openai / openai/codex

[Desktop] Support background cross-task image inspection from connected hosts

Open
#43,486 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app enhancement remote
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

What variant of Codex are you using?

Codex Desktop on macOS 26.901.51231 (build 8109), coordinating a task on a connected Windows host.

What feature would you like to see?

A supported, bounded way for a coordinating task to inspect an explicitly selected image from another authorized task or connected host without navigating the user's foreground Codex window.

Example workflow: a Windows task captures and saves a UI screenshot, and a Mac coordinating task needs to independently inspect the pixels before accepting the result. The remote screenshot can be opened successfully in the native Codex preview, but doing this through foreground navigation interrupts the user's work.

An artifact-read tool accepting a task/host and an explicit artifact reference, returning a model-visible image or a local readable copy, would address this. It should preserve authorization boundaries and avoid importing entire image-heavy transcripts. Alternatively, documentation for an existing supported background route would be welcome.

Additional information

This is a capability request, not a claim that the current text-summary API promises cross-task vision.

In the current installed implementation, read_thread serializes its response as text. With includeOutputs:true, mcpToolCall and dynamicToolCall records still omit their result content; imageView exposes a path, and other image item types may expose serialized references/data. Therefore a screenshot emitted by an MCP tool is not automatically delivered as image input to the coordinating model. A path on the remote Windows filesystem is not directly readable by a Mac-local image viewer.

The full callable tool registry available in the diagnostic task exposed read_thread, wait_threads, local view_image, and open_in_codex, but no dedicated remote artifact/image retrieval capability. open_in_codex is documented as a UI operation. Supplying another task ID can queue a panel until that task is shown; this does not return image pixels to the calling model.

For same-host artifacts already accessible as local files, a separate view_image call is a supported background route. The missing capability in this report is remote artifact retrieval for model inspection without foreground navigation or asking the producing task to encode/retransmit the file manually.

Nearby reports #41532 (mobile preview), #35759 (cross-host task-creation attachments), and #39148 (unbounded serialized image-generation payloads) cover different surfaces. This request is specifically for selective background cross-task image inspection. Raw screenshots, task IDs, account details, and private paths are omitted.

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 reviewing the current read_thread, view_image, and open_in_codex behavior, including how includeOutputs handles image records. Done means an explicitly selected artifact from an authorized task or connected host can be inspected in the background without foreground navigation, while preserving authorization boundaries and avoiding full transcript import.

Written by the indexing model from the issue text.

Assessment

Tech stack
rust
Domain
ai-infra-agents, desktop
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.