Task: Optimize Robot configuration editor controls

Identity

Use the signed-in Xiaozhi role-configuration page as a read-only interaction reference for the Makelore Robot configuration editor.
Replace avoidable free-text and numeric entry with selection-oriented controls while preserving every supported public configuration field and clear_fields behavior.
Add the smallest safe option-catalog seam required for dynamic model, language, and voice choices; do not expose Xiaozhi credentials or arbitrary upstream paths to Renderer.
Add focused Renderer/API/route tests and preserve the existing revision-conflict, idempotency, error-redaction, and credential boundaries.

Nickname and system prompt remain free-form; VAD/ASR/LLM/SLM/VLLM/Intent/Mem/TTS/language/voice should follow Xiaozhi's selection pattern when an authoritative option source exists.
TTS volume, rate, and pitch must be adjustable without keyboard entry and must retain the full supported integer range, including existing non-preset values.
Dynamic choices must use stable machine values with human-readable labels, preserve the currently configured value, and fail safely when catalogs are unavailable.
The page must remain usable at the existing desktop viewport, with accessible labels, keyboard interaction, busy/error states, and at least 40px interactive targets.
This feature task owns only its isolated worktree and task-scoped record; canonical project memory remains integration-owned.

Verify the Xiaozhi UI and option sources, then define the minimal safe catalog contract.
Implement selection controls and dependent option behavior without weakening update validation.
Add TDD coverage for catalogs, selected values, clearing, numeric controls, and unchanged conflict recovery.
Run focused/full checks, obtain independent review, then integrate through the existing serialized main task.

Added a strict Renderer catalog API and fixed Electron Main proxy route for Xiaozhi model/voice choices without exposing provider credentials or arbitrary upstream paths.
Reworked the Robot configuration dialog into grouped Basic, Model, and Language/Voice sections. Model IDs, language values, voices, and chat-history modes now use selects; SLM reuses the LLM directory.
Replaced numeric TTS inputs with bounded -100..100 sliders and explicit provider-default controls while preserving nullable clear_fields semantics.
Preserved unavailable current values, added safe catalog retry/error states, refreshed dependent voices when TTS changes, and retained existing revision rebase and operation-identity behavior.
Applied private, no-store and no-cache to every local catalog success/failure response, and kept all new retry/default controls at a minimum 40px interaction height.

corepack pnpm exec vitest run tests/unit/ai-hardware-api.test.ts tests/unit/ai-hardware-routes.test.ts tests/unit/ai-hardware-page.test.tsx tests/unit/module-navigation.test.tsx tests/unit/main-layout-module-gate.test.tsx -> 5 files, 84 tests passed.
corepack pnpm run typecheck -> passed.
Focused ESLint on the six changed TypeScript/TSX files -> passed.
corepack pnpm run build:vite -> passed; only the existing chunk-size/dynamic-import warnings remained.
git diff --check -> passed with Git line-ending warnings only.
Independent Sol review -> PASS after cache-boundary and 40px target fixes; no P0-P3 findings remain.

Deploy the matching Xiaozhi and Works Square catalog endpoints before releasing this client; catalog failure is safe and preserves the current configuration, but choices cannot populate until both services are updated.

Target: .project-docs/20-architecture/system-overview.md and Robot integration contract. Proposal: record the USER-scoped safe configuration-catalog flow from Xiaozhi through Works Square Main-owned credentials to Makelore. Evidence: the three task records and focused cross-layer tests. Future impact: model/voice selections remain dynamic without exposing provider configuration. No semantic conflict identified; integration owner decides promotion.