MatchDay Command is a demonstration app for the Google PromptWars competition. All match-day data is simulated — there is no real user data, no authentication, and no personal information by design. Still, the app follows sensible web-security practice.

Open a private security advisory on the GitHub repository, or email the maintainer. Please do not file public issues for security reports.

• Security headers (next.config.ts): X-Frame-Options: DENY, X-Content-Type-Options: nosniff, Referrer-Policy: strict-origin-when-cross-origin, Permissions-Policy, and a production Content-Security-Policy.
• Input validation: every API route validates its request body and returns a 400 on malformed input rather than throwing; free-text fields are length-bounded.
• Rate limiting (src/lib/rate-limit.ts): the AI and mutating routes are rate-limited per client to prevent a single caller from exhausting Gemini quota or disrupting the shared demo (429 with Retry-After).
• Constrained AI actions: the copilot can only invoke a fixed allowlist of tools (open incident, dispatch task, generate briefing), executed client-side through an exhaustive switch — the model cannot induce arbitrary actions.
• Secret handling: GEMINI_API_KEY is read server-side only and never sent to the client. The Google Maps browser key is delivered at runtime and must be HTTP-referrer + API restricted.
• No injection sinks: no dangerouslySetInnerHTML, eval, or new Function; React escapes all rendered content. The only outbound fetch targets fixed, non-user-controlled hosts (Gemini, Open-Meteo).

• The app is deployed as a single Cloud Run instance (--max-instances 1); match-day state and rate-limit windows are in-memory and reset on restart, which is intentional for the demo.
• Simulation controls (/api/sim) are intentionally global and unauthenticated — the demo is a shared control room, not a per-viewer session.