SaMD-os: Getting Started in 30 Minutes

01

What is SaMD-os

2 min

SaMD-os is a structured working environment for teams building software medical devices. It sits upstream of your eQMS: a draft layer where you author regulatory documents and get AI pre-review feedback.

Key concept

Skills generate documents. Agents review them. No agent creates controlled artifacts. They only find what's missing.

It doesn't replace your eQMS. Documents only become controlled when you export them after human review.

02

Clone & Customize

3 min

No npm. No pip. No Docker. Clone and open in Claude Code.

# Clone the repo
$ git clone https://github.com/mc-barnes/samd-os.git
$ cd samd-os

# Open in Claude Code
$ claude

# Verify setup
$ ./scripts/status.sh
    

Open CLAUDE.md and fill in:

Team roster: names, roles, Slack/GitHub handles
Device context: classification, predicate, clinical domain
Standards: confirm or adjust the pinned editions

03

Explore the Structure

3 min

Each folder maps to a regulatory function. Each has its own CLAUDE.md that loads context on demand.

product/PRDs, strategy, competitive, customers regulatory/Design controls, risk, submissions, DHF clinical/Intended use, usability, clinical evaluation quality/CAPA, complaints, audit prep engineering/Bugs, RFCs, IEC 62304 SDLC, SOUP analytics/Post-market surveillance, metrics team/Onboarding, retros, decisions .claude/skills/17 skills + 5 agent personas examples/Pre-generated reference artifacts

04

Draft Your First PRD

10 min

Use the SaMD PRD skill to generate a regulatory-ready product requirements document with intended use, clinical requirements, design inputs, risk sections, and EHR integration.

# Say this in Claude Code:
"Write a SaMD PRD for a neonatal SpO2 overnight triage system"

# Or use the trigger directly:
$ /prd-writer-samd
    

Compare your output against examples/samd-prd-example.md.

All SaMD skills:

prd-writer-samd design-controls risk-management change-impact design-review fhir-builder

05

Run an Agent Review

8 min

Run a reviewer on your draft. Each agent returns structured findings with severity, standards citations, and a verdict.

# Single reviewer:
"Run regulatory-reviewer on product/prds/spo2-triage-v1.md"

# Full panel (all 5 in parallel):
"Run review-panel on product/prds/spo2-triage-v1.md"
    

Agent	Focus	Example Finding	Verdict
Regulatory	IFU, traceability	Intended use too broad	NEEDS REVISION
Safety	Risk controls, AFAP	No AFAP justification	ACCEPTABLE
QA	CAPA, audits	Missing decision records	ACCEPTABLE
Cybersecurity	SBOM, 524B	47 entries no versions	CONCERN
Clinical	Algorithms, alarms	Threshold not cited	ACCEPTABLE

Important

Most conservative wins. In a panel review, one FAIL blocks the panel. Cost: ~$0.50 per full panel review.

06

What's Next

4 min

You've completed the core loop: draft → review → fix → export.

Recommended first week:

Day 2: Draft a real PRD, run review panel, compare findings to your RA's feedback
Day 3: Try reverse-engineer skills on your codebase
Day 4: Decide to adopt, customize, or defer

Reverse-engineering skills:

code-to-soup-register code-to-design-inputs code-to-hazard-candidates

File	What you'll find
`examples/`	Full reference artifacts (design controls XLSX, risk analysis XLSX, FHIR bundle JSON, SaMD PRD). Use these to benchmark your own outputs.
`docs/adoption-guide.md`	How to roll out to your team: who runs what, where it fits alongside your eQMS, cost breakdown ($5–$15/month typical), and a readiness checklist.
`docs/responsible-use.md`	What agents can and can't do, known limitations (frozen standards knowledge, no institutional memory, citation accuracy), and validation evidence.
`docs/auditor-briefing.md`	How to explain AI-assisted drafting to auditors. Covers the draft/controlled boundary, human accountability, and what to include in your QMS documentation.