backend

docs/superpowers/plans/2026-07-30-mode-profile-library.md

@@ -0,0 +1,26 @@
+# Mode Profile Library Plan
+
+## Goal
+
+Add a long-term cabin mode configuration layer that can be matched from natural-language requests before normal CAN compilation.
+
+## Scope
+
+- Add repository-backed mode YAML files under `config/modes/`.
+- Convert matched mode operations into the same model-plan requests already used by seat, climate, roof, and ambient-light adapters.
+- Preserve current behavior for unmatched or extra user goals by letting the AI/local parser generate temporary requests.
+- Allow a recent successful action set to be saved as a long-term mode profile when the user explicitly asks to solidify a mode.
+- Expose available modes to the web console and wire the existing quick-mode/save controls to the new workflow.
+
+## Tests First
+
+- Unit-test mode profile loading, matching, request conversion, and save-to-YAML behavior.
+- Integration-test zero-gravity matching alone.
+- Integration-test zero-gravity plus a temporary roof request producing a cross-domain `SceneConfig`.
+- Integration-test saving the previous successful action set as a long-term mode file without emitting CAN.
+
+## Design Notes
+
+- The mode library stores semantic operations, not CAN frames. CAN remains generated by the existing domain compilers.
+- Seat-related operations in modes still pass through vehicle-profile angle/percentage calibration.
+- Save-mode detection is intentionally narrower than seat posture preset saving to avoid changing existing preset behavior.