pipecat-talk
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts as an interface for voice commands, which serves as a data ingestion point for untrusted input.
- Ingestion points: User input is received via the
listen()command as described in the flow section ofSKILL.md. - Boundary markers: The skill explicitly requires the agent to display proposed changes in the terminal and obtain verbal confirmation before proceeding with modifications to files, PRs, or issues.
- Capability inventory: The skill enables the agent to edit files, run commands, list windows, and perform screen captures.
- Sanitization: The human-in-the-loop confirmation process serves as the primary safeguard to prevent the execution of malicious instructions embedded in user voice input.
- [COMMAND_EXECUTION]: The instructions permit the agent to run commands to perform requested tasks.
- Evidence: This capability is mentioned in the flow section and is subject to the confirmation guideline that mandates checking with the user before finalizing any changes.
Audit Metadata