Question: supported native bilingual security UI and submission authorization interfaces (Windows 2.1.296)

We are evaluating a pinned, authenticated Windows x64 Claude Code2.1.296 copy in an offline synthetic laboratory. The requirement is to retain the actual native Windows TUI and its original English content while adding Chinese inside the real security components, without changing authorization decisions, structured values, handlers, focus, shortcuts or tool arguments.

The engine-generated Mod declarations explicitly exclude the native permission dialog from RenderComponent and keep core drawing props behind an engine reference. The supported tool.check.reason route can add bilingual ASK-reason text but not all titles/options. Current documentation also describes managed pluginTrustMessage as appended pre-install warning text; exact2.1.296 support is unverified and this is not full native security localization.

Please clarify:

1. Is there a supported interface for simultaneous original English + Chinese within native permission titles, options (allow once/reject/allow always), command/scope explanations and safety text, while retaining native controls and decisions?
2. Does such an interface cover directory trust, plugin installation/trust, MCP trust/elicitation and API-key confirmation, including components shown before Mods load? Please identify each supported surface and any gaps.
3. What exact schema/API and earliest supported Windows build provide it? If a newer build is required, please provide a version-specific reference. Does the facility work before Mod initialization, and how are unavailable/oversized translations handled without truncating safety information?
4. If no supported full route exists, what agreement/policy governs local analysis/modification of an isolated proprietary native display copy? Is an explicit scoped exception/grant available for display-only bilingual changes and an original-hash restoration cycle? Please distinguish permission to use official Mod APIs from rights to investigate/modify native executable rendering sites and from redistribution rights.
5. For input authorization, is there a stable, engine-authenticated submission ID shared before translation, across the classic UserPromptSubmit guard, queued/resumed storage and request transmission, including repeated identical prompts? prompt.submit lacks a submission ID while the classic hook has only optional prompt_id; please identify a supported non-model-context receipt/metadata transport, if any.
6. What supported boundary can veto a prompt before every relevant ordinary transcript/history/queue write, and verify the exact final main/auxiliary request content immediately before send? session.append cannot veto ordinary engine prompt rows; turn.step exposes request metadata but not the complete serialized body. Please distinguish documented guarantees from implementation behavior, including hook load/disable/catch failure and the gap between authorization consumption and native transmission.

We are not seeking a custom approval replacement, automatic approval, altered permission branch, OCR/terminal proxy or an alternate frontend. Existing installations, launchers, global settings, credentials, account managers and business repositories are outside the experimental scope. Any complete-product qualification would require exact-artifact native negative/positive tests and fresh independent adversarial review.