The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates music operations by executing the listenhub CLI binary. It constructs command lines for tasks like song generation, track isolation, and audio splitting, using user-provided strings for prompts, lyrics, and file references. A hard-gate confirmation mechanism is enforced to prevent unauthorized execution.
[EXTERNAL_DOWNLOADS]: The skill utilizes curl to fetch generated audio files and stem ZIP archives from remote URLs generated by the Mureka service. These downloads are saved locally to the user's working directory, following a predefined artifact naming convention.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input—including song prompts, lyrics, and media files—which creates a surface for indirect prompt injection targeting both the AI model and the local shell environment.
Ingestion points: User-supplied text (prompts, lyrics) and media files (audio, images, video) processed by the interaction flow in SKILL.md.
Boundary markers: A mandatory <HARD-GATE> is implemented, requiring the agent to present all parameters to the user for explicit confirmation before any action is taken.
Capability inventory: The skill can execute subprocesses via the listenhub CLI, perform network downloads via curl, and manage local configuration files and music artifacts.
Sanitization: The skill uses jq with variable binding for configuration management to prevent injection in JSON operations. For CLI command execution, it relies on the agent platform to appropriately escape user-provided variables within shell templates.

music