That's the whole brief. The app can be anything you want: a tool, a dashboard, a side project you've been meaning to start, a game, an internal utility, a weird experiment. The only constraints are:

• AWS Kiro is the IDE. Spec mode or vibe mode, your call. Use specs, hooks, steering — whatever fits the project. The build happens inside Kiro.

• Kane CLI verifies the app. At a minimum at demo time. The strongest submissions wire Kane into Kiro directly, a hook that fires Kane on save, an agent that reads Kane's NDJSON output and iterates, a spec that compiles down to Kane flows.

• It runs, it ships, and you can demo it live at the end of the day.

You can work solo or in teams of up to four. Bring a laptop, your usual stack, and your AWS account. Everything else is on us.

When you walk up to demo, your project needs to clear three bars. Miss any of them and the judges won't be able to score the rest of what you built.

The app works end-to-end. A user can load it, complete the primary flow, and get a result. Not a screenshot, not a mock — the real thing, deployed or runnable locally with one command.

Kane CLI caught something or proved something. You can show one of two things on stage: "Kane caught this bug during the build, here's the failed run," or "Kane verifies these flows pass, here's the green run." Either counts. What doesn't count: Kane installed but never run, or one trivial flow tacked on at the end to qualify.

Kiro and Kane talked to each other. Kiro's output triggered a KaneCLI run, or a Kane result triggered Kiro to do something. The cleanest version: a Kiro hook fires Kane on save, Kane fails, Kiro reads the failure, Kiro edits the code, the next save fires Kane again. Show us that moment. The tighter the loop, the higher the score.

Each team gets a 3-minute demo and 2 minutes of questions in front of a panel of TestMu AI engineers. Scoring is on four dimensions, weighted equally:

A polished todo app with one KaneCLI flow tacked on at the end will lose to a weirder, scrappier thing where a Kiro hook fires KaneCLI and Kiro re-prompts itself based on what Kane finds. Both are valid submissions. The second is the one we're really hoping to see.