Auto-compact fails
Resolved 💬 2 comments Opened Feb 13, 2026 by le876 Closed Feb 13, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
### Steps to reproduce
- Start a long, multi-turn diagnostic conversation.
- Accumulate many context-heavy messages (commands/results/analysis).
- When context window is near limit, continue with another query.
- Observe hard error asking to start new thread / clear history instead of graceful summarization.
### Expected behavior
- Before hard cutoff, Codex should trigger automatic context compression (or offer an explicit “compress and continue” option).
- It should preserve key context (e.g., session goal, environment state, diagnostic findings, IDs, executed commands).
- Conversation should continue without requiring manual new-thread handoff.
### Actual behavior
- No auto-compression observed.
- Session abruptly stops with context window error.
- User must manually start a new thread or clear history, interrupting troubleshooting flow.
2 Comments
There are not enough details in the bug report for us to diagnose this, so I'm going to close.
Feel free to open a new bug report, but please fill in all of the fields in the bug report template. Also, for a bug like this, it's really important that we have access to your logs and session details. You can upload those using the
/feedbackslash command. Then include the thread ID in the bug report so we can reference the logs.When will this be worked on. Codex-5.3 and codex-5.3-spark are completely useless and they keep churning tokens without editing any code until they "run out of room"