The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions include a shell command template that uses curl to call an external API and pipes the result to a Python script. This script writes binary image data directly to the local file system at apps/web/public/images/books/.
[COMMAND_EXECUTION]: The skill directs the agent to modify application source code, specifically updating COVER_MAP in .tsx files located in apps/web/app/books/drafts/ to register new covers.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to Google's Generative Language API (generativelanguage.googleapis.com) to retrieve generated image content based on constructed prompts.
[PROMPT_INJECTION]: The skill ingests untrusted data from external sources, including book.yaml, project outlines, and chapter text, to inform the design process. This creates a surface for indirect prompt injection.
Ingestion points: Reads book.yaml, book outlines, and individual book chapter files.
Boundary markers: No explicit delimiters are used to wrap or isolate the untrusted book content during the analysis phase.
Capability inventory: Execution of curl, python3, and file write operations to project directories.
Sanitization: No explicit sanitization or filtering of the input text is performed before it is processed by the agent to create the image generation prompt.

arcanea-book-cover