macOS: reopening a task with Data Analytics artifacts crashes AppRoutes with React error #185

What version of the Codex App are you using?

Codex Desktop 26.715.70719 (build 5650)

What platform is your computer?

macOS 26.5.2 (25F84), Apple silicon

What issue are you seeing?

One specific task becomes completely inaccessible after it contains a failed Data Analytics artifact card followed by a corrected artifact render.

The corrected MCP resource read succeeds, but reopening the task replaces the entire Codex window with:

Oops, an error has occurred

The desktop log then records an AppRoutes error boundary with React error #185. React's decoder identifies #185 as "Maximum update depth exceeded."

This is a renderer-level failure: navigating back to another task works normally, but reopening the affected task reproduces the crash every time.

Steps to reproduce

1. In a task, render a Data Analytics report whose table manifest references fields that are missing from the selected dataset rows.
2. The artifact card displays "Artifact validation failed."
3. Correct the artifact so the table points to a dataset containing all declared fields, then render it again in the same task.
4. Navigate to another task.
5. Reopen the original task.

Expected behavior

The task should open normally. A malformed artifact should remain isolated to its own card, and the corrected artifact should render without affecting the task route.

Actual behavior

The complete task route crashes and the task can no longer be opened in Codex Desktop.

The relevant sanitized log sequence is:

info  [AppServerConnection] response_routed ... method=mcpServer/resource/read ... errorCode=null
error [electron-message-handler] error boundary ... errorMessage="Minified React error #185; visit https://react.dev/errors/185 ..." ... name=AppRoutes

The component stack includes bundles named:

artifact-tab-content.electron
notebook-preview-panel
local-conversation-thread
app-main

Immediately before the error, the log also shows a rapid series of browser-sidebar owner sync/rebind events for the affected route.

Additional information

The task contains two artifact cards: the original failed card and the corrected render. Both were produced from the same report manifest metadata, so duplicate artifact identity/state may be involved, but I have not confirmed that as the exact trigger.

The artifact's separate validation-contract problem is tracked at:

https://github.com/openai/role-specific-plugins/issues/14

That plugin issue explains why the first malformed artifact was accepted by the MCP validator. It does not explain why Codex Desktop later enters a React update loop and crashes the whole task after the corrected render.

I have not attached raw desktop logs because they contain local paths and task content. I can provide a narrowly redacted excerpt if needed.

Workaround

I can navigate away to other tasks, but I have not found a supported way to remove the persisted artifact card from the affected task, so that task remains inaccessible.