The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it processes untrusted audio files provided via URL for inpainting and outpainting tasks.
Ingestion points: The audio parameter in the input JSON for the acestep-ai/ace-step/audio-inpaint and acestep-ai/ace-step/audio-outpaint endpoints (SKILL.md).
Boundary markers: The skill uses structured JSON for inputs, though it lacks explicit instructions to the underlying model to ignore content embedded within the audio files.
Capability inventory: The skill relies on the runcomfy CLI (command execution) to interact with remote model APIs.
Sanitization: The documentation explicitly states that the CLI does not perform shell expansion on input content, which mitigates direct shell injection risks.
[COMMAND_EXECUTION]: The skill's core functionality is delivered through the execution of the runcomfy CLI tool.
Evidence: The skill provides multiple examples of invoking runcomfy run with various model paths and JSON inputs to generate audio.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the necessary CLI tool from a public package registry.
Evidence: The installation steps recommend npm i -g @runcomfy/cli or using npx -y @runcomfy/cli --version. These downloads target the official NPM registry, which is a well-known service.

ace-step