yutori-browse（游鸟浏览）

Browser Automation

Help the user automate browser-based tasks using Yutori's Navigator agent.

Process

1. Understand the task

- What website needs to be automated? - What actions are required? (clicking, typing, extracting data) - Does it require login or authentication?

1. Define the task clearly

- Break complex workflows into clear steps - Specify what data to extract if applicable - Note any buttons or elements to interact with

1. Start the browsing task Use run_browsing_task with:

- task: Clear natural language instructions - start_url: The URL to begin browsing - max_steps: 25 (default) to 100 for complex flows - require_auth: true: For login forms or other authenticated workflows - browser: "local": When the task should run through Yutori Local with the user's desktop session - output_fields: For structured data extraction (e.g., ["name", "price", "url"]) - webhook_url: HTTPS URL for completion notification - webhook_format: scout (default), slack, or zapier

1. Poll for results

- Browsing typically takes 30-120 seconds depending on complexity - Use get_browsing_task_result to check status - Poll every 10-15 seconds until complete

1. Review and validate

- Check the extracted data or confirmation - Verify the task completed as expected

Task Writing Tips

• Be specific about UI elements: "Click the blue 'Submit' button"
• Reference visible text when possible
• For forms, specify which fields get which values

$ARGUMENTS