The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to construct prompts for AI image generation and to populate parameters in Python scripts. This creates a potential vulnerability where malicious instructions embedded in a user request could influence the agent's behavior. \n
Ingestion points: User's natural language requests are used to define the 'prompt_text' and instructions in the generated scripts (found in SKILL.md and examples/). \n
Boundary markers: The skill lacks explicit delimiters or instructions for the agent to ignore potentially malicious content within user-provided descriptions. \n
Capability inventory: The skill utilizes Bash to execute code, Write to save files, and WebFetch to interact with APIs. \n
Sanitization: No explicit sanitization or validation of the user-provided prompt content is mentioned in the instructions. \n- [DYNAMIC_EXECUTION]: The skill's operational model involves the agent generating Python scripts on-the-fly and executing them via the Bash tool. While these scripts follow provided templates for interacting with legitimate APIs, the pattern of dynamic script generation and execution is a known attack surface. \n- [COMMAND_EXECUTION]: The instructions direct the agent to use the Bash tool for several tasks, including package installation (pip install) and running the custom Python scripts used to interface with the image generation services.

image-generation