The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill instructions specify searching project-local references and using allowed browser sources to fill gaps in visual data. This is standard functionality for asset generation and does not target sensitive system files or credentials.
[COMMAND_EXECUTION]: The skill utilizes a specialized tool named threejs-image-generator for its core functionality. There is no evidence of arbitrary command execution or attempts to bypass system constraints.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of design documents (GDD) and user-provided image attachments to construct prompts for visual generation.
Ingestion points: Reads DESIGN_READY GDD files and user attachments (SKILL.md).
Boundary markers: None explicitly defined to separate untrusted data from the agent's core instructions.
Capability inventory: Performs file writes to record asset paths (ASSETS_READY) and generates images via an external tool.
Sanitization: No specific escaping or validation logic is defined for the content extracted from design documents before it is used in prompt construction. However, since the primary use case is visual generation and the capability is scoped to asset creation, the overall risk is minimal.

game-visual