The Agent Skills Directory

[COMMAND_EXECUTION]: The vedit.py script executes ffmpeg and ffprobe shell commands to perform audio extraction and media metadata probing. These calls use the recommended list-based argument format which helps mitigate command injection risks.
[EXTERNAL_DOWNLOADS]: The skill's setup documentation instructs the user to download approximately 7 GB of external resources, including Python libraries from the official PyTorch registry and a large-scale transcription model from Hugging Face. These sources are well-known and reputable.
[INDIRECT_PROMPT_INJECTION]: The skill processes listing.txt, which contains verbatim transcripts of user-provided video files. This creates a surface for indirect prompt injection where malicious instructions spoken in a video could potentially influence the agent's behavior during the editorial pass.
Ingestion points: Transcribed text from external media files is loaded into the agent's context from listing.txt and segments.json (created in scripts/vedit.py).
Boundary markers: The instructions lack specific delimiters or safety warnings to prevent the agent from following instructions found within the transcribed text.
Capability inventory: The skill has the capability to execute shell commands (ffmpeg) and perform DaVinci Resolve operations (via MCP tool calls).
Sanitization: There is no evidence of filtering or sanitization of the transcription output before it is read by the agent.

davinci-resolve-video-edit