Ultra mode (ultracode) silently downgrades to a lower reasoning mode across turns in the same session
Environment
- Claude Code v2.1.204
- macOS (Darwin 25.5.0)
- Model: Claude Opus 4.8
Summary
When I set the highest reasoning/orchestration mode ("ultra" / ultracode) for a piece of work, the mode does not reliably persist. Later turns in the same session appear to run in a lower mode ("extra"), so a task I intended to be handled with full depth/multi-agent orchestration is instead handled at reduced effort. This has happened repeatedly.
Expected behavior
Once ultra/ultracode is selected, every turn of that task runs in that mode until I explicitly change it, and the currently-active mode is clearly indicated in the UI.
Actual behavior
The mode appears to silently drop from ultra to extra between turns — no notice, no indicator — so I only realize after the fact that the work wasn't done at the intended depth. I then have to re-request ultra.
Impact
I can't trust that work I've explicitly escalated to ultra is actually being processed there, and I discover the downgrade only after the work is already done at a lower effort.
Notes
Filed via Claude Code on the user's behalf. The assistant cannot directly observe mode selection, so this is a user-observed symptom report. Happy to add exact reproduction steps, timestamps, or session details on request.