The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text strings via the --style-mod parameter, which are interpolated directly into the prompts used for image generation. This represents a potential vulnerability surface where malicious instructions could be embedded in the style data.
Ingestion points: The --style-mod command-line argument and input photo reference path/URL.
Boundary markers: The prompt construction uses natural language labels but lacks formal delimiters or "ignore previous instructions" safety markers around the interpolated user input.
Capability inventory: The skill calls external image generation APIs (via internal runners) and performs file system operations to save generated images and manifests.
Sanitization: No explicit sanitization or validation of the --style-mod input is documented in the skill instructions.
[DYNAMIC_EXECUTION]: The entry point script scripts/run.py utilizes os.execv to perform process replacement. The script dynamically locates a local virtual environment (.runners-venv) and re-executes itself using the specific Python interpreter from that environment. While a standard utility pattern, it involves dynamic execution paths.
[COMMAND_EXECUTION]: The skill configuration requires the Bash tool to initiate the Python-based execution pipeline through the scripts/run.py script.

avatar-maker