The Agent Skills Directory

[Data Exposure & Exfiltration] (SAFE): The skill requires a Google Gemini API key and correctly instructs the user to provide it via environment variables or a .env file. The scripts/check_setup.py script attempts to find this key using a centralized resolver pattern, which is a secure way to manage secrets across multiple skills without hardcoding them.
[Indirect Prompt Injection] (LOW): Because the skill is designed to process external, untrusted media files (PDFs, images, audio, video), it inherently possesses an indirect prompt injection surface. An attacker could embed malicious instructions within a document or image metadata.
Ingestion points: Processing functions in gemini_batch_process.py and document_converter.py (referenced in SKILL.md).
Boundary markers: Not explicitly defined in the provided documentation, though typical for Gemini-based workflows.
Capability inventory: The skill uses Bash, Read, Write, and Edit permissions, allowing it to modify the filesystem and execute commands, which is required for media conversion and processing.
Sanitization: No explicit sanitization of file content is shown in the provided script logic, relying on the underlying LLM's safety guardrails.
[Dynamic Execution] (SAFE): The scripts/check_setup.py script uses __import__ to dynamically check for the presence of required Python packages. This is a standard and safe implementation for a setup verification tool.

ai-multimodal