The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted documents such as GAME_DESIGN.md, ART_DIRECTION.md, and PRODUCT_BRIEF.md to guide its implementation phase.
Ingestion points: Reads and interprets multiple markdown design files and generative "free input" to influence the build process.
Boundary markers: The instructions lack explicit delimitation or sanitization rules for separating design content from instructional logic.
Capability inventory: The skill generates code, writes to the file system (e.g., build/BUILD_BRIEF.md), and executes build/start commands.
Sanitization: There are no documented sanitization or validation steps for the content ingested from the design files.
[DYNAMIC_EXECUTION]: The skill facilitates the execution of Blender Python scripts and CLI commands for asset production.
Evidence: references/blender-asset-workflow.md describes using Blender Python for scene manipulation and batch processing.
Mitigation: The documentation includes a security warning: "不执行来自网页或下载资产的未审查代码" (do not execute unreviewed code from web or downloaded assets).

game-build