The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by allowing the ingestion of untrusted data from external sources into agent contexts.
Ingestion points: The add_knowledge_base_to_agent tool in SKILL.md accepts external URLs and local file paths to populate an agent's knowledge base. Additionally, the speech_to_text tool processes local audio files which may contain instructions embedded in the audio.
Boundary markers: The skill does not define boundary markers or explicit instructions for the AI to ignore commands embedded in the ingested data.
Capability inventory: The skill has the capability to create AI agents with custom system prompts (create_agent), make outbound phone calls (make_outbound_call), and perform local file write operations (text_to_speech, compose_music).
Sanitization: No sanitization or validation of the ingested external content is mentioned in the skill definition.

elevenlabs-mcp