The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes uv run to invoke local video generation and media management utilities. These commands are consistent with the skill's described functionality for video editing.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading necessary models via comfy-models. It explicitly directs the agent to describe the download to the user and wait for approval before using the --yes flag, adhering to safe operational practices.
[DATA_EXFILTRATION]: Network activity is restricted to a local gallery service running on 127.0.0.1. This does not pose a risk of data exfiltration.
[PROMPT_INJECTION]: The skill defines a surface for potential indirect prompt injection by interpolating user-provided prompts and file paths into shell commands.
Ingestion points: User-provided text for --prompt and file paths for --input-video and --reference-image (SKILL.md).
Boundary markers: Arguments are wrapped in double quotes in example commands, but instructions do not mandate input sanitization.
Capability inventory: Shell command execution via uv run comfy-videogen (SKILL.md).
Sanitization: Absent from instructions; relies on agent-level handling of command line arguments.

comfy-bernini-videoedit