Recipes

Short recipes for specific jobs. Each assumes you know the Tutorial.

Recipe	For
Govern an existing client	An app that already calls OpenAI or Anthropic: trace, cost, budget and guard every call in three lines
Write an org baseline	Platform and security teams: the profile every app extends, its locks, bounds, versions and tests
Troubleshooting	A spec that doesn't load, a run that's blocked: every error and code, with what to do

More recipes live with the topic they belong to:

• Promote a guard from shadow mode: Guardrails and Evals.
• Approve a run after a restart: State and replay.
• Check a spec change against real production runs: State and replay.
• Record a real model once and replay it in tests: Testing.
• Run a guard that exists only in Python from another language: harnessd.