Skip to content
Advanced trackVerified 2 Nov 2026

From good prompts to governed agents

Four strands and a capstone for people who already use Copilot daily. Every module comes with a small, disposable repository: break it, fix it with Copilot, and let npm run verify decide.

Your progress0 of 5 knowledge checks passed
Start: Prompt patterns that hold up

New here? Do the nine core modules in Learn first.

How the labs work

Node.js 22+ and Git, no dependencies, no network. Each lab starts failing on purpose and ships with a rubric. The free pages teach the method; Pro adds the lab ZIP and reference solutions.

Labs ZIP (Pro)
  1. Strand A1 · 12 min read · 60 min practice

    Prompt patterns that hold up

    Contracts, context, critique, schemas and untrusted input, scored by tests.

    You'll be able to: Write a task contract with acceptance criteria, constraints and a verification command before asking for code.

    Before this: Ask, Plan, Agent & Autopilot, Custom instructions

    FreeSix patterns, worked example, rubric, check

    ProLab repository, reference solution and walkthrough

    Open
  2. Strand A2 · 10 min read · 75 min practice

    Build and test a custom agent

    A narrow agent with least-privilege tools, proven by five eval cases.

    You'll be able to: Decide whether a job needs a custom agent, a prompt file, a skill or an instruction.

    Before this: Custom agents, Permissions & terminal sandbox, Prompt patterns that hold up

    FreeDesign rules, worked example, rubric, check

    ProLab with linter and eval harness, reference agent

    Open
  3. Strand A3 · 12 min read · 75 min practice

    Issue to pull request, with guardrails

    Agent-ready issues, safe workflows, setup steps and review.

    You'll be able to: Write an agent-ready issue with testable acceptance criteria and an explicit out-of-scope list.

    Before this: Dev Tunnels & GitHub Environments, Build and test a custom agent

    FreeRules for setup steps and CI, rubric, check

    ProLab with workflow checker, runbook and solution

    Open
  4. Strand A4 · 14 min read · 90 min practice

    Govern agents, MCP and secrets

    Intake, pinning, masked secrets, policy layers and honest metrics.

    You'll be able to: Run an MCP server through an intake review and pin the approved version.

    Before this: Permissions & terminal sandbox, Issue to pull request, with guardrails

    FreeControl layers, worked example, rubric, check

    ProLab with checkers, playbook and threat-model answers

    Open
  5. Capstone · 8 min read · 150 min practice

    Capstone: ship a reviewed change

    Fix a bug, add a feature, and catch a bad agent patch.

    You'll be able to: Deliver a bug fix and a feature with acceptance tests written before the code.

    Before this: Prompt patterns that hold up, Build and test a custom agent, Issue to pull request, with guardrails, Govern agents, MCP and secrets

    FreeBrief, rubric and review method

    ProSeeded repository, reference tests and model submission

    Open