talk
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation indicates that voice models are downloaded during the initial user connection to the Pipecat MCP server. This is standard initialization behavior for the vendor's service.
- [COMMAND_EXECUTION]: The instructions describe a workflow where the agent can run commands or edit files when requested by the user. This capability is moderated by a strict guideline requiring the agent to seek verbal confirmation from the user before proceeding with any changes.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted input (voice conversation) and tool outputs (window listings and screen captures) which are then used to inform actions. The risk is mitigated by a required human-in-the-loop confirmation step for all file modifications or command executions.
Audit Metadata