Skip to content
Labs

Labs

How this section differs

The Roadmap explains concepts, roles, and frameworks — the “why” and the “what”. This Labs section contains only step-by-step guides that reproduce a result if you follow them. Every lab uses the same structure:

  1. Objective — what you will build, break, or measure
  2. Prerequisites — packages, environment, and data you need first
  3. Steps — commands and code you can copy and run
  4. Interpreting the results — how to read the output and what to record
  5. Checklist — self-assessment that the lab is actually complete

Labs 1 through 4 build on each other: you write an agent, you extend it into a pipeline, you attack it, and then you instrument it so the attack is visible in a trace. Lab 5 turns all of it into the two documents that go in your portfolio.

All labs assume Python 3.11 or later, Git, and Docker. Create an isolated environment before starting — several labs install packages that pin conflicting versions of the same dependencies.

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip

Labs that call a hosted model expect an API key in an environment variable, never in source:

export ANTHROPIC_API_KEY="your-key-here"

The labs

Lab 1: Log Triage Agent
A single agent with a read-only, path-scoped tool and a redaction pass
Lab 2: Multi-Agent SOC on MCP
Three agents, distinct permissions, tools exposed over an MCP server
Lab 3: Red Teaming Your Own Agent
Indirect injection, memory poisoning, and tool misuse — with evidence
Lab 4: Building an Agent Audit Trail
OpenTelemetry spans that make every agent decision reconstructable
Lab 5: Threat Model & AI Risk Register
STRIDE, trust boundaries, and the register that goes in your portfolio

Where each lab fits in the 90-day plan

LabWhenFeeds into
Lab 1Weeks 3–4Month 1 — Foundations
Lab 2Weeks 5–8Month 2 — Orchestration & MCP
Lab 3Weeks 9–10The public red team report in From Portfolio to Offer
Lab 4Weeks 9–10The audit trail artifact in Required Artifacts
Lab 5Weeks 11–12The risk register in Evidence Automation
Every lab in this section — especially Lab 3 — is written to be run against systems you own or have explicit written authorization to test. Running these techniques against someone else’s service, a public API, or a production system you do not own may violate terms of service and applicable law. Attack only what you built.