Problem

When trying to interact with elements inside cross-origin iframes (such as Apple's authentication iframe on icloud.com), agent-browser cannot access or interact with the iframe content.

Current Behavior

• agent-browser snapshot does not include elements from iframes
• Mouse clicks via agent-browser mouse move/down/up click on the iframe but don't focus internal elements
• Keyboard input via agent-browser press doesn't reach elements inside the iframe
• agent-browser fill "iframe >> input" "text" doesn't work for cross-origin iframes
• agent-browser eval cannot access iframe content due to cross-origin security restrictions

Expected Behavior

Playwright supports iframe interaction via page.frameLocator() which properly handles cross-origin frames:

const frame = page.frameLocator('iframe[name="auth-widget"]');
await frame.locator('input[type="text"]').fill('[email protected]');

Proposed Solution

Add frameLocator support to agent-browser, potentially via:

1. A new --frame flag for commands:

   agent-browser fill --frame "iframe[name='auth']" "input" "text"
   agent-browser click --frame "iframe" "@e1"

2. Or a frame selector syntax:

   agent-browser fill "frame:iframe[name='auth'] >> input" "text"

3. Or frame-specific commands:

   agent-browser frame "iframe[name='auth']" fill "input" "text"

Use Case

This is critical for automating authentication flows (Apple, Google, OAuth providers) where login forms are embedded in cross-origin iframes for security.

Environment

• agent-browser version: latest (via npx)
• macOS
• Tested with iCloud authentication (idmsa.apple.com iframe)

Workaround

Currently, users must use Playwright directly or tools like mcp__next-devtools__browser_eval which properly expose frameLocator functionality.