skills/jeremylongshore/tons-of-skills-marketplace/abridge-reference-architecture/Gen Agent Trust Hub
abridge-reference-architecture
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data in the form of clinical transcripts and ambient audio captures which can be used to influence agent behavior.
- Ingestion points: Data enters the system context via webhook events (encounter completion) and the Abridge Cloud API (transcript fragments), as referenced in
src/webhooks/handler.tsandsrc/abridge/client.ts. - Boundary markers: The architecture includes HMAC signature verification in
src/webhooks/signature.tsand PHI redaction insrc/security/phi-redactor.ts. However, there are no explicit boundary markers or instructions to ignore potential commands embedded within the clinical transcripts themselves. - Capability inventory: The skill possesses the capability to write to the local file system (
allowed-tools: Write, Edit) and perform network operations to external EHR APIs (Epic, Athena) viasrc/ehr/fhir-client.ts. - Sanitization: While the architecture includes FHIR resource schema validation and PHI redaction, it lacks specific sanitization to prevent natural language instructions within patient encounters from being interpreted as commands by the processing agent.
Audit Metadata