The Agent Skills Directory

[SAFE]: The skill utilizes the @runcomfy/cli tool, which is a vendor-provided package hosted on the official NPM registry. This is standard practice for providing specialized CLI capabilities.\n- [SAFE]: Sensitive information such as authentication tokens are managed using secure defaults, including restricted file permissions (mode 0600) and support for environment variables in CI environments.\n- [COMMAND_EXECUTION]: Shell execution is strictly limited to the runcomfy CLI as defined in the allowed-tools configuration. The skill provides clear instructions on using JSON to wrap inputs, which effectively mitigates common shell injection risks.\n- [INDIRECT_PROMPT_INJECTION]: The skill identifies risk of indirect prompt injection from untrusted image URLs and provides guidance to the agent on handling user-provided content.\n
Ingestion points: External data enters through the image_urls and prompt keys within the --input JSON parameter in SKILL.md.\n
Boundary markers: The skill uses JSON structured strings to encapsulate inputs, preventing shell interpretation of the payload.\n
Capability inventory: The skill is scoped to the runcomfy CLI via the Bash tool.\n
Sanitization: Documentation states the CLI avoids shell expansion for prompt content.

image-outpainting