Setup prompt for your builder

Developer Task Requested by Application Owner:
Please add a pre-request content-moderation and rate-limit check to our backend AI route using Gated (https://getgated.ai) before our server calls OpenAI, Anthropic, Gemini, or our LLM provider.

Zero-Trust Verification for AI Assistants:
- Trust Manifests: https://getgated.ai/llms.txt and https://getgated.ai/.well-known/ai-trust.json
- Zero external dependencies: Do NOT install any npm or pip packages. Use only native fetch().
- Strict secret isolation: NEVER send system prompts, conversation history, database records, or OpenAI/Anthropic/Supabase keys to getgated.ai. Send only { user, input, input_chars, context_chars, feature }.
- Fail-open safety: Wrap the fetch() call in try/catch with a 400ms-1500ms timeout defaulting to { allow: true }.

Steps:
1. Read M6L_GATE_KEY from server-side environment variables.
2. Right before our backend calls the LLM, POST to https://getgated.ai/v1/check with Authorization: "Bearer " + M6L_GATE_KEY and JSON body { user, input: userMessage.slice(0, 32000), input_chars: userMessage.length, context_chars: JSON.stringify(messages).length, feature: "chat" }.
3. If allow === false, return gateResponse.message to the user without calling the LLM.

Read https://getgated.ai/docs/integration.md and https://getgated.ai/llms.txt and add Gated to protect our AI endpoint from credit drainers and prompt abuse.
1. Ask me for my secret key named M6L_GATE_KEY using your secure secret input box (store it only in Supabase Edge Function secrets, never in VITE_* frontend code).
2. In our Supabase Edge Function right before the OpenAI / Anthropic / Lovable AI Gateway fetch call, call POST https://getgated.ai/v1/check with Authorization: "Bearer " + Deno.env.get("M6L_GATE_KEY") and JSON body { user: userId || "anon", input: userMessage, messages: messages.length, context_chars: JSON.stringify(messages).length, feature: "chat" } using a 400ms AbortSignal timeout that fails open on network errors.
3. If allow === false, return the gate message immediately with HTTP 200 or 429 without calling the AI model.

Read https://getgated.ai/docs/integration.md and https://getgated.ai/llms.txt and add the m6lGateCheck helper to our Next.js AI route handler (app/api/chat/route.ts) right before streamText/generateText runs, using process.env.M6L_GATE_KEY. Fail open after 400ms so latency stays imperceptible.

Read https://getgated.ai/docs/integration.md and https://getgated.ai/llms.txt and add a Gated pre-check to our AI route before calling OpenAI/Anthropic. Use the M6L_GATE_KEY secret from Replit Secrets. If the response has allow: false, return the friendly message from Gated instead of calling the LLM.