Prior Authorization Review
Automates payer prior-authorization review and produces auditable approval, denial, or pend decisions.
The files show three healthcare connectors, pre/post-call notifications, human confirmation, audit records, coverage disclaimers, and a requirement not to overwrite existing waypoints. However, there is no complete least-privilege model, retention, encryption, de-identification, or end-to-end data-flow disclosure for member, clinical, and provider data. Demo mode also bypasses one NPI lookup, and the workflow generates authorization decisions and provider notifications. Points are deducted for incomplete sensitive-data governance, external-effect controls, and rollback.
The files show a two-stage workflow with prerequisite checks, actionable error messages, low-confidence handling, output validation requirements, and resume behavior. However, SKILL.md conflicts with README and the rubric about DENY behavior, notification output alternates between TXT and PDF, several referenced files or directories are not verified by the supplied evidence, and no committed test suite or reproducible run is provided. Static review supports a plausible happy path but not strong abnormal-path reliability.
The target users, input files, trigger phrases, and Medicare LCD/NCD limitation are reasonably clear. Non-fit boundaries, real deployment assumptions, Chinese interaction, and mainland-China network reachability are not established. The skill depends on three healthcare MCP connectors and CMS WebFetch, while commercial and Medicare Advantage cases may use policies that differ from the cited sources. Points are deducted for incomplete trigger boundaries and environment-fit evidence.
The evidence provides an overview, prerequisites, staged subskills, a rubric, tool reference, templates, and error-handling guidance, showing some progressive disclosure. However, naming is inconsistent across the path, README, and skill name; license metadata is unknown; versioning, changelog, explicit maintenance ownership, and update path are absent; and output and decision semantics are inconsistent. The documentation is readable but governance and maintainability are incomplete.
The files define collection, code and provider validation, policy matching, medical-necessity assessment, human confirmation, and notification generation, so the intended workflow could reduce manual effort. There is no execution evidence, test result, or representative verified output, and policy applicability, extraction correctness, and legal/compliance quality still require substantial professional review. Static evidence supports only limited core-task plausibility and not directly usable high-stakes results.
The evidence consists primarily of author-authored procedures, pseudocode, schemas, and repository CI configuration. The CI workflows concern code review rather than this skill and do not constitute key-path tests. No execution logs, third-party evaluation, or independently reproducible coverage are supplied, so verifiability remains limited.
- This is a high-stakes healthcare prior-authorization workflow; qualified professionals must review it, and generated APPROVE, PENDING, or notification content must not be treated as final decisions without that review.
- Before deployment, reconcile DENY/PENDING semantics, TXT/PDF output conventions, and referenced paths, and add tests covering the key paths.
- Member and clinical records are sensitive data; the supplied files do not specify de-identification, access control, retention, encryption, or cross-border processing requirements.
- Core operation depends on three healthcare MCP connectors and CMS WebFetch. The supplied files do not establish reachability from mainland China or provide an equivalent safe fallback when connectors are unavailable.
- The demo NPI bypass is restricted to a specific sample combination, but production deployment should isolate it and prevent accidental activation by sample identifiers.
What it does & when to use it
This skill is designed for health insurance payers, including Medicare Advantage, commercial, and Medicaid MCO organizations. It processes authorization forms and clinical documentation, validates providers and codes, and assesses medical necessity against coverage policies. Its workflow has two subskills—intake and assessment, followed by decision and notification—and produces assessment data, a final decision, and a provider notification letter. It requires three healthcare MCP connectors and readable input files.
Checks for and resumes incomplete authorization requests; reads the PA request form, clinical notes, imaging reports, laboratory results, and optional supporting documentation; verifies providers through the NPI MCP, validates ICD-10 diagnoses through the ICD-10 MCP, validates CPT/HCPCS codes through the CMS Fee Schedule, and searches coverage policies through the CMS Coverage MCP; extracts clinical information and maps evidence to policy criteria; generates an APPROVE, DENY, or PEND recommendation; creates waypoints/assessment.json, waypoints/decision.json, and outputs/notification_letter.txt; and documents an audit trail.
- A health insurer needs to assess medical necessity for a prior-authorization request with supporting clinical documentation.
- A Medicare Advantage or Medicaid MCO reviewer needs to validate whether a requested service aligns with CMS coverage policy.
- A payer review team needs to verify provider NPI data, ICD-10 diagnosis codes, and CPT/HCPCS procedure codes during intake.
- A utilization-management team needs decision-specific rationale, documentation requests, and provider notification letters for approval, denial, or pend outcomes.
Pros & cons
- Covers intake, provider and code validation, policy lookup, medical-necessity assessment, decisioning, and notification in one workflow.
- Produces explicit approval, denial, and pend outputs with supporting rationale.
- Supports resuming incomplete requests through waypoint files and documents an audit trail.
- The README describes the relevant hosted MCP servers as requiring no setup.
- Requires CMS Coverage, ICD-10, and NPI MCP connectors to be available together.
- Depends on readable PDF documentation; extraction confidence below 60% requires a user choice about continuing, requesting more documentation, or aborting.
- The supplied source does not provide a test suite, performance validation, or production deployment evidence.
- The README states that the collection is provided under Anthropic's terms of service, but does not specify an independent license or pricing.
How to install
Install the healthcare plugin using the repository README commands:
/plugin marketplace add anthropics/healthcare
/plugin install healthcare@healthcare
The repository bundles 11 skills; this profile covers only plugins/healthcare/skills/prior-auth/SKILL.md. The README does not provide a separate installation command for this skill.
How to use
After installing the healthcare plugin, invoke:
Use the prior-auth-review-skill
Then choose either to upload your own PDF files or to load the configured CT-guided lung-biopsy sample case. Your own files should include a PA request form, clinical notes/H&P, diagnostic imaging reports, and laboratory results; additional supporting documentation is optional. All three required MCP connectors must be configured.