The Agent Skills Directory

[SAFE]: The skill is designed to provide an interface for the RunComfy platform. It uses a scoped tool definition Bash(runcomfy *) to limit execution to the specific CLI tool.
[COMMAND_EXECUTION]: The skill uses the runcomfy CLI for model invocation and status polling. This is the primary intended purpose of the skill and is constrained by the frontmatter configuration.
[EXTERNAL_DOWNLOADS]: The skill downloads model output files from *.runcomfy.net and *.runcomfy.com. These are the official domains for the RunComfy service described in the skill.
[CREDENTIALS_UNSAFE]: The skill manages authentication tokens stored in ~/.config/runcomfy/token.json. It specifies that the file is created with 0600 permissions (owner-only access), which is a standard security practice for CLI credentials.
[REMOTE_CODE_EXECUTION]: While the skill mentions a curl | bash installation option for the CLI, it explicitly instructs users and agents to manually inspect such scripts before execution and recommends using verified package managers like npm or npx instead.
[PROMPT_INJECTION]: The skill includes a dedicated 'Security & Privacy' section that addresses 'Indirect prompt injection' (Category 8). It warns the agent that third-party content processed by models (like URLs or web search results) is untrusted and provides specific mitigation strategies, such as only processing user-provided URLs and defaulting to conservative search settings.

runcomfy-cli