Medical Protocol
A workflow system that gives Claude Code the domain knowledge, verification tools, and privacy guardrails to build medical software correctly.
The Problem
Edge cases fail silently. A Winters formula off by 0.3 won't throw an error — but it changes the clinical decision. LLMs don't know what they don't know.
How It Works
Paste the protocol URL into Claude Code. It fetches workflows, skills, and CLI tools automatically.
Tell the agent what you need in clinical language. “I need an acid-base interpreter with compensation checks.”
The agent cross-checks every calculation against the medprotocol CLI — 1,198 test assertions run at build time.
Domain Skills
Each skill encodes domain-specific knowledge — formulas, edge cases, clinical thresholds, and UI patterns.
CLI Verification
9 clinical calculators with deterministic outputs. The agent runs medprotocol CLI, compares against the UI, and flags discrepancies.
Test Coverage
Every formula edge case. Every compensation threshold. Every unit conversion. Validated at build time before any component ships.
Architecture
Pre-tool hooks block patient data from leaving the machine. Zero external API calls with PHI.
TypeScript strict mode + CLI tests run before deploy. Runtime errors caught before they reach patients.
Components installed via CLI — no manual wiring. Dependency graph resolved automatically.
Full strict mode across all components. No implicit any, no unchecked nulls.
Who It's For