Captions not auto-enabled: CC button detection fails

[smian1]

Bug

When joining a Google Meet with --auth, the bot logs:

WARNING: Could not verify captions are on — capture may not work

And the transcript file remains empty despite participants speaking.

Environment

• macOS (arm64), Node v25.5.0
• openutter 0.1.6
• Playwright Chromium
• Joined with --auth (37 cookies, authenticated session)

What happens

The bot tries 4 methods to enable captions (click CC button, press 'c', Ctrl+Shift+C, more options menu) but none succeed. The caption container MutationObserver never fires.

Expected

Captions should be auto-enabled after joining, or the bot should retry enabling them periodically.

Workaround

Joining with --headed and manually clicking CC in the bot's browser window.

[sumansid]

@smian1 Thanks for pointing it out. Can you run npx openutter screenshot in a different terminal window, while you still have the main playwright instance running? I want to see if the CC is blocked due to some dialog boxes.

[Hugo-DL]

I ran into the same issue and did some debugging. Here is what I found:

Root Cause

The enableCaptions() function dismisses overlays (8× Escape + "Got it"/"Dismiss"/"Continue" buttons), but it does not handle the Gemini in-meeting prompts that Google shows to authenticated users.

When joining with --auth, Google Meet recognizes you as a real user and shows Gemini-related UI elements after joining (e.g. "Take notes with Gemini", "Try Gemini" banners). These appear after the post-join dialog dismissal runs, and they overlay or intercept the CC button area and keyboard shortcuts.

Evidence:

• --anon mode: Gemini banners appear on the pre-join page (where dismissOverlays() handles them), CC button works fine after joining
• --auth mode: Gemini prompts appear in-meeting (after join), but only dismissPostJoinDialogs() runs there — which does not check for Gemini elements

The Gemini dismissal logic (text=/Use Gemini/i + Escape) exists only in dismissOverlays() (pre-join), not in enableCaptions() or dismissPostJoinDialogs().

Suggested Fix

Add Gemini banner dismissal to enableCaptions() (or dismissPostJoinDialogs()) before attempting to click the CC button:

// In enableCaptions(), after the initial overlay dismissal:
try {
  const gemini = page.locator('text=/Use Gemini|Gemini|Take notes/i').first();
  if (await gemini.isVisible({ timeout: 1000 })) {
    await page.keyboard.press("Escape");
    await page.waitForTimeout(500);
  }
} catch {}

I can submit a PR if you want.

[sumansid]

@Hugo-DL Feel free to raise a PR

[Hugo-DL]

Update on this — after implementing the Gemini dismissal fix I suggested above, the issue still persists. The current codebase (v0.1.6+) already includes the Gemini banner dismissal in enableCaptions(), but auth mode captions still fail.

So the root cause is deeper than just unhandled Gemini overlays. My current thinking:

1. The Gemini selectors may not match — Google may be rendering different text/elements than Use Gemini|Take notes with Gemini|Try Gemini
2. Gemini side panel vs. banner — auth users may get a persistent Gemini companion panel (not just a dismissable banner) that steals focus from the toolbar and intercepts keyboard shortcuts
3. Toolbar behavior differs in auth mode — the bottom toolbar auto-hide logic may work differently, so page.mouse.move(640, 680) may not reliably surface the CC button
4. Keyboard shortcut c intercepted — the Gemini panel may capture keystrokes before they reach the Meet caption toggle

Next steps I'm planning:

• Run a --headed --auth session and take debug screenshots at each step of enableCaptions() to see what the bot actually sees
• Identify the exact element(s) blocking CC activation
• Build a more robust auth-mode caption flow based on findings

This will likely be a bigger PR than I initially thought — not just a selector fix but potentially a new auth-mode-specific caption enablement path. Will report back with findings before submitting.