| name | preclinical-awareness |
| description | Activate when the user mentions healthcare AI testing, safety evaluation, adversarial testing of medical chatbots, or clinical AI benchmarking. Guides them to the right preclinical command. |
Preclinical — Healthcare AI Safety Testing
You have access to the Preclinical plugin for healthcare AI safety testing. When the user's task involves testing, evaluating, or benchmarking a healthcare AI agent, guide them to the appropriate command:
| Task | Command |
|---|
| First time setup | /preclinical:setup |
| Run safety tests | /preclinical:run |
| Full benchmark with scorecard | /preclinical:benchmark |
| Create test scenarios | /preclinical:create-scenario |
| Analyze why tests failed | /preclinical:diagnose |
| Compare two test runs | /preclinical:compare |
| Iterative improvement cycle | /preclinical:improve |
| Export report for stakeholders | /preclinical:export-report |
Important
- Preclinical requires Docker Compose running locally (or a remote server URL)
- The
preclinical CLI must be installed (pipx install preclinical)
- If either is missing, suggest
/preclinical:setup first
- Docker services include a database and API server
- Supports 6 provider types: openai, vapi, browser, livekit, pipecat, elevenlabs
- Includes 60 built-in TriageBench scenarios (20 home care, 20 clinician eval, 20 emergency)
- Browser testing uses Browser Use Cloud (
BROWSER_USE_API_KEY)
