Engineering an AI Agent that Ships Real Roblox Games

Описание и применение

Инженерный подход к созданию AI‑агента, который автоматически строит, тестирует и публикует полноценные Roblox‑игры, используя библиотеку проверенных компонентов, строгую типизацию и автоматический цикл исправления ошибок.

Задача

How to produce shippable Roblox games from an LLM‑driven agent while ensuring reliability, testability, and deployability.

Результат

The agent shipped three monetized Roblox games that survived real‑device testing, demonstrating that a library‑based, type‑checked, automated playtest loop can reliably produce shippable content.

Рабочий процесс

• 1. Agent loads a library of 136 tested kits.
• 2. Agent composes a game inside Roblox Studio via a local WebSocket.
• 3. CI runs luau‑lsp strict type checks on all kits.
• 4. Agent snapshots the place, runs a 30‑second playtest, collects errors, and sends fixes back to the LLM.
• 5. After three failed attempts, the agent restores the snapshot.
• 6. Successful build is built with Rojo and deployed via a curl POST to the Roblox API.
• 7. Post‑deployment tasks (icons, thumbnails) are handled manually or via dashboard.

Доказательства

The agent has now shipped three Roblox games: a pet daycare, an outback tycoon, and a hidden‑object island hunt.

Источник

Every kit passes luau‑lsp analyze in strict mode on CI, with a Rojo‑generated sourcemap so cross‑kit requires resolve.

Источник

The newest piece is a self‑repair loop. After the build phase, the companion snapshots the place, runs the game with RunService:Run() for 30 seconds, collects every error since the run started, classifies them, and hands them back to the LLM to fix with a script update.

Источник