The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool restricted to the runcomfy command to interact with the RunComfy API. It provides structured examples that pass inputs via JSON strings to the --input flag, which effectively prevents shell injection even when processing user-supplied prompts.
[EXTERNAL_DOWNLOADS]: The skill instructs the operator to install the @runcomfy/cli package from the official NPM registry. It also downloads generated image files from the platform's official domains (*.runcomfy.net and *.runcomfy.com) into a user-specified output directory.
[PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection because it can process external data from untrusted sources (reference images and web search results).
Ingestion points: External content enters the agent's context through image URLs (e.g., image_urls, images) and the enable_web_search feature, which can influence the image generation model.
Boundary markers: Data is encapsulated within a JSON object passed to the CLI's --input parameter.
Capability inventory: The skill can execute the runcomfy command, make network requests to the RunComfy API, and write files to the local file system.
Sanitization: The skill documentation includes a dedicated 'Security & Privacy' section that warns about these risks and recommends mitigations such as only using user-provided URLs and keeping web search disabled by default.

ai-image-generation