The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted lyrics data and possesses file-writing capabilities, creating an attack surface for indirect prompt injection.\n
Ingestion points: Lyrics are ingested from track files or direct user input via the $ARGUMENTS parameter in SKILL.md.\n
Boundary markers: The instructions do not define any delimiters or boundary markers to encapsulate and separate the untrusted lyrics data from the agent's core instructions.\n
Capability inventory: The skill is permitted to use Read, Edit, and Write tools to modify track files, and utilizes the bitwize-music-mcp tool for scanning.\n
Sanitization: There is no evidence of input sanitization or filtering logic to prevent malicious instructions embedded in the lyrics from being executed or processed by the agent.

pronunciation-specialist