The Agent Skills Directory

[SAFE]: The skill implements the principle of least privilege by restricting shell access to the runcomfy command specifically via the allowed-tools frontmatter configuration.
[EXTERNAL_DOWNLOADS]: Installs and utilizes the official @runcomfy/cli package from the NPM registry, which is the expected tool for the vendor's infrastructure.
[CREDENTIALS_UNSAFE]: Manages API tokens using standard and secure practices, such as environment variables (RUNCOMFY_TOKEN) and local configuration files with restricted access permissions (0600).
[DATA_EXFILTRATION]: Network activity is restricted to the platform's official domains (runcomfy.com, runcomfy.net) for media processing and asset delivery.
[PROMPT_INJECTION]: Contains a dedicated safety section instructing the agent to refuse requests involving non-consensual deepfakes or harmful synthetic media.
[COMMAND_EXECUTION]: Shell commands are structured to pass inputs as JSON via the --input flag, which prevents shell-injection vulnerabilities when processing user-supplied prompts or URLs.
[SAFE]: The skill identifies the potential for indirect prompt injection from third-party assets (images/audio) and provides explicit mitigation strategies for the agent.

ai-avatar-video