The Agent Skills Directory

[PROMPT_INJECTION]: The "FINAL STEP" section utilizes a hardcoded user critique ("It isn't perfect enough...") to manipulate the agent into a specific behavioral loop of obsessive refinement. This constraint is used as a prompt engineering technique to ensure high-quality output.
[PROMPT_INJECTION]: The skill processes external user input to derive conceptual artistic themes, representing a surface for indirect instructions.
Ingestion points: User input provided in the initial request (SKILL.md).
Boundary markers: Absent; instructions encourage the agent to use input as a creative foundation.
Capability inventory: Local file creation of .md, .pdf, and .png documents.
Sanitization: No specific validation or sanitization of conceptual input is performed.
[EXTERNAL_DOWNLOADS]: The instructions permit the agent to download external font assets as needed. The local canvas-fonts directory contains licenses associated with well-known and trusted organizations such as Google, IBM, Vercel, and Red Hat, indicating the assets are from reputable sources.
[SAFE]: The skill metadata and core instructions align with its stated purpose of visual art creation without requesting unnecessary system permissions or tools.

space-canvas-art