The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/media_optimizer.py utilizes the eval() function to process the r_frame_rate metadata field extracted from media files via ffprobe. Because metadata in media files can be arbitrarily modified by an attacker, passing this data directly to eval() creates a risk of arbitrary code execution.
[COMMAND_EXECUTION]: The scripts/media_optimizer.py script uses subprocess.run() to execute ffmpeg and ffprobe commands. While necessary for the skill's media optimization functionality, this allows for the execution of external system binaries based on user-supplied file paths.
[INDIRECT_PROMPT_INJECTION]: This skill is designed to ingest and process external, untrusted media files including PDFs, images, audio, and video. The content extracted from these files (such as OCR text or transcriptions) is fed into the AI agent's context, which could allow for indirect prompt injection attacks if the source files contain malicious instructions.
Ingestion points: Media files are processed in scripts/gemini_batch_process.py and scripts/document_converter.py.
Boundary markers: None identified in the script code for delimiting extracted content within the prompt.
Capability inventory: The skill has access to Bash, Read, Write, and Edit tools.
Sanitization: There is no evidence of sanitization or filtering of the content extracted from media files before it is processed by the agent.

ai-multimodal