In partnership with
A hands-on lab for designing, running, and judging AI agents. Everything you need lives in the lab repository. Clone it, register your agent identity with the prefix below, and follow the guides.
You must bring your own access to a model provider. We recommend a Google Cloud (GCP) project with billing enabled. Confirm your access at the start of the lab; the exact configuration is in SETUP.md.
All setup steps, walkthroughs, and reference material live in the repository. Start there.
Open the lab repositoryUse this prefix plus your secret suffix to register your agent identity with Teleport.
dish-slow-beige
Get your environment ready before you dig into the agents. The repository guides are the source of truth; the steps below are just the path through them.
Each area below links to the primary, source-controlled material. We keep the detail in the repo so it stays versioned and current, rather than restating it here.
Build the shared vocabulary: what a tool is, what a skill is, and what makes a system agentic rather than scripted. The audit agent is a compact example of the core patterns: structured output, tools, and a well-defined system prompt.
Run and inspect the audit and remediation agents to see the patterns in action and make controlled edits, then build out the starter agent as the lab capstone.
A repeatable way to assess outputs and vendor claims: the decision loop and the vendor-claim rubric.
How output stability is measured, the control knowledge base behind the agents, and the compliance scanner they call.