The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to interact with the runcomfy CLI binary. The scope of execution is restricted to the runcomfy command prefix as defined in the allowed-tools configuration.
[EXTERNAL_DOWNLOADS]: The runcomfy run command downloads generated assets from *.runcomfy.net and *.runcomfy.com. These domains are official infrastructure for the RunComfy service and are documented for transparency.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface where third-party content (images, videos, or web search results) provided via URLs could contain malicious instructions designed to influence model output.
Ingestion points: Input parameters for the runcomfy run command, including asset URLs and web search result fields (SKILL.md).
Boundary markers: Not explicitly defined within the prompt structure, although the CLI enforces JSON schema validation for inputs.
Capability inventory: The agent can execute the runcomfy CLI which has capabilities for network requests to the RunComfy API and file system writes for downloads (SKILL.md).
Sanitization: The skill provides explicit mitigation instructions for the agent, such as preferring user-provided URLs and defaulting enable_web_search to false.
[DATA_EXPOSURE]: The documentation describes the location of the API token at ~/.config/runcomfy/token.json and its file permissions (0600). It includes clear instructions to never log or expose this token in prompts.

runcomfy-cli