The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/media_optimizer.py uses the eval() function to parse frame rate information from ffprobe output (info['fps'] = eval(stream.get('r_frame_rate', '0/1'))). This is dangerous because metadata in media files can be manipulated; a malicious file could inject Python code into fields processed by eval(), potentially allowing for remote code execution.
[PROMPT_INJECTION]: The skill ingests and processes untrusted data from images, videos, and documents, creating a vulnerability to indirect prompt injection.
Ingestion points: scripts/gemini_batch_process.py and scripts/document_converter.py load file content and metadata for model processing.
Boundary markers: Absent; files are processed without explicit delimiters or instructions to the model to ignore potential embedded instructions.
Capability inventory: The skill uses subprocess.run to call ffmpeg and performs various file system write operations.
Sanitization: Absent; no filtering or validation is applied to content extracted from processed files before it is sent to the model.

ai-multimodal