Pokédex workshop: prerequisites

Read and complete this the day before the session. It is not session time, and none of it should come out of the sixty you have in the room.

Everything below was confirmed by a real run on 2026-08-07 on Windows 11 (10.0.26200). Where a version is given, that is the version the run used, not a minimum — anything close should be fine, and if yours differs and something breaks, that is worth knowing before the session rather than during it.

One thing is still blank, and it is not something you can supply:

(contact name goes here, not yet assigned)

The account in the authentication command is yours and differs per attendee:

(your own Northwell account goes here)

Tools you need installed

--version works on all three — confirmed, not assumed:

1agents-cli --version
2uv --version
3gcloud --version

You do not need a particular Python on your PATH. The project pins >=3.11,<3.14, and uv downloads and uses its own interpreter regardless of what you have. The build run had Python 3.14.4 installed system-wide, which is outside that range, and uv quietly built the virtual environment on CPython 3.12.11 instead.

Authentication

The agent talks to Gemini through Vertex AI, which needs Application Default Credentials, not a plain gcloud auth login. Run both:

1gcloud auth application-default login --account=(your own Northwell account goes here)
2gcloud auth application-default set-quota-project p-d-digex-vertex-001

p-d-digex-vertex-001 is the project the workshop runs against. This was confirmed during the build run: agents-cli scaffold create printed ✓ Connected to project: p-d-digex-vertex-001 without being told which project to use, having picked it up from the credentials above.

If either command errors rather than completing, do not spend the evening guessing at local fixes. Use the contact at the end of this document. A permission denial in particular usually needs an administrator, not a change on your machine.

Environment variables: there are none

Nothing needs setting in your shell. agents-cli scaffold create writes a .env file into the project containing the three values the agent needs:

1GOOGLE_GENAI_USE_VERTEXAI=true
2GOOGLE_CLOUD_PROJECT=p-d-digex-vertex-001
3GOOGLE_CLOUD_LOCATION=global

This was verified by running the agent with all three, and PYTHONIOENCODING, explicitly unset. It worked, including non-ASCII output.

Two things that will interrupt you if you meet them first in the room

The telemetry prompt, twice. agents-cli playground asks Enable telemetry? [Y/n]: in the terminal before it starts, and the browser UI then asks again in a modal. Neither is a trick question and either answer is fine — but answer both once now, so the first thing you do in the session is not read a consent dialog.

Live reload does not work on Windows. The playground starts with --reload_agents and immediately turns it back off:

1WARNING: The --reload flag is not supported on Windows because it forces
2Uvicorn to use SelectorEventLoop, which does not support subprocesses
3(needed for executing tools). Forcing --no-reload.

You will edit app/tools.py several times during the session. Each edit needs the playground stopped and restarted before it takes effect. Nothing warns you at the time; the old code just keeps running.

Prove your setup works

agents-cli info prints CLI and project metadata. It does not check authentication, so a clean run from it proves very little:

agents-cli info

1CLI version:        1.2.1
2OS info:            Windows-11-10.0.26200-SP0
3Deployment target:  none

The command that actually exercises your credentials is the scaffold itself, which is Step 1 of the workshop. If you want to prove your setup end to end before the session, run it in a throwaway directory and delete the result:

agents-cli scaffold create prereq-check --agent adk --prototype

A line reading ✓ Connected to project: p-d-digex-vertex-001 means you are ready. An authentication or permission error means you are not, and that is exactly what this document exists to find out a day early.

One network check

The agent calls PokéAPI (https://pokeapi.co/api/v2/) directly. No key, no signup. It sits behind Cloudflare, so if your machine reaches the internet through a corporate proxy, confirm you can reach it:

(Invoke-WebRequest https://pokeapi.co/api/v2/pokemon/gengar).StatusCode

200 is what you want. Anything else, use the contact below.

(A 403 from your own Python code later in the session is a different problem with a known cause, and the workshop covers it: Cloudflare rejects urllib's default user agent, and the tool you write sends its own.)

If anything above will not complete

Contact:

(contact name goes here, not yet assigned)