The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install standard, well-known Python packages including google-genai, python-dotenv, and pillow to interact with the Google Gemini API.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data from various media formats (PDF, audio, video, images). These files can contain embedded instructions intended to influence the agent's behavior.
Ingestion points: Files are processed via scripts/gemini_batch_process.py and the Google GenAI SDK.
Boundary markers: The provided documentation and scripts do not demonstrate specific delimiting or 'ignore instruction' warnings for the multimodal content.
Capability inventory: The skill is configured with allowed-tools including Bash, Write, and Edit, providing a significant capability surface if an injection succeeds.
Sanitization: There is no evidence of sanitization or filtering for instructions embedded within the processed media.
[CREDENTIALS_UNSAFE]: The skill requires a GEMINI_API_KEY. It correctly guides the user to provide this through environment variables or .env files rather than hardcoding credentials, which is a safe practice for secret management.

ai-multimodal