The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the host system by executing the runcomfy command-line interface to perform image generation tasks as described in the invocation examples.
[EXTERNAL_DOWNLOADS]: The skill documentation specifies the requirement to install the @runcomfy/cli package via NPM and describes the process of downloading generated image assets from the service provider's official domains (.runcomfy.net and .runcomfy.com).
[CREDENTIALS_UNSAFE]: The skill provides clear guidance on managing API authentication tokens, recommending the use of the RUNCOMFY_TOKEN environment variable or a configuration file at ~/.config/runcomfy/token.json with restricted permissions (mode 0600), which aligns with standard security practices for CLI tools.
[INDIRECT_PROMPT_INJECTION]: The skill identifying a surface for indirect prompt injection as it processes user-provided text for image generation.
Ingestion points: User-supplied text is mapped to the prompt field within the skill's input schema in SKILL.md.
Boundary markers: The skill documentation notes that input is passed as a JSON string to the CLI, providing a structural boundary that separates instruction from data.
Capability inventory: The skill uses the runcomfy CLI to perform network operations and writes generated files to a user-specified output directory.
Sanitization: The skill explicitly states that the CLI JSON-encodes the prompt to prevent shell injection and mentions upstream safety tolerance settings provided by the model provider.

nano-banana-2