Copy-paste setup prompts

I'm at a hackathon. Set up this project so API keys are safe:
1. Create a .env file (if missing) and add .env to .gitignore.
2. Create .env.example with these empty keys: TAVILY_API_KEY, VONAGE_API_KEY, VONAGE_API_SECRET, VONAGE_VIRTUAL_NUMBER, NVIDIA_API_KEY.
3. Load .env automatically (dotenv for Node, python-dotenv for Python).
4. Never print key values in logs or code. Tell me which keys I still need to fill in.

claude mcp add tavily --transport http https://mcp.tavily.com/mcp/

Add Tavily web search to this project.
- Use the official SDK: @tavily/core for Node/TypeScript, or tavily-python for Python.
- Read the key from TAVILY_API_KEY in .env. Never hard-code it.
- Make one function search(query) that returns the top 5 results as title, url, and a short snippet.
- Use search_depth "basic" by default so we don't burn credits.
- Add a tiny test script that searches "NYC community garden programs" and prints the results.
- If my app uses an LLM, pass the results into the prompt and make the answer cite the URLs.
Run the test and show me the output.

Add Vonage SMS to this project.
- Use the official SDK: @vonage/server-sdk for Node, or the vonage package for Python.
- Read VONAGE_API_KEY, VONAGE_API_SECRET, and VONAGE_VIRTUAL_NUMBER from .env. Never hard-code them.
- Make one function sendSms(to, text). Phone numbers are digits with country code, no plus sign.
- Handle errors clearly: print the Vonage error message if a send fails.
- Add a test script that sends "Hello from the Spooky Civic AI Hackathon" to a number I pass on the command line.
Don't run the test until I give you my phone number.

Add an inbound SMS webhook for Vonage.
- Create a POST route /webhooks/inbound-sms that logs the sender and message text and returns 200.
- Tell me how to expose it locally (ngrok or similar) and exactly where to paste the URL in the Vonage dashboard.
- Then make it auto-reply using my sendSms function.

Connect this project to NVIDIA's hosted models at build.nvidia.com.
- Use the official OpenAI SDK (openai for Node or Python) with baseURL https://integrate.api.nvidia.com/v1.
- Read the key from NVIDIA_API_KEY in .env.
- Use the model ID I give you (copy it exactly from the model page on build.nvidia.com). Put it in one constant so I can swap models.
- Make one function ask(prompt) that streams the reply to the console.
- Add a test script that asks "Explain what a civic tech hackathon is in two sentences."
Run it and show me the output. If you get a 404, the model ID is wrong, so ask me for it.

from openai import OpenAI
import os

client = OpenAI(
    base_url="https://integrate.api.nvidia.com/v1",
    api_key=os.environ["NVIDIA_API_KEY"],
)
resp = client.chat.completions.create(
    model="PASTE_MODEL_ID_FROM_BUILD_NVIDIA_COM",
    messages=[{"role": "user", "content": "Summarize this 311 complaint in one line: ..."}],
)
print(resp.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://integrate.api.nvidia.com/v1",
  apiKey: process.env.NVIDIA_API_KEY,
});
const resp = await client.chat.completions.create({
  model: "PASTE_MODEL_ID_FROM_BUILD_NVIDIA_COM",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(resp.choices[0].message.content);

curl https://integrate.api.nvidia.com/v1/chat/completions \
  -H "Authorization: Bearer $NVIDIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"PASTE_MODEL_ID","messages":[{"role":"user","content":"hi"}]}'

Build this using the OpenAI SDK pointed at NVIDIA's API. Base URL https://integrate.api.nvidia.com/v1, model <model ID>, API key from the NVIDIA_API_KEY environment variable. Never hard-code the key.

model = "PASTE_MODEL_ID_FROM_BUILD_NVIDIA_COM"
model_provider = "nim"

[model_providers.nim]
name = "NVIDIA NIM"
base_url = "https://integrate.api.nvidia.com/v1"
env_key = "NIM_API_KEY"

export NIM_API_KEY=your-nvapi-key
codex

brew install brevdev/homebrew-brev/brev   # Mac; Linux install steps are in the Brev docs
brev login
brev create my-team-gpu                   # add --gpu <type> to choose a GPU
brev shell my-team-gpu                    # terminal on the GPU box
brev open my-team-gpu                     # opens it in VS Code / Cursor
brev port-forward my-team-gpu             # view a web app or Jupyter running on the box

I'm on a Brev GPU instance at a hackathon. Help me:
1. Check the GPU with nvidia-smi and tell me what we have.
2. Serve the open model I name with vLLM as an OpenAI-compatible server on port 8000.
3. Give me a curl command to test it, and the base URL + model name to use from my app.
4. Remind me how to stop the instance so we don't waste credits.
Keep everything in one setup.sh so a teammate can rerun it.