creativeclaw-elevenlabs-v3
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted script text that incorporates instructions for the speech generation model via square-bracketed audio tags (e.g., [excited], [pause]). This presents a standard surface for indirect prompt injection where data content can be misinterpreted as control instructions. * Ingestion points: The text parameter passed to the generate_speech tool in SKILL.md. * Boundary markers: None are specified; the skill encourages interleaving tags directly with the script content. * Capability inventory: The skill uses media generation, merging, and assembly tools provided by the creative-claw MCP. * Sanitization: The instructions do not define any validation or escaping mechanisms for the input text.
- [EXTERNAL_DOWNLOADS]: The agent is configured to connect to an external service at https://app.creativeclaw.co/mcp/chatgpt to access the necessary tools. This endpoint belongs to the identified author (creativeclawco) and is essential for the skill's media processing capabilities.
- [COMMAND_EXECUTION]: The skill outlines workflows for merging and assembling media files using specialized tools. Evidence: The processes described in references/media-assembly.md rely on structured tool calls to the MCP server rather than executing arbitrary shell commands or untrusted scripts.
Audit Metadata