CLI intermittently halts with `{"detail":"Bad Request"}` and requires manual resume
Open 💬 2 comments Opened Jul 23, 2026 by uditgoenka
Summary
For roughly the last two weeks, Codex CLI has intermittently stopped mid-session and displayed only:
■ {"detail":"Bad Request"}
After this appears, the active work pauses instead of recovering or continuing. The failure is random and has occurred repeatedly during otherwise normal CLI use.
Steps to reproduce
- Start an interactive Codex CLI session.
- Work normally for an extended session, including tool calls and continued turns.
- At an unpredictable point, the CLI prints the
Bad Requestresponse above. - Observe that the session stops progressing and requires a manual resume/retry or CLI restart.
Because the failure is intermittent, there is no single deterministic command that triggers it.
Expected behavior
- A transient API-side bad request should either be retried safely or return a useful, actionable error.
- The CLI should not silently pause the active workflow.
- Ideally the error should include a request/correlation ID and enough sanitized context to diagnose the rejected request.
Actual behavior
- The CLI prints only
{"detail":"Bad Request"}. - Ongoing work stops immediately.
- No request ID, failing endpoint, validation detail, or recovery guidance is shown.
- Manual intervention is required to continue.
Frequency / impact
- Repeated intermittently over approximately two weeks.
- Causes loss of continuity in long-running tasks and can leave background work waiting without a clear failure state.
Environment
- Codex CLI:
0.145.0 - macOS:
26.5.1 - Architecture: Apple Silicon (
arm64) - Shell:
zsh 5.9
Suggested diagnostics
It would help if the CLI logged a sanitized request ID, response status/body, retry classification, and which request field failed validation—without exposing prompt or repository contents.
No private project data, prompts, credentials, or repository details are included in this report.
2 Comments
I’m seeing the same user-visible error in Codex Desktop on Windows:
A useful trigger pattern from my recent occurrences: it happens while Codex is generating HTML files and while generating/exporting PDF files. I have not yet established whether the underlying cause is output size, rendering/conversion, context compaction, or a particular tool result, so I’m reporting this as an observed correlation rather than a root-cause claim.
The next time it occurs I can capture the Codex app version, thread/feedback ID, whether the thread was near compaction, output file size, and the exact tool/action immediately before the error. Please let me know if there are other diagnostics that would be most useful.
Observed again on 2026-08-22 in the macOS Codex IDE during a Backend JavaScript/Prisma editing session. The IDE transport returned
{"detail":"Bad Request"}after a workspace edit. Git inspection immediately afterward showed a coherent dirty worktree andgit diff --checkpassed; focused tests passed after repairing only one independently detected logic defect. No Undo or batch replay was used.Impact: the task required manual continuation despite repository state remaining valid. No credentials, customer data, or repository-specific secrets are included here.