The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (prompts and images) which are subsequently passed to external AI models and providers. \n
Ingestion points: scripts/gen.mjs accepts user-supplied prompts via --prompt or --prompt-file, and images via --images.\n
Boundary markers: The skill instructions include prompt templates but lack explicit delimiters or instructions to the underlying models to ignore malicious commands embedded within the input data.\n
Capability inventory: scripts/lib/providers.mjs performs network operations (API calls to OpenAI, Google, Fal, etc.) and scripts/gen.mjs performs file system operations (writing generated images to the disk) via child_process.spawn and fetch.\n
Sanitization: No specific input sanitization or filtering is implemented for the prompt text before it is interpolated into the external API requests.\n- [EXTERNAL_DOWNLOADS]: The skill documentation describes downloading configuration templates from the vendor's repository (github.com/dlazy-ai/ecommerce-skills) and installing the vendor's CLI tool (@dlazy/cli) via npm. These resources are hosted on platforms associated with the skill's authoring organization.\n- [COMMAND_EXECUTION]: The execution logic in scripts/lib/providers.mjs uses child_process.spawn to run the vendor-provided dlazy command-line utility. This is a primary mechanism for the skill to interact with the image generation backend.\n- [DATA_EXFILTRATION]: To perform its primary function, the skill reads local image files and transmits them along with prompts and API keys (retrieved from environment variables) to external services such as OpenAI and Google Gemini.

creative-scene