The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external media (video and audio) for transcription and analysis. This creates a surface where malicious instructions could be embedded in the content or metadata to influence the agent.
Ingestion points: External video and audio source files (SKILL.md).
Boundary markers: None specified in the protocol.
Capability inventory: File system access, tool execution (ffprobe, Blender), and code generation for video frameworks.
Sanitization: No sanitization or validation steps are defined for external data inputs before they are used in the workflow.
[DYNAMIC_EXECUTION]: The protocol directs the agent to generate and run scripts for Remotion, HyperFrames, and Three.js/Blender to build video scenes. Generating executable code for rendering purposes is a known dynamic execution surface.
[COMMAND_EXECUTION]: The skill utilizes command-line utilities such as ffprobe for file inspection and Blender for headless rendering tasks as part of its deterministic pipeline.

master-editor-protocol