The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted video transcripts; however, it includes strong mitigation instructions for the agent to ignore any commands within the data.
Ingestion points: Raw video files and transcripts downloaded via mcp__google-drive__download_file and processed by whisper.cpp (referenced in SKILL.md).
Boundary markers: Present. A prominent warning blockquote at the top of the file instructs the agent to treat transcripts as untrusted material.
Capability inventory: Execution of local shell commands (ffmpeg, remotion), Python script execution, and extensive Google Drive and Airtable API access.
Sanitization: The instructions direct the agent to quote and stop if it encounters text that appears to address it directly.
[COMMAND_EXECUTION]: The skill requires the use of local command-line tools such as ffmpeg for video trimming and remotion for rendering, as well as the execution of Python snippets for data cleaning.
[EXTERNAL_DOWNLOADS]: The skill downloads content from Google Drive and references external software like whisper.cpp and Remotion for core functionality.

edit-video