tutti/packages/agent/runtimeprep/skill_templates/browser-use.md

name, description

name	description
browser-use	Use to operate a web browser — open URLs, read page content, click, fill forms, take screenshots — through the Tutti CLI.

Browser Use

Use this skill for browser tasks: open URLs, read pages, click, fill forms, run page JS, or capture screenshots.

Drive the browser only through {{.CLICommand}} browser. The Tutti daemon owns the browser session. Do not launch open, xdg-open, start, google-chrome, chromium, or direct browser automation; those are outside the managed session.

Protocol

1. Open a requested URL directly with {{command "browser.open" (args "url" "<url>")}}. This creates and selects a managed Browser page before loading the URL. Do not use a cold navigate call as the first command.
2. Read the current page with {{command "browser.snapshot"}}; use returned uid values for interactions.
3. Act with {{command "browser.click" (args "uid" "<uid>")}} or {{command "browser.fill" (args "uid" "<uid>" "value" "<text>")}}.
4. Use {{command "browser.list-pages"}} to inspect the workspace Browser and any session-owned Agent Browser tabs. Select a stable target with {{command "browser.select-page" (args "page-id" "<id>")}} when needed.
5. Create another managed Browser page with {{command "browser.new-page" (args "url" "<url>")}} and close a page you own with {{command "browser.close-page" (args "page-id" "<id>")}}.
6. Navigate the selected page later with {{command "browser.navigate" (args "url" "<url>")}}.
7. Use {{command "browser.eval" (args "script" "'() => document.title'")}} for page JS and {{command "browser.screenshot"}} for a PNG path.
8. Re-run snapshot after navigation or UI-changing actions because uid values can change.

Append --json only when the exact command's help advertises it. Browser commands support JSON, but plain output is normally sufficient for page text.

Guardrails

• The desktop browser session includes visible in-app BrowserNode tabs for the current workspace. Website App tabs are outside this automation scope.
• User Browser tabs may contain the user's active work. Read or modify them only when that is required by the request.
• If the desktop BrowserNode host or the headless managed Chrome backend is unavailable, report that error instead of falling back to shell/browser tools.