The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input in the form of music captions and lyrics which are interpolated into API requests. This creates an attack surface where malicious instructions could be embedded in data processed by the agent.
Ingestion points: The prompt and lyrics fields used in REST API request examples in SKILL.md.
Boundary markers: No explicit sanitization, escaping, or boundary delimiters are provided for these user-supplied fields in the documentation.
Capability inventory: The skill utilizes network capabilities to communicate with the ACE-Step API.
Sanitization: The instructions lack specific guidance on validating or sanitizing user-provided text before inclusion in API payloads.
[EXTERNAL_DOWNLOADS]: The skill references external resources for model weights, documentation, and source code.
Evidence: Links to the official GitHub repository (github.com/ace-step/ACE-Step-1.5) and Hugging Face model cards (huggingface.co/ACE-Step/Ace-Step1.5).
Context: These downloads and references target well-known services and the official vendor of the ACE-Step tool, representing standard functionality for this type of integration.

ace-step