Medical Priority Scorer

Produce deterministic priority scores for medical messages without mutating any state.

Quick Triggers

• Rank messages by medical urgency for callback priority
• Classify messages into P1/P2/P3 queue
• Score follow-up priority from triaged messages

Recommended Chain

medical-triage -> medical-priority-scorer -> medical-entity-extractor

Execute Workflow

1. Accept input from medical-triage containing triaged messages
2. Score each message with:
   • urgency_score in range [0, 1] (based on triage category)
   • sentiment_score in range [-1, 1] (anxiety, distress, frustration)
   • recency_score in range [0, 1] (how recent the message is)
   • patient_context_score in range [0, 1] (chronic conditions, known patient)
3. Compute priority_score on a 0-100 scale:
   • priority_score = 100 * (0.50*urgency_score + 0.25*sentiment_score_risk + 0.15*recency_score + 0.10*patient_context_score)
   • sentiment_score_risk = max(0, -sentiment_score) (negative sentiment = higher risk)
4. Assign buckets:
   • P1 for priority_score >= 75 (immediate attention)
   • P2 for priority_score >= 50 and < 75 (same-day)
   • P3 for < 50 (routine)
5. Produce plain-language evidence tokens that explain the score

Input Format

[
  {
    "id": "msg-123",
    "category": "urgent",
    "subject": "Medication side effects",
    "from": "patient@example.com",
    "date": "2026-02-27T10:30:00Z",
    "body": "I've been feeling dizzy since starting the new medication..."
  }
]

Output Format

[
  {
    "id": "msg-123",
    "priority_score": 78,
    "priority_bucket": "P1",
    "urgency_score": 0.8,
    "sentiment_score": -0.4,
    "recency_score": 1.0,
    "patient_context_score": 0.6,
    "evidence": "Urgent triage category + negative sentiment (concern) + very recent message + known patient"
  }
]

Enforce Boundaries

• Never write to databases or files
• Never send messages or trigger outbound channels
• Never create reminders or execute actions
• Never bypass routing or approvals

Handle Errors

1. Reject schema-invalid inputs
2. Return field-level reasons when scoring cannot be computed
3. Fail closed if required scoring features are missing

Integration

This skill can be invoked via the OpenClaw CLI:

openclaw skill run medical-priority-scorer --input '[{"id":"msg-1","category":"urgent",...}]' --json

Or programmatically:

const result = await execFileAsync('openclaw', [
  'skill', 'run', 'medical-priority-scorer',
  '--input', JSON.stringify(triagedMessages),
  '--json'
]);

Recommended Model: Claude Sonnet 4.5 (openclaw models set anthropic/claude-sonnet-4-5)