The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to process untrusted user input (e.g., scene descriptions) to automate scene building and verification. This presents an indirect prompt injection surface.
Ingestion points: User instructions for scene generation (e.g., "Create a scene", "Make a character") ingested in SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between its own logic and potentially malicious instructions embedded in user-provided node names or paths.
Capability inventory: File creation (mcp_godot_create_scene) and process execution (mcp_godot_run_project) via MCP tools.
Sanitization: There are no instructions to sanitize or validate user-provided strings before they are used as arguments for the scene building tools.
[COMMAND_EXECUTION]: The skill utilizes MCP tools that execute external processes, specifically mcp_godot_run_project and mcp_godot_launch_editor, to spawn the Godot editor. While these are legitimate tools for the skill's purpose, they represent a capability that could be abused if the agent is misled by malicious input.

godot-mcp-scene-builder