| name | safety-md-hazard-workshop |
| description | User-invoked skill for running a clinical safety hazard-identification workshop and updating a SAFETY.md hazard log. |
| disable-model-invocation | true |
SAFETY.md Hazard Workshop
Use only when the user explicitly asks for this skill or asks to identify clinical safety hazards.
Read first
GLOSSARY.md- Existing
SAFETY.md, SAFETY-CASE.md, and HAZARD-LOG.md if present.
references/hazard-archetypes.md
Steps
- Establish intended use, clinical context, users, patient population, and deployment assumptions. Completion: unknowns are explicit.
- Walk every hazard archetype in
references/hazard-archetypes.md. Completion: each archetype is represented by hazards, marked not applicable, or recorded as a question.
- For each candidate hazard, separate description, cause, effect, controls, evidence, residual risk, owner, and review status. Completion: no hazard is just a vague concern.
- Update or draft
HAZARD-LOG.md. Completion: new hazards have stable IDs and [PENDING] fields where needed.
- Add unresolved clinical questions to
queries.md. Completion: the user can see exactly what a CSO/SME must answer.
Never do
- Do not treat starter hazards as a completed hazard workshop.
- Do not invent risk acceptance or CSO review.
- Do not hide high/unknown residual risks by wording them vaguely.
Completion criteria
- Candidate hazard families have been systematically considered.
- Each recorded hazard has cause, effect, controls or control gaps, residual-risk status, and review status.
- Unanswered clinical/domain questions are captured.
Common Gotchas
- “User enters wrong value” is usually a cause, not the full hazard.
- Security and interoperability issues become clinical safety hazards only when they can plausibly contribute to patient harm.
- A short hazard log often means insufficient analysis, not a safe product.
