00-prerequisites

Chapter 00 — Prerequisites

A few things to install before chapter 1. Should take about 10 minutes if you've never used these CLIs before, less if you have.

What you'll do here

• Install Claude Code, Codex CLI, and Gemini CLI
• Get API keys for all three
• Check your budget (the goal: stay under $5 for the whole course)
• Run check-env.sh to confirm everything works

Install the three CLIs

Claude Code

Official install guide.

npm install -g @anthropic-ai/claude-code
claude --version  # should print a version

Codex CLI

Official install guide.

npm install -g @openai/codex
codex --version

Important: this tutorial requires codex-cli v0.125+. Earlier v0.121 shipped with gpt-5.5 as the default, which is no longer supported. Upgrade if you have an older version.

Gemini CLI

Official install guide.

npm install -g @google/gemini-cli
gemini --version

Set API keys

# In your ~/.bashrc or ~/.zshrc:
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GEMINI_API_KEY="..."

If you use ChatGPT account auth for Codex CLI instead of an API key, follow Codex's docs for the OAuth flow. Note: ChatGPT-account Codex can't use the bare gpt-5 model, so use the default ChatGPT-allowed model.

macOS users: install GNU timeout

The scripts in this repo use timeout to bound how long any one LLM call can run. macOS doesn't ship GNU timeout by default:

brew install coreutils    # this gives you `gtimeout`

The scripts auto-detect either timeout or gtimeout, so once you've installed it you don't need to do anything else.

Check your environment

bash check-env.sh

Expected output:

✓ Found timeout: /opt/homebrew/bin/gtimeout
Checking Claude Code...
Checking Codex CLI...
Checking Gemini CLI...
✓ All checks passed.
  Add to your shell rc to skip auto-detect on every run:
    export REDTEAM_TIMEOUT_CMD=/opt/homebrew/bin/gtimeout

If anything fails, the script tells you which command and what to install.

Budget

Per full red-team run on a typical plan:

Plan size	Tokens (3 models)	Cost (USD)
Small (sample-plan.md, ~50 lines)	~5k in, ~5k out	~$0.10–0.20
Medium (chapter 4 BQ plan, ~100 lines)	~10k in, ~12k out	~$0.30
Large (chapter 5 GCP plan, ~150 lines)	~15k in, ~15k out	~$0.50

The whole course (chapters 1–6, about 6 runs total): under $5.

Things that go wrong

A short list of things readers have actually hit:

• macOS without coreutils. Error: timeout: command not found. Fix: brew install coreutils.
• Codex CLI v0.121 or older. Error mentioning "model not supported". Fix: upgrade to v0.125+.
• ChatGPT account using gpt-5 directly. Returns a 400. Use the default ChatGPT-allowed model instead.
• Gemini CLI in some regions. First call fails with "API not enabled". Fix: enable the relevant API in your GCP console.
• Windows native (no WSL). Bash & and wait parallel syntax isn't reliable there. Use WSL.

Next

Chapter 01 — Why one LLM isn't enough

← Back to README · 中文版