Hands-on labs and reusable recipes for cloud-native security and agentic AI. Each lab is a self-contained, runnable demonstration of one idea — small enough to read, big enough to learn from.
| Lab | What it shows |
|---|---|
| Lab_6_1_Model_Armor | Prompt injection defenses and guardrails with Model Armor |
| Lab_8_1_ADK_Observer | Observability for Google ADK agents — notebook + SQL for inspecting agent behavior |
| Lab_4_1_AgentSpace_Security | AgentSpace deployment security (in progress) |
kata/ — single-lesson exercises: one bug, one fix, one test. Walkthroughs on devnull.fyi/labs.
| Kata | Lesson |
|---|---|
| strict-mocks | A mock that answers anything proves only that your code asked |
recipes/ — self-contained Google ADK agent recipes. Each recipe is a single folder you can run locally or deploy to Google Cloud.
| Recipe | Tools |
|---|---|
| hello | Time-telling agent (local tool) |
| search | Grounded research with google_search |
| maps | Places, weather, routes via Maps MCP |
| registry | Discover and invoke registered agents |
| bigquery | Query datasets via BigQuery MCP |
| gcs | List buckets and read objects |
| mcp | Custom MCP server on Cloud Run |
See recipes/README.md for prerequisites and quick start.
Three reasons:
- Showing beats telling. The labs are the proof. Frameworks describe the problem; runnable code demonstrates the answer.
- Sharing what works. Patterns I've used in real deployments, distilled into the smallest reproducible form.
- A place to tinker. New tools, new threats, new ideas — this is the bench.
PRs welcome. See CONTRIBUTING.md. Code of Conduct applies.