The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted media files, which could contain instructions intended to override agent behavior or extract data.
Ingestion points: The skill uses inlineData and ai.files.upload to process external images, videos, and audio in Sections 6.2, 6.3, 8.2, and 10.2 of SKILL.md.
Boundary markers: Absent. The provided prompt templates do not include delimiters or instructions to the model to ignore potential malicious content embedded in the media.
Capability inventory: The skill utilizes file system writes (fs.writeFileSync, ai.files.download) and makes network requests to the Gemini API.
Sanitization: The skill does not demonstrate any validation or sanitization of content extracted from the media files before processing.
[EXTERNAL_DOWNLOADS]: The skill installs the official Google Gen AI SDK and downloads generated media assets from Google's infrastructure.
The skill uses npm install @google/genai to manage API interactions.
The skill uses ai.files.download to fetch generated video files from Google's servers to the local environment.
[COMMAND_EXECUTION]: The skill includes instructions and templates that execute shell commands to process data.
Section 5.1 of SKILL.md demonstrates using a pipeline of curl, jq, and base64 --decode to save image data to the local disk.

google-gemini-media