The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill correctly identifies and utilizes the official Google Generative AI SDK (@google/genai). All external references and documentation links point to official ai.google.dev or github.com/googleapis domains, which are recognized as trusted services.
[COMMAND_EXECUTION]: Includes a diagnostic shell script (check-versions.sh) designed to verify the local environment. It safely uses npm list and node -v to check dependency versions and warns users if they are using the deprecated SDK.
[INDIRECT_PROMPT_INJECTION]: As the skill facilitates processing external data (PDFs, images, and video) via LLM models, it inherently involves an attack surface for indirect injection.
Ingestion points: Multimodal templates (multimodal-image.ts, multimodal-video-audio.ts) read file data from the local system for transmission to the Gemini API.
Boundary markers: The documentation includes a "Production Best Practices" section that explicitly recommends input validation and sanitization.
Capability inventory: The templates demonstrate filesystem read capabilities (fs.readFileSync) and network transmission to trusted Google API endpoints.
Sanitization: The skill provides architectural guidance on sanitizing user inputs and managing API keys securely using environment variables.
[SAFE]: The skill serves an educational and developmental purpose, adhering to secure coding standards such as avoiding hardcoded secrets and using official SDKs.

google-gemini-api