NEW BUILD PATH · MICROSOFT AGENTS

Build an agent you can explain.

A support agent with a reusable skill, scoped tools and evidence you can defend.

2–3 hours · Guided notebook + terminal · Local tests first, optional real Azure model next.

STEP 01

One project. Two execution modes.

Download the complete project and open lesson.ipynb or the terminal guide. Python 3.12 is recommended. The local simulation uses fictional data and no model or Azure account. It teaches the host boundaries before you connect a real agent.

Extract the ZIP, then open a terminal in 07-microsoft-agent-lab. The first test run deliberately fails until you implement the TODO functions.

Use the notebook or Windows PowerShell

Open lesson.ipynb in your existing Jupyter environment, or install the optional notebook interface:

TERMINAL

python -m pip install jupyterlab
python -m jupyterlab lesson.ipynb

Windows setup:

TERMINAL

py -3.12 -m venv .venv
.venv\Scripts\Activate.ps1
python -m unittest -v test_exercise

TERMINAL

python3 -m venv .venv
source .venv/bin/activate
python -m unittest -v test_exercise

Agent runtime source

Done when The starter tests run and you can identify the unfinished functions.

I completed this step

Sign in to save your checkpoints and notes. They do not certify model quality or cloud deployment.

YOUR AGENT EVIDENCE

Keep the result, not just the demo.

Tests, observed tool calls, failure cases and tradeoffs

Only you can see these notes.

Go to the Microsoft source.

This original learning project is based on Microsoft Agent Framework. Reviewed 2026-10-03; pinned revision e9857bd7. The fictional policies and exercises are ours.

Real SkillsProvider.from_paths and Agent integration example Actual file-skill frontmatter and reference/script convention Approval request and response round-trip using same session Agent Skills Using function tools with human in the loop approvals