provider-denial-workup
Pass
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions define a purely reasoning-based workflow for medical claim appeals. It does not utilize any dangerous shell commands, network operations, or external code execution.
- [PROMPT_INJECTION]: The skill processes untrusted input from medical denial documents (EOB/835 ERA), creating an indirect prompt injection surface. The analysis identified the following security posture: (1) Ingestion points: Denial text provided by the user to the agent; (2) Boundary markers: Relies on structured reasoning tables rather than explicit delimiters; (3) Capability inventory: Limited to text generation (appeal letters) with no execution or filesystem writes; (4) Sanitization: Explicitly mandates human-in-the-loop approval before any output is used. This surface is considered safe due to the reasoning-only nature and lack of dangerous capabilities.
- [DATA_EXFILTRATION]: No network access or data exfiltration vectors were identified. The skill operates solely on user-provided context within the session boundary and includes guardrails for handling Protected Health Information (PHI).
Audit Metadata