Skip to main content

Open-source design-controlled development infrastructure for medical device and SaMD teams — includes dhfkit

Project description

MedHarness

AI harness and DHF tooling for medical device software teams.

PyPI License: MIT Python 3.11+

MedHarness structures how AI agents interact with a Design History File under IEC 62304 / FDA-regulated software projects. It pre-computes DHF context before an agent runs, enforces approval gates the agent must pass through, and commits decisions back into the DHF — so the engineer controls the feedback loop, not the agent.

It combines two packages:

  • medharness — CLI harness, CI gates, CR workflows, project scaffolding (init)
  • dhfkit — standalone DHF engine for items, traceability, document generation, schema validation

Install

pip install medharness[full]

[full] pulls in optional extras: ai (Gemini-based AI review) and docs (PDF export via WeasyPrint). Omit for a minimal install — the DHF engine (dhfkit) is always included.

Verify:

medharness --help
dhfkit --help

From source (development):

git clone https://github.com/itercharles/MedHarness
cd MedHarness
pip install -e ".[dev]"
pytest dhfkit/tests/ tests/

Quick Start

medharness init is zero-prompt — it scaffolds a single-repo project in the current directory. The project name is derived from the directory name.

mkdir my-medical-device && cd my-medical-device
python -m venv .venv && source .venv/bin/activate
pip install medharness
medharness init

After init completes, here's what exists on disk:

my-medical-device/                  # single repo — DHF + source together
├── DHF/
│   ├── config/
│   │   ├── global.yaml             # project name, lifecycle states
│   │   └── doc_types/              # one YAML per type (SYS, CRS, SRS, SWDD, CR, …)
│   ├── items/                      # one YAML file per requirement / risk / CR
│   │   ├── 01_crs/                 # Customer Requirements (CRS-NNN.yaml)
│   │   ├── 02_sys/                 # System Requirements (SYS-NNN.yaml)
│   │   ├── 03_srs/                 # Software Requirements (SRS-NNN.yaml)
│   │   ├── 06_cr/                  # Change Requests (CR-NNN.yaml)
│   │   └── ...                     # Use Cases, SOUP, Risk, Defects, etc.
│   ├── test-results/
│   ├── documents/
│   │   ├── specs/                  # Jinja2 spec templates (.j2)
│   │   └── plans/                  # development_plan.md, verification_plan.md, …
│   └── README.md
├── .github/
│   ├── workflows/
│   │   ├── engineering-control.yml # main CI: CR validation + coverage gate + evidence
│   │   ├── cr-analyze.yml          # CR analysis from issues
│   │   ├── cr-develop.yml          # CR development with AI
│   │   ├── cr-spec-iterate.yml     # iterate on spec with review feedback
│   │   ├── cr-transition.yml       # transition CR state in DHF
│   │   ├── cr-complete.yml         # auto-close CR on PR merge
│   │   └── review-pr.yml           # AI-assisted PR review
│   └── prompts/                    # AI prompt templates
├── .claude/skills/                 # Claude Code skills
├── tests/                          # product test suite
├── CLAUDE.md                       # agent entrypoint
├── .gitignore
└── README.md                       # project README

The scaffolded items are starter samples — replace them with your project's real requirements, architecture, and plans before using this for a regulated product.

Initialize git and push:

git init && git add -A
git commit -m "feat: initialize My Medical Device with MedHarness"
git remote add origin https://github.com/<org>/my-medical-device
git push -u origin main

How a Change Request flows

Every non-trivial change starts as a Change Request (CR) in the DHF. CRs move through AI-assisted stages:

Issue → cr-analyze → cr-spec-iterate → cr-develop → cr-transition → cr-complete
Stage Trigger What MedHarness does
cr-analyze Issue labeled CR Pre-computes DHF context, runs Claude to write a technical spec, commits the spec to DHF/documents/cr-specs/
cr-spec-iterate Review feedback on spec Resumes Claude session with feedback, updates spec, pushes revisions
cr-develop Spec approved Injects $DHF_CONTEXT, runs Claude to implement code, opens a PR
cr-transition PR events Transitions the CR to in_review/approved in the DHF
cr-complete PR merged Transitions the CR to complete, generates closing evidence

At each stage MedHarness:

  1. Pre-computes contextmedharness dhf context for-stage <stage> returns focused JSON
  2. Injects into agent environment$DHF_CONTEXT is available to Claude
  3. Captures decisions backmedharness dhf item transition --commit --push
  4. Stores session IDsmedharness ci claude-session put/get for iterative review loops

The workflow YAML files for each stage are scaffolded by medharness init into .github/workflows/.


Test Coverage Gate

The CI gate (medharness ci test-coverage) enforces that every verifiable requirement has at least one passing test linked to it.

JUnit XML contract

Tests must emit JUnit XML with properties linking to DHF item IDs:

<testcase name="test_TC_SYS_005_001_validates_link_format">
  <properties>
    <property name="medharness.id" value="TC-SYS-005-001"/>
    <property name="medharness.links" value="SYS-005"/>
  </properties>
</testcase>

All property names are defined as constants in medharness/contracts.py:

Property Purpose
medharness.id Test case identifier (e.g. TC-SYS-005-001)
medharness.links Comma-separated DHF item IDs the test covers
medharness.title Human-readable test title (optional)
medharness.reviewer Reviewer name (optional)
medharness.review_date Review date (optional)
medharness.review_status Review status (optional)

Python / pytest

Use pytest's record_property in conftest.py:

@pytest.fixture(autouse=True)
def _inject_medharness_metadata(request, record_property):
    doc = request.function.__doc__ or ""
    tc_id = extract_tc_id_from_name(request.node.name)
    links = parse_links(doc)   # extract @links:SYS-005 from docstring
    if tc_id:
        record_property("medharness.id", tc_id)
    if links:
        record_property("medharness.links", ",".join(links))

TypeScript / Vitest / Playwright

Use custom JUnit reporters that emit <properties> blocks for medharness.links. Reference implementations are available in the WebTPS repo.

Running the gate locally

# From project root
pytest tests/ -q --junitxml=test-results/results.xml
medharness --dhf DHF ci test-coverage --junit-dir test-results

Expect output like:

[test-coverage] SRS: 12/14 covered
      ↳ uncovered: SRS-012
      ↳ uncovered: SRS-008

The command exits non-zero when gaps exist, blocking CI.


CLI Reference

Scaffold

medharness init                     # zero-prompt single-repo project setup

DHF operations (run with --dhf DHF)

medharness --dhf DHF dhf item list --type SYS
medharness --dhf DHF dhf item get SYS-001
medharness --dhf DHF dhf item create --type SYS --data '{"title": "My req"}'
medharness --dhf DHF dhf item update SYS-001 --data '{"title": "Updated"}'
medharness --dhf DHF dhf item delete SYS-001
medharness --dhf DHF dhf item transitions CR-001
medharness --dhf DHF dhf item transition CR-001 approved --by "Alice"
medharness --dhf DHF dhf validate schema
medharness --dhf DHF dhf validate traceability
medharness --dhf DHF dhf doc list
medharness --dhf DHF dhf doc generate SYS
medharness --dhf DHF dhf doc export SYS          # PDF output (requires `[docs]`)
medharness --dhf DHF dhf test list
medharness --dhf DHF dhf config doc-types

CI gates

medharness ci dhf-validate --dhf DHF
medharness ci test-coverage --dhf DHF --junit-dir test-results
medharness ci evidence bundle --dhf DHF --out-dir artifacts

CR workflow commands

medharness cr workflow intake-github-issue-ci      # CR intake from issue
medharness cr workflow complete-from-github-pr     # CR completion on PR merge

Agent session helpers

medharness ci claude-session put <pr_number> <session_id>
medharness ci claude-session get <pr_number>

Python API

Use DHFClient for high-level operations (recommended for product repo automation):

from medharness.client import DHFClient

client = DHFClient(Path("DHF"))

cr   = client.get_item("CR-034")
spec = client.get_cr_context("CR-034")   # {"cr": {...}, "spec": "..."}
client.transition_item("CR-034", "in_review", performed_by="alice")

Or use dhfkit standalone (no dependency on medharness):

from dhfkit.local_adapter import LocalDHFAdapter

adapter = LocalDHFAdapter(Path("DHF"))
items  = adapter.list_items("SRS")

Repository layout

Directory Purpose
medharness/ CLI harness, CI gates, CR workflows, init scaffolding
dhfkit/ DHF engine: items, lifecycle, traceability, document generation
dhfkit/templates/ Starter DHF scaffold — config, specs, plans, sample items, CI workflows
tests/ MedHarness and dhfkit test suites
docs/ Architecture, ADRs, compatibility contracts

dhfkit has no dependency on medharness — the engine can be used standalone.


Docs


License

MIT — see LICENSE.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

medharness-0.3.0.tar.gz (121.4 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

medharness-0.3.0-py3-none-any.whl (159.3 kB view details)

Uploaded Python 3

File details

Details for the file medharness-0.3.0.tar.gz.

File metadata

  • Download URL: medharness-0.3.0.tar.gz
  • Upload date:
  • Size: 121.4 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for medharness-0.3.0.tar.gz
Algorithm Hash digest
SHA256 d79fc96d84f7f8ffeaec5f71bbbf3ff2876725606fd04fc364f270f817f7bd24
MD5 ef0756040b8d346a63d7eb0805a76dea
BLAKE2b-256 e8bbf28477de373b066084f4d06aefdf475f3f8a721518bd1ad708c3e9e8de7b

See more details on using hashes here.

Provenance

The following attestation bundles were made for medharness-0.3.0.tar.gz:

Publisher: release.yml on itercharles/MedHarness

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file medharness-0.3.0-py3-none-any.whl.

File metadata

  • Download URL: medharness-0.3.0-py3-none-any.whl
  • Upload date:
  • Size: 159.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for medharness-0.3.0-py3-none-any.whl
Algorithm Hash digest
SHA256 43c839c1f3e526b0144a109647cdd6d14d6944d6b80ad61e92566d9844833af3
MD5 9961ed358ee6fb1cca4087fe384da6c7
BLAKE2b-256 d9b5660c114436563dc946125f64c969d0e517e7cfadcff4298c9f5495b9d7de

See more details on using hashes here.

Provenance

The following attestation bundles were made for medharness-0.3.0-py3-none-any.whl:

Publisher: release.yml on itercharles/MedHarness

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page