The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses multiple shell script blocks to automate the cover generation process. These scripts execute commands such as curl for network requests, jq for JSON manipulation, base64 for decoding image data, and magick or sips for image resizing and cropping.
[CREDENTIALS_UNSAFE]: The skill requires a sensitive GPT_IMAGE_API_KEY to be set as an environment variable. This key is passed in the Authorization header of curl requests.
[DATA_EXFILTRATION]: The API key is sent to the address defined by GPT_IMAGE_BASE_URL. While it defaults to OpenAI's official endpoint (a well-known service), this variable can be overridden. If redirected to an untrusted domain, the user's API credentials would be exposed to that third party.
[EXTERNAL_DOWNLOADS]: The skill can download reference images from arbitrary URLs provided by the user using curl -fsSL.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data to influence generated prompts and file system operations. 1. Ingestion points: User-supplied book titles and author names are ingested via the SKILL.md instructions. 2. Boundary markers: The skill does not employ delimiters or specific instructions to the agent to ignore malicious patterns within the user-provided text. 3. Capability inventory: The skill utilizes shell execution for network operations (curl), data processing (jq, base64), and image manipulation (magick, sips), along with file writing and directory creation. 4. Sanitization: Input is passed to shell environment variables without explicit sanitization for shell metacharacters, though jq is used for the API request body.

story-cover