speech-engine

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes speech-to-text transcripts derived from user audio, which represents an untrusted data source. 1. Ingestion points: WebSocket callbacks like onTranscript (SKILL.md, references/javascript-sdk-reference.md) and on_transcript (SKILL.md, references/python-sdk-reference.md). 2. Boundary markers: Documentation in SKILL.md mandates a 'validation boundary' and 'deterministic validation'. 3. Capability inventory: Facilitates streaming text responses and potentially tool calls (SKILL.md). 4. Sanitization: SKILL.md warns transcript text is 'untrusted' and requires conversion to 'trusted application state'.
  • [SAFE]: The skill follows security best practices for credential management, advising users to store the ELEVENLABS_API_KEY in environment variables and never expose it in browser-side code.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 08:15 PM
Security Audit — agent-trust-hub — speech-engine