A playable product built through an AI harness.
CS Brasil is both a browser FPS and a test of AI-native product engineering: specialist skills define visual, gameplay, map, regression, and bug-hunt contracts while real usage telemetry closes the loop.
A game exposes weak AI workflows quickly. A change that makes one screenshot prettier can reduce readability, break collision, destabilize frame time, or regress another map. Speed only matters when the evaluation loop can catch that.
- 01
Repository skills encode repeatable contracts for maps, characters, visual quality, gameplay, and bug hunting.
- 02
Independent visual critics inspect screenshots and measurements; builders do not approve their own work.
- 03
Automated issue capture turns player-visible failures into structured GitHub reports for triage and repair.
- 04
Private product telemetry records rounds, maps, characters, scores, and session behavior so iteration follows evidence.
The public alpha became a measurable production system rather than a demo: player behavior informs the backlog, the harness turns evidence into scoped work, and regression checks protect previous gains.