Skill: Droid Browse

Use this skill when you need live browser automation during a Factory session—opening sites, clicking through flows, gathering structured data, or capturing screenshots.

Inputs

• Target URL or task description (natural language)
• Optional structured extraction schema (JSON field → type)

Behavior

1. Ensure Chrome is running via Stagehand with the local profile stored in .chrome-profile.
2. Support these commands:
   • navigate <url>: open a page and capture a screenshot.
   • act "<instruction>": perform natural-language actions.
   • extract "<instruction>" '{"field":"type"}': return structured data.
   • observe "<goal>": list suggested steps the agent can take.
   • screenshot: capture the current viewport.
   • close: shut down the session when finished.
3. Save screenshots to agent/browser_screenshots and report the file path in the response.
4. When tasks finish, summarize what happened plus any follow-up steps for the user.

Verification

• If a navigation/action fails, include the error message and prompt the user for next steps.
• Before ending the session, ensure close has been run so Chrome processes don’t linger.

Notes

• This skill expects ANTHROPIC_API_KEY (for Stagehand) and, if used via droid exec, FACTORY_API_KEY to already be configured.
• The working directory should remain inside the cloned skill folder so relative paths resolve correctly.