The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted inputs, including image URLs and user prompts, which are passed to the runcomfy CLI tool. This creates a surface for indirect prompt injection where malicious content in third-party images or prompts could attempt to influence the agent's behavior or CLI parameters.
Ingestion points: image_urls and prompt fields within the JSON input provided to the runcomfy command.
Boundary markers: The documentation states that prompts are encapsulated as JSON strings to prevent shell expansion.
Capability inventory: The agent uses the runcomfy CLI via bash to execute image processing tasks.
Sanitization: No explicit sanitization or filtering of external URLs or prompts is defined within the skill itself.
[UNVERIFIABLE_DEPENDENCIES]: The skill references the installation of the @runcomfy/cli package via npm. Although this is an official resource from the tool's vendor, it constitutes an external code dependency downloaded from a public registry.
[DATA_EXPOSURE]: The skill documents the local storage of API credentials at ~/.config/runcomfy/token.json. While this is functional documentation, it identifies a sensitive file path containing user authentication tokens.
[COMMAND_EXECUTION]: The skill uses the runcomfy CLI through bash to perform its primary functions, requiring the execution of local shell commands with user-influenced arguments.

image-outpainting