[Bug] Codex Desktop: benign diagnostics blocked, tool schema error, and broken thread recovery

Summary

I am using Codex Desktop on Windows for a real hardware/software project. Several Codex-side failures are interrupting normal work. These are not project-code, server, or hardware failures.

Confirmed problems

1. Benign local diagnostics can be blocked as a policy violation

A normal, read-only request to check whether my computer could receive 4G data from my own device was rejected with:

Invalid prompt: your prompt was flagged as potentially violating our usage policy. Please try again with a different prompt.

The task was limited to observing a local serial/4G data path. It did not request unsafe behavior, unauthorized access, evasion, or harmful activity.

Expected: allow ordinary local diagnostics, or provide a specific safe alternative instead of an opaque rejection.

2. Internal tool-call schema error is exposed to the user

The desktop app showed:

[ObjectParam] [input[176].namespace] [unknown_parameter] Unknown parameter: 'input[176].namespace'.

I did not provide this internal parameter. The error interrupted the task without identifying the affected tool or offering recovery.

Expected: do not send unsupported parameters; preserve the task context and present a recoverable error if a tool integration fails.

3. Thread is left in systemError despite nearby turns completing

The same hardware-debugging thread became systemError, although some nearby ordinary turns were marked completed. From the UI it is unclear whether the thread can continue, whether context was lost, or whether a new thread is required.

Expected: show the failure time/subsystem/retryability and automatically clear or clearly label recovered threads.

4. Voice-session continuity appears unreliable

Long voice sessions can fail with a "too long"-style error. Starting a new voice session has sometimes shown the same error instead of starting with a clean context, and voice startup can fail.

Expected: a newly created voice session should have an independent context and a clear recovery path.

5. Possible excessive local disk growth

During Codex use, local disk usage has at times grown enough to nearly fill the drive. I do not yet know whether this is caused by logs, worktrees, caches, downloads, or temporary files.

Expected: clear disk-usage attribution, retention controls, and safe cleanup guidance. I can provide a separate disk-usage inventory if useful.

Reproduction context

• Windows desktop app
• Long-running, multi-threaded project work
• Voice interaction plus local agent/tool usage
• Multiple archived and active threads

I have screenshots of the policy rejection and the namespace error, and can provide sanitized diagnostic details privately.

Request

Please advise on the best way to provide diagnostics for these desktop issues. I am willing to continuously reproduce regressions and file structured reports. If OpenAI has a Codex beta/QA program, could you consider providing test access or usage credits (for example, around USD 200) so I can keep testing and reporting issues? This is a request, not a condition for the report.