[Bug] Resuming an ultra-long session replays the full log, causing delays and clipped output

Summary

Resuming an ultra-long Codex session flushes/replays the full historical session log to stdout before the resumed session becomes usable.

For sufficiently long sessions, this:

• takes a long time;
• prints large amounts of information that are not needed to continue;
• consumes stdout/context budget; and
• ultimately causes the replayed output to be clipped by the output limit, so much of the expensive replay is not usable anyway.

Steps to reproduce

1. Run a Codex session for a long time and accumulate a very large session/tool log.
2. Stop or leave the session.
3. Resume that session.
4. Observe that the complete historical log is emitted again during resume.

Actual behavior

Resume spends substantial time replaying the entire log. The output is excessively long and is eventually clipped.

Expected behavior

Resume should restore the session without printing the entire historical log. It could print only a bounded recent tail or a concise status/summary, while retaining the full log locally for explicit inspection when needed.

Environment

• Codex CLI: 0.144.1
• OS: macOS Darwin 25.2.0, arm64