Error running remote compact task

@etraut-openai The uploaded thread ID here (019becfa-bd48-7f63-882a-9a90d6052cb9) should be really useful — thanks @nshkrdotcom.This feels like part of the broader “remote compaction can fail hard and strand the session” family (e.g. #9666 is a timeout waiting for the child process; there are a few other remote compact task reports too), just with a different failure mode (context_length_exceeded).

Two thoughts:

• If the compaction step itself can hit context_length_exceeded, it likely means the compactor request is still too large (maybe including too much history/tool output). Server-side truncation/chunking of compactor input might help.
• - UX-wise, if remote compaction fails, a fallback to local compaction (or at least a “continue without compaction” escape hatch) would prevent the user from getting stuck.

If you want to reduce fragmentation, maybe track these under a single canonical “remote compaction failures” issue and link the various thread IDs.

Hasnt happened since, and I have since updated codex.

Still experiencing this issue. It may have to do with too much tool output in my case.

started happening to me as well

id: 019c1eda-5393-7d70-a40c-a93a64ae17d8

Same for me. Having PRO plans. So for my case it happened 2 times and maybe seems related to plan mode. Because i was close to full context length and when using Plan mode it made the plan and didnt replay to launch the plan.

This has been happening to me for a few days. Previously I was a few versions of Codex behind, but I just upgraded and the problem continues.

019c2b9e-9310-7e43-af8f-ba6a031258db

This has been happening to me as well. I'm running codex on windows using ubuntu wsl.

019c9c8c-8d5e-7d20-96d6-06ddcabcb1d1

edit: the specific error I am seeing is below

■ Error running remote compact task: exceeded retry limit, last status: 429
Too Many Requests, request id: <some request id>

Error running remote compact task: {
"error": {
"message": "Unknown parameter: 'prompt_cache_retention'.",
"type": "invalid_request_error",
"param": "prompt_cache_retention",
"code": "unknown_parameter"
}
}

This is happening to me very often now:

Error running remote compact task: {
"error": {
"message": "Unknown parameter: 'prompt_cache_retention'.",
"type": "invalid_request_error",
"param": "prompt_cache_retention",
"code": "unknown_parameter"
}
}

Error running remote compact task: { "error": { "message": "Unknown parameter: 'prompt_cache_retention'.", "type": "invalid_request_error", "param": "prompt_cache_retention", "code": "unknown_parameter" } }

Latest version: very often at pro plan gpt-5.4 highx:
Error running remote compact task: {
"error": {
"message": "Unknown parameter: 'prompt_cache_retention'.",
"type": "invalid_request_error",
"param": "prompt_cache_retention",
"code": "unknown_parameter"
}
}

Running fine again after wait some minutes...

gpt-5.4 xhigh

■ Error running remote compact task: {
"error": {
"message": "Unknown parameter: 'prompt_cache_retention'.",
"type": "invalid_request_error",
"param": "prompt_cache_retention",
"code": "unknown_parameter"
}
}

GPT 5.4

Error running remote compact task: { "error": { "message": "Unknown parameter: 'prompt_cache_retention'.", "type": "invalid_request_error", "param": "prompt_cache_retention", "code": "unknown_parameter" } }

I wait about 5 mins, and it return running normally.

For those who were seeing the "prompt_cache_retention" error, that should be fixed now.

Error running remote compact task: { "error": { "message": "Your input exceeds the context window of this model. Please adjust your input and try again.", "type": "invalid_request_error", "param": "input", "code": "context_length_exceeded" } }
I am a 5x Pro package. Can anyone solve this problem for me?
019e4df6-f2ad-7bd3-a8b3-c0b65c4c4565
<img width="1874" height="848" alt="Image" src="https://github.com/user-attachments/assets/b768ac9c-11c2-4014-8545-c1fd6891fbc6" />

Thanks for reporting this problem. Until recently, Codex used a separate endpoint and server-side logic for compaction. We recently switched to a more robust approach where the compaction logic is moved locally using the same endpoint as normal turns. This eliminates "remote compaction" errors and increases the stability and reliability of compaction operations. If you see further problems with compaction, please use /feedback and open a new bug report.