# HugoScore CAIHL Draft Report: Your Healthcare Agent

Status: Draft for human review
Last reviewed: 2026-06-22
Review method: Deep public-source review of yourhealthcareagent.org home, setup, instructions, about, guides, raw agent spec, GitHub README and repository metadata, Anthropic Privacy Policy, Consumer Terms, Consumer Health Data Privacy Policy, Usage Policy, and Claude Projects help documentation. No hands-on Claude Project setup, Cowork portal testing, security/privacy audit, legal review, medical review, or outcome validation.
Service: Your Healthcare Agent
Vendor: Nick Dawson / independent open-source project
URL: https://yourhealthcareagent.org/
Category: Patient care navigation and health copilot AI

## Summary

Your Healthcare Agent is a source-available public instruction layer and guide series that helps a patient or caregiver configure Claude as a persistent healthcare navigation agent. The "product" is mostly a `CLAUDE.md` behavioral specification plus guides, not a separate hosted medical AI application. Users paste the spec into Claude Project instructions, upload insurance cards, plan documents, EOBs, denial letters, bills, and other records, then ask Claude to help interpret, draft, track, and act.

From a CAIHL perspective, the design is strongly patient-directed. The project explicitly starts from presumed agency: the user is treated as someone with standing in their own healthcare or in the care of someone they support. The agent is meant to help with coverage, billing, denials, records access, appointments, prescription coverage, and appeal tracking. The main caution is that the workflow shifts sensitive health, insurance, and financial documents into Claude/Anthropic's consumer platform, and full functionality is framed around Claude Pro, Projects, and Cowork. The project gives patients leverage, but users still need privacy controls, verification habits, and access to documents and portals.

## Evidence Reviewed

- Official homepage: https://yourhealthcareagent.org/
- Setup guide: https://www.yourhealthcareagent.org/start/
- Agent instructions page: https://www.yourhealthcareagent.org/instructions/
- Raw agent spec: https://www.yourhealthcareagent.org/agent-spec.md
- About page: https://www.yourhealthcareagent.org/about/
- Guides index: https://www.yourhealthcareagent.org/guides/
- Denial appeal worked guide: https://www.yourhealthcareagent.org/guides/appeal-a-denial/
- GitHub repository: https://github.com/nickpdawson/yourhealthcareagent
- GitHub README: https://raw.githubusercontent.com/nickpdawson/yourhealthcareagent/main/README.md
- Claude Projects help: https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects
- Anthropic Privacy Policy: https://www.anthropic.com/legal/privacy
- Anthropic Consumer Terms: https://www.anthropic.com/legal/consumer-terms
- Anthropic Consumer Health Data Privacy Policy: https://www.anthropic.com/legal/consumer-health-data-privacy-policy
- Anthropic Usage Policy: https://www.anthropic.com/legal/aup

## CAIHL Profile

- Who does this AI serve? Patient-directed and rights-aware. The official materials frame the agent as serving the patient or caregiver, not an insurer, provider, or payer workflow.
- Can patients tell AI is involved? Yes. The setup explicitly uses Claude, Claude Projects, and the published `CLAUDE.md` instruction block.
- Can patients meaningfully choose? Yes, with platform and cost limits. Use is voluntary and forkable, but full use depends on Claude account access, often Claude Pro, and comfort uploading sensitive documents.
- Can patients correct or challenge what the AI produces? Partial. The spec tells the agent to draft, explain, and help the user verify; users can edit all drafts, but no independent QA or correction workflow was found.
- Does it help patients understand or act? Yes. It supports plan-type identification, rights explanation, EOB/bill translation, appeal drafting, record requests, appointment prep, prescription/prior-auth support, and portal tracking.

## Agency Interpretation

Your Healthcare Agent is one of the more CAIHL-native patient agency projects in the directory because it is explicitly designed around critical reflection and strategic action. It does not merely answer health questions. It helps users understand the governing system: plan type, ERISA limits, Medicare Advantage differences, Medicaid fair hearing rights, HIPAA right of access, No Surprises Act context, denial deadlines, and where to verify official rules.

The strongest agency value is practical stamina. The denial guide describes the agent reading across documents, finding contradictions, drafting the appeal, asking for a collections hold, and tracking insurer and provider portals until a refund is issued. That is not clinical decision support; it is administrative self-defense with AI assistance.

The constraint is that this empowerment depends on a general-purpose AI platform that the project does not control. Anthropic's privacy policy says consumer Inputs and Outputs may be used for model training unless the user opts out; connected third-party services may receive inputs, outputs, and instructions when Claude acts on the user's behalf; and health data submitted through Claude may fall under Anthropic's consumer health data disclosures in relevant states. The project's own static site appears lightweight, but the actual patient data handling happens inside Claude and any portals, cloud files, or services the user connects.

## Key Unknowns

- Whether most users understand the privacy implications of uploading insurance cards, EOBs, denial letters, bills, plan documents, and medical records into Claude.
- Whether the setup guide should more prominently instruct users to review Claude privacy settings and opt out of model training before uploading sensitive health documents.
- Whether Claude Cowork/browser portal actions work reliably and safely for insurer, provider, pharmacy, and billing portals.
- Whether the agent spec's legal, insurance, and health-adjacent guidance is accurate across current state, federal, ERISA, Medicare, Medicaid, and plan-specific contexts.
- Whether organizations using this workflow for clients have qualified human review, disclosure, consent, authorization, and data-handling safeguards.
- Whether the $30,000 refund case in the denial guide is reproducible beyond one anecdotal worked example.
- Whether multilingual, low-literacy, disability, mobile-only, and caregiver/proxy needs are supported.
- Whether the public GitHub repository license metadata fully matches the README's Creative Commons Attribution claim; no separate `LICENSE` file was found in the repository root during this review.

## Publication Recommendation

Ready for human review as a draft profile. Publish as a patient-directed healthcare navigation agent spec or instruction layer, not as a standalone clinical AI product. Confidence should stay medium until the privacy/onboarding language, Claude settings guidance, Cowork action safety, accessibility, and independent review needs are addressed.