The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-uploaded video files to generate narrative text that is subsequently used in prompts for image generation and captioning.
Ingestion points: User-provided video files (in.mov, small.mp4).
Boundary markers: The prompts do not use specific delimiters or instructions to ignore potentially malicious embedded content within the video analysis results.
Capability inventory: Shell execution of ffmpeg/ffprobe and network API access via the gemini_client helper.
Sanitization: No explicit filtering or validation of the grounded storyline is performed before it is interpolated into image generation prompts.
[COMMAND_EXECUTION]: Utilizes local system tools ffmpeg and ffprobe to analyze video files, extract frames, and compress clips. This is the core functionality for video processing.
[DYNAMIC_EXECUTION]: Loads helper scripts from a local path (/mnt/skills/user/invoking-gemini/scripts) to facilitate communication with AI models. This represents standard modular development practices within the platform's skill environment.

drawing-video