The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes untrusted user-provided topics and text to generate images, creating a surface for indirect prompt injection.
Ingestion points: auto_generator.py (reads from sys.stdin) and interactive_generator.py (reads from input()).
Boundary markers: Absent; user content is interpolated directly into the image generation logic without delimiters or safety instructions.
Capability inventory: File system write operations (generating PNG images) in the /mnt/user-data/outputs directory.
Sanitization: No evidence of input escaping or validation for the text rendered into images.
Unverifiable Dependencies & Remote Code Execution (SAFE): The skill relies on the 'Pillow' library for image processing. Installation instructions use the standard pip install method. No remote scripts are downloaded or executed.
Data Exposure & Exfiltration (SAFE): No hardcoded credentials, sensitive file access (e.g., SSH keys), or network-based exfiltration patterns were detected in the analyzed scripts.
Privilege Escalation (SAFE): While the documentation mentions sudo apt-get for font installation, the scripts themselves do not attempt to escalate privileges or perform unauthorized system modifications.
Command Execution (SAFE): No use of eval(), exec(), or dangerous subprocess calls with untrusted input was identified. File system operations are limited to creating the output directory and saving images.

card-news-generator