Codex app in-app browser: coordinate drag does not trigger HTML5 drag/drop interactions

Open 💬 0 comments Opened May 17, 2026 by Alien-AV

Summary

In the Codex app in-app browser, agent-driven coordinate dragging appears available, but it does not trigger a web app's HTML5 drag/drop event flow (dragstart, dataTransfer, drop). This makes it impossible for the agent to validate UIs that rely on native drag/drop interactions.

Environment

  • Codex desktop app on Windows
  • In-app browser target: local dev server at http://localhost:8787/
  • Browser Use / in-app browser plugin was otherwise connected and able to inspect DOM, click buttons, and read console logs.

Repro

  1. Open a local HTML app in the Codex in-app browser.
  2. Use an app that has draggable cards implemented with native HTML drag/drop:
  • source element has draggable="true"
  • dragstart sets event.dataTransfer.setData(...)
  • target listens for dragover/drop
  1. Ask the agent to drag one card onto another using the in-app browser automation.

Observed

  • tab.cua.drag({ path: [...] }) moves through the requested coordinates but does not trigger the app's dragstart/drop handlers.
  • No app state changes occur.
  • No debug logs from the app's drop path appear.
  • The Playwright surface exposed in this runtime did not include locator.dragTo(...), so there was no alternate DOM-aware drag method available.

Expected

The in-app browser automation should provide a supported way for agents to perform native web drag/drop interactions, either by:

  • making coordinate drag generate normal browser drag/drop events with dataTransfer, or
  • exposing a Playwright-compatible dragTo/drag-and-drop API, or
  • documenting the current limitation and recommended workaround.

Impact

Agents can click, inspect, and validate many UI flows, but cannot validate drag/drop-heavy local apps without manual user intervention or app-specific test hooks.

View original on GitHub ↗