Kahozue/thesis-admin-console

# thesis-admin-console Admin console MVP for a privacy-preserving social engineering detection system in enterprise instant messaging (Microsoft Teams). Developed as the implementation artifact for a master's thesis on LLM-assisted threat detection. ## Overview This frontend dashboard supports three analyst roles — Security Analyst (SA), Compliance Manager (CM), and Authorized Re-identifier (AR) — in reviewing pseudonymized chat conversations flagged by the LLM risk scoring pipeline, and authorizing controlled identity re-identification under a strict audit trail. The system deliberately operates on **pseudonymized data only**: real user identities are never exposed unless a full CM approval + AR reveal workflow completes and is logged in the audit trail. ## Architecture Context Microsoft Teams ↓ System A (data collection + PII filtering + token mapping) ↓ pseudonymized text only System B (LLM risk scoring) ← this UI consumes System B output ↓ Admin Console (this repo) — dashboard / alerts / re-identify / audit The PII detection pipeline (System A component) lives in [thesis-pii-pipeline](https://github.com/KahoKozue/thesis-pii-pipeline). ## Tech Stack - React 18.3 + TypeScript 5.4 (strict) - Vite 5.4, Tailwind CSS 3.4, shadcn/ui - react-router-dom 6, TanStack Query 5, Zustand 4 - react-hook-form 7 + Zod 3 - Recharts 2 (line charts, doughnut, custom heatmap) - MSW 2 (Service Worker mocks — no backend required to run) - Vitest 2 + Testing Library 16 (46 unit/component tests) ## Getting Started npm install npm run dev # http://localhost:5173 npm run test # run test suite npm run typecheck # tsc --noEmit npm run build # production build Copy `.env.example` to `.env.local` (defaults work out of the box with MSW mocks): VITE_USE_MOCK=true VITE_API_BASE_URL=/api/v1 ## Demo Walkthrough 1. Open `http://localhost:5173` → redirects to `/login` 2. Select role: - **SA** — direct login, can view dashboard and alerts - **CM** — triggers MFA dialog (enter `000000`) - **AR** — same as CM 3. Full re-identification flow: 1. SA: `/dashboard` → heatmap → click high-risk event → `/alerts/{id}` 2. SA: click "Initiate Re-identification" (requires ≥ 20-char justification) 3. CM login → `/reidentify/queue` → approve → MFA 4. AR login → `/reidentify` → Reveal → second MFA → view real identity 5. AR: "End session & clear" → `/audit` shows 5-entry trail Edge state testing: /dashboard?mockState=empty /dashboard?mockState=5xx /dashboard?mockState=forbidden /dashboard?mockState=delayed ## PII Pipeline Integration When running alongside the PII API server, the Ingest page shows live pseudonymization: # Terminal 1: PII API cd ../thesis-pii-pipeline/pii && python api_server.py # port 8503 # Terminal 2: this app npm run dev Navigate to `/ingest`, load built-in samples, and inspect the alias map showing `PERSON_A`, `EMAIL_B`, etc. ## Project Structure src/ ├── pages/ 11 pages (dashboard, alerts, reidentify, audit, settings, ...) ├── components/ │ ├── ui/ shadcn/ui base components │ └── domain/ heatmap, stability indicator, identity badge ├── features/ │ ├── auth/ roleStore (Zustand) + AR candidate list │ └── dashboard/ risk trend and distribution charts ├── api/ 12 endpoint modules (aligned to spec) ├── types/ graph / analysis / audit / api / role / common ├── schemas/ Zod runtime validators ├── mocks/ MSW handlers + in-memory store + scenario data └── lib/ permissions / alias / dateTime / risk / queryClient ## Implemented Features | ID | Feature | Status | |----|---------|--------| | F-1 | Login + MFA trigger | Done — role switcher replaces SSO; CM/AR require 000000 | | F-2 | Dashboard heatmap | Done — Tailwind grid, click-through to filtered alert list | | F-3 | Risk trend line chart | Done — High/Mid/Low, D/W/M toggle | | F-4 | Risk distribution doughnut | Done — sector click filters alert list | | F-5 | Alert list with pagination and filters | Done | | F-6 | Event context page | Done — pseudonymized, three-panel layout | | F-7 | Threat message highlight + reason sync | Done — bidirectional scrollIntoView | | F-8 | Re-identification request | Done — RHF + Zod, ≥ 20-char reason, audit write | | F-9 | Approval queue + MFA | Done — CM-only | | F-10 | AR reveal (one-time token) | Done — state-only, countdown, beforeunload cleanup | | F-11 | Rejection path | Done — mandatory reason + second MFA | | F-12 | Mark as handled | Done — irreversible, audit write | | F-13 | Audit trail query | Done — SA/AR see own; CM sees all | | F-14 | Report export | Done (CSV) | | F-15 | System settings (CM) | Done (frontend state) | ## Known MVP Limitations | Item | Reason | |------|--------| | Real SSO (SAML/OIDC) | Out of MVP scope; frontend uses role switcher | | Session idle timeout | Requires backend session; reveal 15-min countdown is implemented | | Real backend API | Full MSW mock; Zod schemas ready for direct API wiring | | XLSX report export | Deferred pending spec clarification |

标签：自动化攻击