The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it is designed to ingest and analyze untrusted external data (such as essay text, blog posts, or datasets) to drive the visual generation process.
Ingestion points: Workflows/Essay.md and Workflows/Visualize.md directly process user-provided text content to determine the emotional register, metaphors, and composition of generated images.
Boundary markers: The workflows do not utilize explicit delimiters or instructions to ignore potential directives embedded within the ingested content.
Capability inventory: The agent can execute shell commands (magick, cwebp) and perform network operations to several AI service providers.
Sanitization: While filename generation logic includes character replacement, the content used to build LLM prompts for image generation does not appear to have robust escaping or sanitization.
[COMMAND_EXECUTION]: Multiple tools within the skill (Tools/Generate.ts, Tools/ComposeThumbnail.ts) execute system commands using the spawn method to perform image processing tasks.
Evidence: These scripts invoke magick for cropping and resizing, and cwebp for converting images to WebP format. These operations depend on parameters derived from user instructions or input filenames.
[EXTERNAL_DOWNLOADS]: The skill interacts with several external AI platforms and image processing APIs to fulfill its primary purpose.
Evidence: It sends prompts and images to, and receives data from, Replicate, OpenAI, Google GenAI, Discord (for Midjourney integration), and remove.bg. These services are recognized as well-known technology providers.

Art