The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes untrusted user input ($ARGUMENTS) to define directory and file names. While it includes validation for genres, the album names are passed directly into a tool with file-writing capabilities.
Ingestion points: User-supplied album names and genres in the $ARGUMENTS variable.
Boundary markers: None present around the interpolated arguments.
Capability inventory: Access to Bash and Write tools allows for significant file system modifications.
Sanitization: The skill provides a whitelist for genres and specific parsing logic, though it lacks validation for the album-name string itself.
Command Execution (SAFE): Although the skill specifies the Bash tool as allowed, the instructions explicitly prohibit the agent from manually running shell commands (mkdir, cp) and instead mandate the use of a structured MCP tool (create_album_structure). This significantly reduces the risk of arbitrary command execution via user-supplied names.
Data Exposure (SAFE): The skill references local configuration paths (e.g., ~/.bitwize-music/config.yaml). This is standard behavior for local automation skills and no patterns of exfiltration were detected.

new-album