Use when implementing the Detective Agent FE investigation desk with natural-language dialogue, BE/SSE integration, accessibility, clean components, and frontend observability.
설치
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Use when implementing the Detective Agent FE investigation desk with natural-language dialogue, BE/SSE integration, accessibility, clean components, and frontend observability.
Build the FE as a single-screen investigation desk that makes natural-language questioning feel central while reflecting BE session truth and SSE events.
Visual Target
The FE must match target/chatgpt-shared-detective-interface.png as closely as practical. This target image is the primary UI spec for the MVP.
Required first-screen layout:
Dark noir full-screen dashboard with near-black page background, subtle blue-gray borders, glassy panels, and red danger/accent color.
Header: scale icon + 알리바이 교차검증형 추리 게임 on the left; 사건 파일, 증거 목록, 메모, settings gear on the right.
Left suspect column titled 용의자: portrait cards, active card red border, status dot, suspect name/role/description, red active status text, bottom 인물 관계도 button.
Center interrogation stage titled 심문 대상: <name> (<role>): cinematic background image, large character art, speech bubble, natural-language input, red send button, example prompt, and 증거 제시 button.
Right evidence column titled 증거: 3-column image tile grid with labels, locked evidence tile, then red-bordered 모순 사항 list.
Bottom system-flow strip: 사용자 입력 → Character Agent → Light Rule Check → GameMaster Agent → 결과 기록 with icons and arrows.
The above elements should be visible on initial/default load without requiring hidden navigation.
Implementation guidance:
Use the image as a pixel/spacing/color reference, not just conceptual inspiration.
Prefer game-like composition and cinematic panels over generic web/admin UI.
If assets are missing, use existing public/assets/* images and dark overlays/placeholders while preserving the same composition.
After changing UI, run the app/build and compare the first screen against the target; report remaining visual mismatches.
Must Preserve
Do not regress to forced question buttons as the primary UX.
FE never calls AI directly.
FE never stores or infers hidden culprit/solution truth.
Suggested questions are hints/examples only.
First loaded screen should visibly include selected speaker, conversation area, input, and send action.
The first loaded screen must visually resemble target/chatgpt-shared-detective-interface.png before secondary pages/features are optimized.
Development Flow
Read AGENTS.md, Docs/implementation.md, ../PRD.md, and ../Docs/structure-audit.md.
Inspect target/chatgpt-shared-detective-interface.png, src/App.tsx, src/api.ts, src/types.ts, src/storage.ts, and CSS before editing.
Keep API contracts typed and centralized.
Use reducers/adapters for SSE/session payload changes.
Add accessible status labels for pressure, new evidence, event connection, and errors.
Run npm run build before reporting done.
Visually compare the rendered first screen with the target image and list remaining differences.
Observability Checklist
Central logging utility exists before logs spread through components.
Dialogue submit logs metadata only: IDs, duration, result, text length; not full free text by default.