A user who asked Claude to manage a batch of Codex sessions uncovered what appears to be a history-related bug in Codex. In the reported case, one session had run continuously for two days, produced 110MB of records, and consumed about 353 million tokens. Yet after reopening the session, Codex only showed the very first sentence. The underlying work was not lost.
According to the report, Codex stores the full chat log locally and also builds a separate index for fast access. The failure appears to sit in that index: it gets stuck near the beginning, leaving later records still present on disk but invisible to Codex. The user checked 439 sessions and found index anomalies in 398 of them.
Similar "false amnesia" reports later appeared in environments including Windows Desktop and Remote SSH. In those cases, the underlying records were still there, but the interface stayed pinned to an older position and did not automatically raise an error or repair itself. Deleting the damaged index let Codex reread a 110MB session in about 20 seconds and restore all 398 affected sessions, though later cases suggested the issue may not always be limited to the index. Some original records may also carry duplicated sequence numbers. The bug remains open, and no related fix PR has been seen yet.
BlockBeats reported that a user who had Claude manage a group of Codex sessions found a history bug in Codex.
In the reported case, one conversation ran for two straight days, generated 110MB of records, and used about 353 million tokens. After the session was reopened, Codex remembered only the first sentence. The actual work was still intact.
The issue appears to be in the session index
Codex stores the full chat history locally and also creates a separate index for quick loading. The reported failure sits in that index: it stopped at an early point, so later records were still present but no longer visible to Codex.
The user reviewed 439 sessions and found index anomalies in 398 of them.
Similar "false amnesia" reports appeared in other environments
Users in Windows Desktop, Remote SSH, and other environments later reported similar behavior. The pattern was described as "false amnesia": the underlying records remained available, but the interface stayed stuck at an older point and did not automatically throw an error or repair the issue.
Deleting the damaged index restored access, but not every case may be the same
After the original poster deleted the corrupted index, Codex reread a 110MB session in about 20 seconds and restored all 398 affected sessions.
Later cases suggested the issue may not be limited to the index alone. Some original records may also contain duplicated sequence numbers, which means deleting the index is not necessarily safe in every case.
The bug is still marked open, and no related fix PR has been seen so far.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
