The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and provides instructions for installing the @runcomfy/cli package from the official NPM registry. This package is a legitimate vendor resource associated with the runcomfy.com service.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted asset URLs and text prompts for generative model pipelines, which is a common attack surface for indirect injection.
Ingestion points: image_url, audio_url, video_url, and prompt fields in JSON inputs for runcomfy run commands (SKILL.md).
Boundary markers: Commands use JSON structure for inputs, but the prompts themselves lack explicit data/instruction delimiters.
Capability inventory: Execution of the runcomfy CLI via Bash tool and file system write access via the --output-dir flag.
Sanitization: The skill provides explicit safety instructions to the agent to refuse harmful user requests (e.g., defamatory or non-consensual content) and notes that the underlying CLI does not perform shell expansion on input content.

face-swap