The Agent Skills Directory

Indirect Prompt Injection Surface: The skill ingests untrusted external data in the form of clinical notes, imaging reports, and laboratory results as documented in references/01-intake-assessment.md. This information is processed by the agent to assess medical necessity against insurance policies. If these source documents were to contain adversarial text or hidden instructions, they could attempt to influence the agent's clinical reasoning. The risk is minimized by the skill's design, which explicitly requires a human professional to review, confirm, or override every draft decision before it is finalized.
Healthcare MCP Tool Integration: The skill utilizes specialized Model Context Protocol (MCP) tools to interact with healthcare databases, including npi_lookup_provider for credentials and icd10_validate for diagnosis codes. While these tools are standard for this application, they represent the skill's primary execution capabilities. The instructions include mandatory user notifications before and after each tool call to provide an audit trail and maintain transparency regarding data sources.
External Resource Fetching: The skill performs web lookups via WebFetch to validate CPT/HCPCS codes using official CMS (Centers for Medicare & Medicaid Services) domains. These resources are well-known, authoritative services necessary for the skill's function. The implementation follows standard practices for medical code verification within a payer environment.

prior-auth-review-skill