codex - 💡(How to fix) Fix Codex in-app browser cannot enter Cyrillic/Unicode text when virtual clipboard is missing

Summary

Codex desktop app in-app browser cannot reliably enter Cyrillic/Unicode text into normal web forms. Standard Browser/Playwright input methods fail with a virtual clipboard error, and the available fallbacks do not provide a reliable Unicode input path.

This blocks production workflows where the agent must preserve non-Latin customer names, cities, and notes exactly.

Environment

• Codex desktop app on Windows
• Codex Browser / in-app browser
• Browser plugin path observed locally: openai-bundled/browser/26.519.41501
• Target page type: authenticated HTTPS CRM web app with normal <input> and <textarea> fields

Reproduction

1. Open Codex desktop app.
2. Open any authenticated HTTPS web app with a normal text input.
3. Focus a text input or use a locator for it.
4. Try to fill Cyrillic/Unicode text:

await tab.playwright.locator('input[name="name"]').fill('Тест Клиент')

or:

await tab.dom_cua.type({ text: 'Тест Клиент' })

Observed behavior

The action fails before text is inserted:

Browser Use encountered an error interacting with this webpage's clipboard:
Browser Use virtual clipboard is not installed
locator.fill failed for selector input[name="name"]

Direct Cyrillic keypress is also not supported:

Unknown key: "А"

The Browser runtime's evaluate() path is read-only, so it cannot be used as a safe workaround to set form values.

Expected behavior

Codex Browser should support reliable Unicode text entry into editable fields:

• locator.fill('Тест Клиент') should work.
• locator.type('Тест Клиент') should work.
• tab.dom_cua.type({ text: 'Тест Клиент' }) should work.
• If a virtual clipboard is required, the app should install/repair it or show a clear setup instruction before the task reaches a live form.

Impact

This is not cosmetic. It can cause real data-entry failures in production systems:

• customer names and cities cannot be preserved in Cyrillic;
• agents may be tempted to transliterate, which is incorrect;
• CRM records, project notes, and customer handoffs become wrong or unreadable;
• normal browser automation cannot complete the task safely.

Requested fix

Please provide a non-clipboard Unicode input fallback or repair the virtual clipboard dependency in the in-app browser. A regression test should cover Cyrillic/Kazakh characters in both <input> and <textarea> fields.