Project: grounded runbook assistant

New teammates at esc bash keep asking the same questions that are already written down in the team runbook. Build an assistant that answers those questions from the runbook itself, and is honest when the answer is not there.

Set up the project

Run this to drop the team runbook at /root/runbook.md:

curl -fsSL https://raw.githubusercontent.com/Esc-Bash/project-init-scripts/main/ai/project-3/init.sh | bash

Read it first. It holds specific facts (rollback steps, who owns what, an incident phone code) that a model could not possibly guess.

What to build

• Write an assistant at /root/ask_doc.py that reads the whole runbook, takes a question, and answers using only what the runbook says. Instruct the model plainly: answer from the document, and if the document does not cover it, say so rather than guessing.
• Run it with a question whose answer is in the runbook (for example, how long a rollback takes, or who owns the billing service) and save the reply to /root/answer.txt.

This is grounding in action: the model has no idea what your team's runbook says until you put it in the prompt. Everything you learned about giving the model the facts and not trusting it to invent them comes together here.

Press Submit once /root/ask_doc.py exists and /root/answer.txt holds an answer grounded in /root/runbook.md.

Your lab setup

This VM comes preconfigured with your AI keys - just type claude to get started. Your key and URL are already set as $OPENAI_BASE_URL and $OPENAI_API_KEY.