llm-wiki-and-gemma4-tests

Source code, prompts, and outputs accompanying two YouTube videos.

Videos

| Gemma 4: 8 Real-World Tests — the demo app, prompts, and model outputs | | | LLM Wiki — a living, agent-maintained wiki for any codebase (uses this repo as the demo) | |

The .agents/ directory contains the portable LLM Wiki package (skills, workflows, rules) that any Antigravity user can drop into their own repo. The wiki/ directory is the generated output. See .agents/README.md for details.

What's in this folder

Tests 1–5 outputs are shown as screenshots/images in the video. Tests 6–8 full responses are in outputs/.

Running the Demo App

Requirements: A Google AI Studio API key (free at aistudio.google.com)

The app reads env.txt at startup and pre-fills the API key field. Alternatively, paste your key directly into the key field in the UI. env.txt is gitignored so your key is never committed.

Enter your API key in the top-right field. Select a model, mode (Text or Image), and optionally enable Thinking Mode.

Test Results Summary

Tests we expected to fail — but didn't

Tests 4, 5, 7, and 8 were designed expecting Gemma to struggle. It didn't. The video tells the honest story — results updated in real time as we ran each test.

Test 6 caveats (passed, but with code quality issues):