argorix-buildlisted

# ARGORIX Build Put guardrails around the LLM calls in this project, in monitor mode first, and prove they work before changing any behaviour. A guardrail is a decision at a chokepoint: the point where text you did not write enters the model, or where the model's output leaves for a user, a tool, or a database. This skill finds those points, routes them through ARGORIX, and leaves the app running exactly as it did until the operator flips to enforce. ## Before anything **Never put the APP_API_KEY in source, in a committed file, or in a prompt.** It goes in the environment. If the user pastes one into the chat, tell them to rotate it in the console and continue with an env var. Ask them to run the export themselves rather than writing the value anywhere: ```bash export ARGORIX_APP_API_KEY="..." # from the ARGORIX console export ARGORIX_APP_NUMBER="..." # the application number it belongs to export ARGORIX_API_URL="https://api.argorix.com" ``` Both values come from **Applications → your app → Runtime** in the ARGORIX console. If the user does not have them yet, stop and say so: nothing below works without them, and guessing wastes their time. ## Step 1 — Read the project before touching it Do not install anything yet. Establish: 1. **Language and package manager.** `pyproject.toml` / `requirements.txt` → Python. `package.json` → Node; check for TypeScript. 2. **Where the LLM is called.** Search for `openai`, `anthropic`, `langchain`, `llama_index`, `litellm`