Points
For detailed commands, load agent-browser skill. Common usage:
agent-browser open <url> - Open pageagent-browser snapshot -i - Get interactive elementsagent-browser click @e1 - Click elementagent-browser get attr @e1 href - Get link attribute
Important Notes
- agent-browser is slow to start; use webfetch/curl when possible
- Search results are more reliable than guessed URLs
- Always use curl for file downloads, never attempt with agent-browser
