The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the installation and use of the @runcomfy/cli Node.js package from the official registry. This is standard behavior for a tool-based skill and targets the vendor's own CLI.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute runcomfy commands. Security risks are mitigated by the use of JSON-formatted inputs to pass data to the CLI, which prevents shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface where instructions could be embedded in third-party audio files processed during inpainting or outpainting tasks.
Ingestion points: The audio URL parameter within the input JSON for the audio-inpaint and audio-outpaint routes in SKILL.md.
Boundary markers: The skill documentation includes a "Security & Privacy" section that instructs the agent to only process audio URLs explicitly provided by the user and to monitor for unexpected output shifts.
Capability inventory: The agent is restricted to executing the runcomfy CLI via the Bash tool, limiting the impact of any potential injection.
Sanitization: Prompt data and URLs are passed as static fields within a JSON object to the CLI, avoiding direct shell interpretation of the untrusted content.
[CREDENTIALS_UNSAFE]: The skill documents the use of an API token stored at ~/.config/runcomfy/token.json. This is the standard configuration path for the CLI, and the skill includes instructions to avoid logging or exposing this token.

ai-music