retellai-reference-architecture

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides architecture implementation examples that ingest and process external, untrusted data from voice call transcripts and webhooks.
  • Ingestion points: In references/architecture-examples.md, the Express.js webhook handler processes call.transcript and req.body.args from the Retell platform.
  • Boundary markers: The provided prompt examples and code snippets do not include explicit boundary markers or directives to ignore instructions embedded in the voice transcript.
  • Capability inventory: The reference code demonstrates capabilities for writing to databases (saveCallRecord) and executing business logic (bookAppointment) based on the ingested data.
  • Sanitization: The provided code examples lack explicit sanitization or validation logic for the external call data before it is used in downstream operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:48 AM
Security Audit — agent-trust-hub — retellai-reference-architecture