Setup
Do this before the workshop if you can. The agent calls Gemini through Vertex AI in your own Google Cloud project. You need a project with billing. Free trial credits work, and the whole workshop makes only a few dozen calls.
Or use Cloud Shell, with nothing to install. Click Open in Cloud Shell. In the dialog, select Trust repo and click Confirm. Cloud Shell copies the repo and opens a tutorial panel next to the editor. It walks you through the setup and every step, and its links open the files for you.
Install the gcloud CLI, then sign in and select your project:
gcloud auth login
gcloud auth application-default login # lets the agent call Vertex AI as you
gcloud config set project YOUR_PROJECT_IDGet the code and run the setup script. On Windows, use Git Bash or WSL.
git clone https://github.com/omarcevi/coding_agent_workshop.git
cd coding_agent_workshop
bash scripts/setup.sh # about a minuteThe script checks the project and its billing, turns on the Vertex AI API, installs uv and the Python packages, writes .env, copies the buggy demo project into workspace/, and makes one call to Gemini. It ends with ✅ Ready. ADK finds the one .env from every agent in the repo.
No billing account? Use a free key from aistudio.google.com/apikey instead. Install uv, then:
uv sync
uv run reset_workspace.py
uv run python scripts/set_key.py # asks for the key; it doesn't show on screenStart the ADK dev UI:
uv run adk web --reload_agents agents
Open localhost:8000 and pick my_agent from Select an app at the top. --reload_agents picks up your code when you save, so you never restart it. Always start it with uv run, so the agent's python -m pytest uses this project's Python.
What's in the repo:
| Path | What it is |
|---|---|
agents/my_agent/ | Your agent. You build it up during the workshop. |
agents/step1_hello/ to agents/step4_safe/ | The finished checkpoint for each step. |
demo_project/ | A tiny shopping cart with two bugs: three of its four tests fail. |
workspace/ | The agent's copy of demo_project/. The agents only touch this folder. |
reset_workspace.py | Puts the original buggy project back in workspace/. |