speech-engine
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface for indirect prompt injection via speech-to-text transcripts, which is the intended functionality of the voice engine.
- Ingestion points: The skill utilizes on_transcript (Python) and onTranscript (TypeScript) callbacks to handle voice recognition data.
- Boundary markers: The documentation provides explicit warnings to treat transcripts as untrusted user input and avoid mapping them directly into model roles or tool calls.
- Capability inventory: The server logic implemented by the developer can invoke tools or generate responses based on these transcripts.
- Sanitization: The skill recommends implementing deterministic validation, allowlisted intents, or explicit user confirmation before acting on transcript data.
Audit Metadata