The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input via the $ARGUMENTS variable to define album names and genres. While instructions suggest converting inputs to slugs, there are no explicit boundary markers or strict validation logic within the prompt to prevent a user from providing malicious strings that could attempt to influence the agent's logic or the underlying tool's behavior.
Ingestion points: User input provided through $ARGUMENTS in SKILL.md.
Boundary markers: None present; the arguments are parsed directly into variables.
Capability inventory: The skill has access to Bash, Write, and bitwize-music-mcp (a vendor-specific tool) in its allowed-tools list, enabling file system and command-line operations.
Sanitization: The instructions mention using "slug form" (lowercase, hyphenated), but the enforcement relies on the model's interpretation rather than programmatic validation.
[COMMAND_EXECUTION]: The skill explicitly allows the use of the Bash tool and provides examples of shell commands (e.g., mkdir, cp) in the "Common Mistakes" section. The combination of shell access and processing user-controlled input (album names and genres) creates a surface for potential command injection if the inputs are not handled carefully by the agent or the underlying system.

new-album