Codex logging bug can write terabytes of data to local SSDs

An issue has been discovered where Codex's SQLite feedback log database continuously writes an excessive amount of data locally.

In one user's environment, approximately 37 TB was written to the main SSD over about 21 days of uptime, which extrapolates to roughly 640 TB per year. This is enough to exhaust the write endurance rating (600 TBW) of a typical consumer SSD in less than a year.

The main causes are as follows:

• TRACE-level logs account for 70.7% of total data
• Excessive logging from specific targets such as codex_api::endpoint::responses_websocket
• Mirrored telemetry and raw WebSocket/SSE payload recording

Analysis suggests that filtering out just the TRACE and INFO log categories could reduce total log data by about 96%.