<!-- # COPYRIGHT NOTICE # This file is part of the "Universal Biomedical Skills" project. # Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu> # All Rights Reserved. # # This code is proprietary and confidential. # Unauthorized copying of this file, via any medium is strictly prohibited. # # Provenance: Authenticated by MD BABU MIA -->

---

name: 'prior-auth-coworker' description: 'Prior Auth Review' measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes. allowed-tools: - read_file - run_shell_command

Prior Authorization Coworker

This skill acts as an automated utilization management reviewer. It takes unstructured clinical notes and a procedure code, compares them against internal policy criteria (e.g., conservative therapy failure), and renders a decision.

When to Use This Skill

• When a user asks to "review a prior auth request".
• When checking if a patient qualifies for a specific procedure (e.g., MRI).
• When you need to generate a structured approval/denial letter justification.

Core Capabilities

1. Policy Matching: Checks against specific criteria (e.g., "Pain > 6 weeks").
2. Trace Generation: Produces an "Anthropic-style" <thinking> trace for auditability.
3. Structured Output: Returns a JSON object with decision, reasoning, and timestamps.
4. Microsoft Payer-Side Multi-Agent Review: Supports Microsoft Agent Framework architecture patterns with four Foundry Hosted Agents for compliance, clinical, coverage, and synthesis review; gate-based decision rubrics; MCP healthcare data access; confidence scoring; audit trails; Azure Container Apps deployment via azd; and human-in-the-loop review before denial or approval support.

Workflow

1. Extract Data: Parse the clinical note and procedure code from the user's input.
2. Execute Review: Run the coworker script.
3. Present Decision: Output the JSON decision and the reasoning trace.

Example Usage

User: "Check if this patient qualifies for an MRI of the Lumbar Spine: Patient has had back pain for 2 months, tried PT but it didn't work."

Agent Action:

python3 Skills/Clinical/Prior_Authorization/anthropic_coworker.py --code "MRI-L-SPINE" --note "Patient has back pain > 2 months. Failed PT."

Supported Policies

• MRI-L-SPINE (Lumbar Spine MRI)

References

• https://github.com/microsoft/Prior-Authorization-Multi-Agent-Solution-Accelerator

<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->