The Agent Skills Directory

[COMMAND_EXECUTION]: Utilizes the runcomfy command-line interface for model execution. Tool access is appropriately scoped via the platform's frontmatter configuration.
[EXTERNAL_DOWNLOADS]: References the official @runcomfy/cli package for installation via the NPM registry, following standard software distribution practices.
[PROMPT_INJECTION]: Identifies the potential for indirect prompt injection via user-provided media URLs and provides explicit guidance for agents to mitigate these risks through careful URL handling and content validation.
Ingestion points: External media URLs (video_url, audio_url, image_url) referenced in SKILL.md.
Boundary markers: Uses structured JSON input strings for CLI arguments to prevent shell expansion issues.
Capability inventory: Access is restricted to the runcomfy CLI tool.
Sanitization: Documentation specifies that agents should only ingest URLs explicitly provided by the user and notes that the skill itself does not gate inputs, placing responsibility on the operator.

lipsync