The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by accepting and processing user-provided natural language guidance and image content which are then transmitted to the Bria AI service. \n- Ingestion points: Natural language guidance provided via the --prompt argument and the input source image (either a local file or a URL) in the bria_resize and bria_resize_submit functions. \n- Boundary markers: There are no specialized delimiters or system instructions used to separate user-provided data from the API request context or to instruct the model to ignore instructions embedded within the data. \n- Capability inventory: The skill utilizes curl for network requests to Bria AI's endpoints and performs file writes for both credential storage (~/.bria/credentials) and output image management. \n- Sanitization: While the shell client script performs character-level escaping for JSON compatibility, it does not perform semantic sanitization to prevent malicious instructions within the prompt or image content from potentially influencing the behavior of the Bria AI service.

ad-resize