Codex automatic context-compaction failure

What version of the Codex App are you using (From “About Codex” dialog)?

Client: Codex macOS desktop 26.814.41407

What subscription do you have?

Pro plan

What platform is your computer?

macOS Tahoe 26.5.2 Darwin 25.5.0 arm64 arm

What issue are you seeing?

Codex automatic context-compaction failure

Affected thread ID:

019fa01b-3db9-7a71-bd21-7292c0230beb

Failure window:

22 Aug 2026, 11:49:36–13:05:57 BST

Model from logs:

gpt-5.6-sol

Reasoning effort:

medium

The thread became stuck repeatedly displaying “Context automatically compacted” without making useful progress.

Local telemetry shows 48 automatic-compaction attempts during the approximately 76-minute failure. total_usage_tokens repeatedly remained approximately 440k–452k despite:

auto_compact_scope_limit=244800

full_context_window_limit=258400

full_context_window_limit_reached=true

token_limit_reached=true

The thread repeatedly attempted to reread the same project skill file and even attempted a no-op true command.

User-visible error:

“I’m blocked by a Codex workspace execution failure: even a no-op tool call is being rejected as an output-overflow error. I therefore cannot safely inspect, edit, test, or release the repositories.”

Representative x-oai-request-id values:

Start:
f7682305-a673-48c0-a2d4-2c19e79f9596

Early:
37d2f50d-0534-42ea-bbe0-bb6a9ce88c94

Middle:
104b6acf-d870-461e-a2f2-596b0d3de0fd

Later:
1a7d3b14-31a6-4044-862b-c53daa75ee16

Near end:
c45ec960-fe8a-4ccf-8749-ca8a43483f64

Final:
7f3b80dd-7ef4-4c85-8e31-339bc2c8b3ab

There are 100 distinct x-oai-request-id values in the affected period.

The server response headers show weekly Codex usage increasing during the malfunction:

11:49:49 BST — x-codex-primary-used-percent: 28

13:05:49 BST — x-codex-primary-used-percent: 34

Both report x-codex-primary-window-minutes: 10080.

I have separately opened a Support case requesting restoration of the weekly allowance consumed by this malfunction. This feedback submission is intended to provide the Codex engineering team with the affected session/logs for investigation.

Thread ID 019fa01b-3db9-7a71-bd21-7292c0230beb
Window: 2026‑08‑22 11:49:36–13:05:57 BST (UTC+1)
Client: Codex macOS desktop 26.814.41407; macOS Tahoe 26.5.2
Model: gpt-5.6-sol + reasoning_effort=medium
Representative x-oai-request-ids (start/middle/end) and the two header excerpts showing x-codex-primary-used-percent 28→34
The exact “output-overflow” error text and that even true failed
The compaction telemetry summary (48 auto-compactions; context staying ~440k–452k vs limits)

What steps can reproduce the bug?

N/A

What is the expected behavior?

_No response_

Additional information

_No response_