openai / openai/codex

Codex Computer Use regression: Custom GPT file upload no longer emits `filechooser` event

Open
#46,585 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app browser bug computer-use tool-calls windows-os
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

Summary

Codex Computer Use can no longer automatically upload files to a Custom GPT through the supported browser-control flow.

This workflow worked successfully in July 2026. Codex uploaded 45 poem JSON files in 45 separate Custom GPT chats, submitted each prompt, captured the responses, and validated the outputs.

On September 18, 2026, automated file upload stopped working. Manual file upload in ChatGPT still works normally.

Environment

  • Windows
  • Codex app: 26.911.7940
  • Bundled browser plugin/native host: 26.911.61220
  • Chrome extension: 1.26.901.11451
  • Tested in both Chrome and signed-in Edge
  • ChatGPT Custom GPT is correctly opened and authenticated

Failure

The Custom GPT composer shows the normal Add files and more control.

The page contains #upload-files with type="file". The input is hidden.

Observed behavior:

  1. Codex clicks Add files and more.
  2. Codex selects Add photos & files.
  3. tab.playwright.waitForEvent("filechooser") is already armed.
  4. No filechooser event is emitted.
  5. The wait times out.
  6. No file is attached.
  7. No prompt is submitted.

Direct Playwright clicking of the hidden file input reports no_visible_match.

Manual upload through the normal ChatGPT interface works.

Exact error from a repeat test

Error: Timed out after 3000ms waiting for file chooser.
Error: Playwright selector deadline exceeded
waiting on click for selector #upload-files
Locator diagnostics: {"kind":"no_visible_match","action":"click","locator":"#upload-files","matchCount":1,"matches":[{"ariaLabel":null,"disabled":false,"role":null,"tag":"input","text":"","type":"file","visible":false}],"visibleCount":0,"truncated":false}

In that repeat test, clicking Add files and more did not leave an upload menu open. No file was attached or prompt submitted.

Troubleshooting already completed

  • Enabled Chrome file-URL access.
  • Enabled all-site extension permission.
  • Verified extension installation.
  • Verified native-host manifest.
  • Verified registry mapping.
  • Reinstalled the extension.
  • Reconnected Computer Use.
  • Refreshed and restarted Chrome.
  • Restarted Codex.
  • Removed Grammarly.
  • Tested in Edge with the same result.
  • Confirmed via bounded CDP diagnostics that a valid file input exists.
  • Did not use CDP to inject a file or bypass the supported Computer Use safety layer.

The evidence points to the supported Computer Use / Playwright file-chooser handoff; the root cause is not confirmed.

Expected behavior

click Add files and moreclick Add photos & filesfilechooser event emittedchooser.setFiles(...) succeeds → uploaded filename appears as an attachment chip in the Custom GPT composer.

Actual behavior

The visible upload action does not emit the Playwright filechooser event.

Regression

The automated workflow worked in July 2026 and no longer works as of September 18, 2026.

Related but distinct: #40046 requests a supported fallback for hidden file inputs when no chooser event fires. This report concerns a previously working Custom GPT upload flow that has regressed.

Request

Please confirm whether this is a known compatibility regression between Codex Computer Use and the current ChatGPT/Custom GPT upload interface.

I need a supported repair for repeatable automated uploads. I am not requesting raw CDP file injection, a browser-security bypass, or another unsupported method.

Please advise:

  1. Whether this is a known regression.
  2. Whether the Codex app, native-host, and extension versions above are expected to be compatible.
  3. Whether a newer Computer Use component fixes the issue.
  4. Which diagnostic logs would be useful.
  5. How to restore the supported filechoosersetFiles() → attachment-chip workflow.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the supported Computer Use flow in Chrome or Edge: click Add files and more, choose Add photos & files, and observe the Playwright filechooser handoff. Inspect the Computer Use, native-host, and Chrome extension logs around the chooser event; done means the event fires, chooser.setFiles() succeeds, and an attachment chip appears without CDP injection.

Written by the indexing model from the issue text.

Assessment

Tech stack
playwright
Domain
devtools, tooling
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.