The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically constructs file paths to load style-specific configuration and assets. It is instructed to read metadata and style definitions from paths like ../../styles/{style-id}/META.md and ../../styles/{style-id}/STYLE.md. Using computed paths based on variable identifiers (style-id) derived from user or agent selection can lead to unauthorized file access or directory traversal if not strictly validated against a known-good catalog.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user content (such as articles or topic descriptions) and interpolates this data directly into a complex image generation prompt template.
Ingestion points: Source material, titles, and topics provided by the user in SKILL.md.
Boundary markers: The final prompt blueprint in references/cover-prompt-blueprint.md does not utilize delimiters (e.g., XML tags or triple quotes) to isolate user-supplied fields such as {title_or_topic}, {summary}, and {visual_subject} from the agent's core instructions.
Capability inventory: The skill leverages an image generation tool and has permission to write files to the local punk-assets/ directory.
Sanitization: While the skill summarizes long articles, it lacks explicit sanitization or escaping mechanisms for user-provided strings before they are injected into the final prompt context.

punk-cover