[Bug] Effort mode setting "max" not propagated to UI and inference runtime

Bug Description
Bug Report: Effort Mode Mismatch After Setting "max" Summary After setting the thinking effort mode to max via /model command, the UI footer continues to display high effort, and runtime behavior reflects high rather than max. Environment - Product: Claude Code CLI v2.1.119 - Model: Opus 4.7 (1M context) (default) - OS: macOS (Darwin 25.3.0) - Date observed: 2026-04-26 Steps to Reproduce 1. Launch Claude Code v2.1.119. 2. Run /model and select Opus 4.7 (1M context) with effort max. 3. Observe the confirmation banner: ▎ Set model to Opus 4.7 (1M context) (default) with max effort 4. Look at the bottom-right status indicator in the prompt footer. 5. Submit any task that triggers extended thinking. Expected Behavior - Footer status should display ● max · /effort. - Thinking-phase telemetry (e.g. Cooking… line) should reflect max effort budget. - Subsequent reasoning should use the max thinking budget. Actual Behavior - Confirmation banner correctly reports max effort, but: - Footer still shows ● high · /effort. - Runtime telemetry shows Cooking… (35s · thinking more with high effort) — i.e. the engine is operating in high mode, not max. - The setting appears accepted at the config layer but is not propagated to either the UI status component or the inference runtime. Evidence Attached screenshot shows: - /model confirmation: Set model to Opus 4.7 (1M context) (default) with max effort - Footer indicator: ● high · /effort - Thinking line: Cooking… (35s · thinking more with high effort) Impact - User cannot actually invoke max effort mode despite the command appearing to succeed. - Silent downgrade to high is misleading — there is no warning that max was rejected, clamped, or unsupported. - Affects task quality on prompts that explicitly require deeper reasoning budget. …
Note: Content was truncated.