The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches and executes installation scripts for the gemini-api CLI from the official google-gemini GitHub repository.
[COMMAND_EXECUTION]: Relies on shell command execution for its core functionality, including the gemini-api CLI, ffmpeg for media processing, and the gcloud CLI for authentication and token management.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled data that is subsequently interpolated into shell commands, creating a potential attack surface.
Ingestion points: User-provided text prompts, local media file paths, and interaction IDs are passed to shell commands in SKILL.md and references/gemini-omni-1-1-flash.md (e.g., via -i and --previous-interaction-id flags).
Boundary markers: Absent. The instructions do not specify the use of delimiters or escaping mechanisms when incorporating user-controlled variables into shell strings.
Capability inventory: Shell command execution via gemini-api, ffmpeg, and gcloud, alongside file system write access for outputting generated videos.
Sanitization: Absent. The skill provides no guidance or examples for validating or sanitizing user-provided strings before they are executed in a shell environment, which could lead to command injection if handled by an unconstrained agent.

gemini-omni