[Windows 10 / Codex Work] Remote compaction repeatedly fails across multiple threads (decode error and SSE timeout) since Oct 3 #53111

Summary

Since 2026-10-03, remote context compaction has repeatedly failed in multiple separate Codex/ChatGPT Work task threads. As of 2026-10-11, the problem is still blocking the reporter's longer-running technical work. This is not a report of a single corrupted conversation: screenshots show recurring errors in a restored long-running Astra HQ session and in other separate task/project threads. The reporter also observed a recurrence in a newly created thread on 2026-10-07 (date and fresh-thread status from their timeline, not independently timestamped in the screenshots).

The user has a paid ChatGPT Pro subscription (reports paying about USD 117/month). Shorter, bounded tasks in fresh sessions may still work, but automatic compaction/continuation of substantial tasks remains unreliable, forcing work to be split into smaller sessions and repeatedly reconstructed.

Environment

• Host: Windows 10 Pro 22H2.
• Surface: ChatGPT Codex/Work desktop tasks. Evidence screenshots were captured from the iOS ChatGPT client, showing the affected task threads (with a NIX workspace label); this does not establish that iOS itself causes the failure.
• Agent/model in the affected workflow: Astra / GPT-6 as selected by the user; errors also appear in several task threads.
• Desktop app builds observed by the reporter during the troubleshooting period: 26.930.79.45.0 and, following an application restart/update, 26.1020.71.24.0. Per-screenshot build/runtime correlation is not available.
• Subscription authentication, thread IDs, backend request IDs, and exact network root cause have not been independently extracted. Please advise which safe diagnostic identifiers would help.

Exact user-visible errors

Most frequent:

Error running remote compact task: stream disconnected before completion: Transport error: network error: error decoding response body

Also observed repeatedly in a different thread:

Error running remote compact task: stream disconnected before completion: idle timeout waiting for SSE

One task also displayed the transport/decode error without the Error running remote compact task: prefix.

Timeline (user-reported dates; screenshots do not contain full calendar timestamps)

• Oct 3: First noticed the persistent remote-compaction failures.
• Oct 3–6: Repeated failures in the main long-running Astra HQ / recovery conversation. One screenshot contains three consecutive instances of the same decoding failure.
• Oct 7: The reporter says the problem also recurred in a new thread; this is why treating the initial conversation as the sole defective object is not sufficient.
• Through Oct 11: Failures persist across multiple conversations, including independent project sessions. Captures include visible context-compaction UI followed by the transport/decode error, and a separate conversation with consecutive SSE idle-timeout errors.

Reproduction pattern

1. Continue a nontrivial or long-running Codex/Work task in an existing thread.
2. When remote context compaction is invoked, the compact operation does not complete successfully.
3. Instead of reducing context and allowing normal continuation, one of the errors above is displayed.
4. Retrying continuation may produce the same error again. Starting smaller tasks in new sessions is only a partial workaround; it does not preserve the working context of a larger project.

The reporter has not established that these failures share a single underlying cause, occur at a fixed context percentage, or are exclusively attributable to the desktop client, network, or VPN.

Expected vs actual

Expected: Automatic/remote context compaction succeeds, or a safe recovery path preserves the session and reports an actionable error.

Actual: Compaction fails with transport/decode errors or an SSE idle timeout, preventing reliable continuation of long technical project sessions. A week's worth of attempts has not produced a usable recovery route for the affected workflow.

Evidence

The reporter supplied nine screenshots from affected threads. They show:

• several identical remote-compaction decode failures in a single recovery thread;
• repetition after the UI explicitly shows context compaction;
• a separate project thread exhibiting the same decode failure;
• consecutive idle timeout waiting for SSE errors in another thread.

The captures were reviewed in preparing this report. The GitHub API integration used to open this issue does not support binary screenshot attachments, so the image files are not attached to this issue yet. They can be added through the GitHub web UI; please do not interpret their absence here as an absence of captures. Personal project details should be redacted before public upload.

Requests to maintainers

1. Is this a known failure mode of remote compaction / transport streaming on the current Codex/Work stack? Can maintainers correlate both error families?
2. What safe, supported workaround lets a paying user resume existing long-running threads without deleting or resetting their data?
3. Which diagnostic identifiers/logs can be collected and shared privately for backend correlation, without exposing tokens, conversation contents, or project files?
4. Can you confirm whether a fix is available or being tracked? A generic status page reporting operational service does not resolve this user's repeated compaction failures.

Related issues

• Local Windows and remote SSH tasks fail during compaction: response body decoding / SSE idle timeout #45216 — Windows/remote SSH compaction: decode and SSE idle-timeout failures.
• [Windows][Codex App][Work] Remote context compaction fails at ~84% with "idle timeout waiting for SSE" even when Responses WebSocket is healthy #45472 — Windows Work/Codex remote compaction SSE timeout.
• Windows Codex Desktop: consecutive remote compaction decode failures prevent continuing a local task #50843 — Windows consecutive remote-compaction decode failures.
• Codex Desktop on macOS: automatic context compaction fails every time with “error decoding response body” #38969 — repeated remote-compaction response decode errors.

This issue is a distinct user-impact report across multiple sessions, not a claim that the root cause is already known.