browser-automation

Clear each field before typing, use input methods appropriate to the field type, and handle dropdowns, checkboxes, and radio buttons explicitly. Group related interactions into a single tool call and always submit the form after filling rather than stopping at filled fields.

Use read_page or get_page_text tools to extract text content, which is preferred over screenshots for tables and structured data. Reserve screenshots for visual content like charts, images, and complex layouts where text extraction loses meaning.

Prefer text extraction whenever possible since it is more efficient and structured. Use screenshots only for visual-heavy applications like Google Docs, Figma, or Canva, or when you need to understand page layout before interacting with elements.

CAPTCHAs cannot be bypassed by the automation. When a CAPTCHA or access-denied page appears, the agent stops and informs the user rather than attempting to circumvent the protection.

Avoid browser automation for general web search, local file reading, or known API endpoints, since dedicated search, file, and API tools handle those tasks better. Browser automation is intended for interactive web apps and complex page interactions.