The Agent Skills Directory

[SAFE]: Detailed analysis confirms the skill functions as described, providing instructions for AI-assisted music generation without any hidden or malicious behaviors.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to interact with the belt CLI; however, it limits the tool's scope to the belt command specifically via the allowed-tools metadata, adhering to the principle of least privilege.
[EXTERNAL_DOWNLOADS]: The skill references installation guidelines and dependencies from the inference-sh organization on GitHub. These sources are official repositories for the platform the skill integrates with and are used for legitimate configuration purposes.
[PROMPT_INJECTION]: The skill handles untrusted user input via musical prompts for model generation. While this presents an indirect prompt injection surface, the risk is mitigated.
Ingestion points: User-provided strings enter the agent context through the --input flag in belt app run examples found in SKILL.md.
Boundary markers: The instructions utilize structured JSON formatting to encapsulate user prompts, which serves as a natural boundary between command instructions and data.
Capability inventory: The skill uses the Bash tool for model inference via the belt CLI.
Sanitization: Content is processed by the external inference.sh platform, which typically employs its own safety and content filtering protocols for model generation.

ai-music-generation