Patch Changes
- Route every Claude Code inference request through the production LFM2 model-tier and ArchRouter thinking worker. The router now selects and applies the Claude model and effort together, preserves compact prior-turn context, and displays both choices in the submitted-prompt hook.