The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes ffmpeg and ffprobe to handle video rendering, audio mixing, and media metadata analysis. Arguments are passed as arrays to prevent shell-based injection attacks.
[EXTERNAL_DOWNLOADS]: The skill connects to Google's official Gemini API to generate narration and character illustrations. It also facilitates the download of official OpenAI Whisper models for automated caption alignment when requested by the user.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content (articles, scripts, or briefs) provided by the user to generate video storyboards and rendering instructions.
Ingestion points: Untrusted data enters through the source material parameter in the setup phase.
Boundary markers: Use of the '5-Stage High-Completion Heartbeat Engine' and specific visual style contracts ensures user content is framed within established templates.
Capability inventory: The skill has capabilities for local file management, system command execution (ffmpeg), and outbound network calls to trusted Google services.
Sanitization: The skill provides instructions to the AI to treat input strictly as narrative content and forbids the generation of unintended visual text or modifications to the core persona.

directing-stickman-videos