The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python script (scripts/create_threejs_game.py) intended to be executed by the agent to scaffold new game projects. The script performs local filesystem operations such as directory creation, copying asset templates, and updating package.json with a sanitized project name. It also references a sibling utility directory (threejs-qa-release) for visual inspection tools.
[INDIRECT_PROMPT_INJECTION]: The core workflow involves the agent interpreting user-defined design briefs, objectives, and mechanics to implement gameplay logic. This ingestion of natural language instructions to drive code generation represents an inherent attack surface for indirect prompt injection.
Ingestion points: User-provided design briefs, primary verbs, and game objectives defined in natural language prompts.
Boundary markers: The workflow mandates a structured Design Brief and Core Loop Contract (references/game-design-level-design.md) to define the implementation scope.
Capability inventory: Local filesystem write access via the Python scaffolding script and suggested shell command execution for project initialization (pnpm install, pnpm dev).
Sanitization: The skill emphasizes rigorous verification including automated bot playtests (tests/bot-playtest.template.ts), visual regression checks, and diagnostic monitoring (window.__THREE_GAME_DIAGNOSTICS__) to ensure implementation matches design intent.

threejs-gameplay-systems