============================================================ PHASE 1: SYSTEM DISCOVERY AND MEDICATION DATA MODEL

1. Identify the medication management platform:
   • Read configuration files, dependency manifests, and environment definitions.
   • Determine the tech stack: backend framework, database, notification services, pharmacy APIs, drug databases, scheduling engine.
   • Map all services, APIs, background processors, and external integrations.
2. Map the medication data model:
   • Medication records: drug name, NDC code, strength, form, route, frequency, prescriber, start date, end date, refill count, pharmacy.
   • Dosage schedules: time-of-day, frequency, PRN conditions, taper schedules.
   • Adherence events: taken, missed, late, skipped with reason, partial dose.
   • Interaction records: drug-drug, drug-food, drug-condition flags.
   • Adverse event records: symptoms, severity, onset timing, suspected medication.
3. Map the medication lifecycle:
   • Prescription creation and verification.
   • Schedule generation and optimization.
   • Reminder delivery and confirmation.
   • Adherence recording and tracking.
   • Refill prediction and ordering.
   • Discontinuation and tapering.
4. Catalog integration points:
   • Pharmacy systems (NCPDP, HL7 FHIR, custom APIs).
   • Drug interaction databases (commercial or open-source).
   • Electronic Health Record (EHR) systems.
   • Smart pill dispensers or medication tracking devices.
   • Caregiver and family communication platforms.
   • Telehealth and prescriber notification systems.

============================================================ PHASE 2: ADHERENCE TRACKING ACCURACY

• Examine all adherence capture mechanisms:
  • Manual confirmation (patient or caregiver marks taken/missed).
  • Smart dispenser integration (automatic detection of dispense events).
  • Barcode or RFID scanning at time of dose.
  • Wearable or ingestible sensor confirmation.
• Check for timestamp accuracy on adherence events.
• Verify that late doses are distinguished from missed doses.

• Read the adherence rate calculation logic.
• Check for standard metrics: Medication Possession Ratio (MPR), Proportion of Days Covered (PDC), dose timing adherence.
• Verify that calculation handles: PRN medications, tapered schedules, hospital stays (medication provided elsewhere), temporary holds.
• Look for individual vs. medication-level vs. regimen-level adherence scoring.

• Check for time-of-day adherence analysis (morning doses vs. evening).
• Look for day-of-week patterns (weekday vs. weekend adherence).
• Examine adherence trend tracking over weeks and months.
• Verify that adherence data can be segmented by medication, condition, or regimen.

• Check for validation on adherence entries (plausibility checks, duplicate prevention).
• Examine how the system handles conflicting adherence data from multiple sources.
• Look for audit trail on adherence record modifications.
• Verify that missed-dose detection has a reasonable grace period before flagging.

============================================================ PHASE 3: INTERACTION CHECKING COMPLETENESS

• Identify the drug interaction database or API used.
• Check for interaction severity levels (contraindicated, major, moderate, minor).
• Verify that interaction checks run on every medication addition and change.
• Examine how interaction alerts are presented to users and clinicians.

• Check for food interaction flagging (grapefruit, dairy, alcohol, high-vitamin-K foods).
• Verify that condition-based contraindications are checked (renal impairment, liver disease, pregnancy, fall risk).
• Look for allergy cross-reference checking.
• Check for age-based dosing warnings.

• Evaluate completeness of the interaction database (last update date, drug count).
• Check for new medication onboarding (how quickly are new drugs added to the database).
• Verify that interaction checks include OTC medications, supplements, and herbal products.
• Look for duplicate therapy detection (two drugs in the same class).

• Check for alert severity filtering (show contraindicated always, suppress minor).
• Examine override workflows (clinician acknowledges and overrides with reason).
• Verify that overrides are logged and auditable.
• Look for alert fatigue metrics (override rate, time to acknowledge).

============================================================ PHASE 4: REFILL PREDICTION AND MANAGEMENT

• Examine the refill prediction algorithm.
• Check for days-supply tracking based on prescribed frequency and quantity dispensed.
• Verify that refill predictions account for actual adherence rate (not just prescribed rate).
• Look for early refill detection (patient requesting refill before expected need).

• Check for automated refill reminders (N days before expected run-out).
• Examine pharmacy integration for electronic refill requests.
• Verify that refill status tracking exists (requested, processing, ready, picked up).
• Look for refill failure handling (insurance denial, prior authorization required, out of stock, prescriber authorization needed).

• Check for on-hand supply tracking (actual count vs. calculated remaining).
• Verify that travel or vacation supply requests are supported.
• Look for 90-day supply optimization recommendations.
• Examine how controlled substance refill restrictions are enforced.

============================================================ PHASE 5: DOSAGE SCHEDULING OPTIMIZATION

• Read the scheduling algorithm that converts prescriptions to daily dose times.
• Check for standard timing rules (with meals, on empty stomach, at bedtime, every N hours, BID/TID/QID optimization).
• Verify that schedules respect drug-specific timing requirements (thyroid medication 30 min before breakfast, separate calcium from iron by 2 hours).
• Look for personalization based on patient daily routine.

• Examine conflict detection between medications that should not be taken together.
• Check for spacing optimization (automatically separating conflicting medications).
• Verify that schedule changes propagate to all downstream systems (reminders, dispensers).
• Look for schedule stability (minimizing changes to established routines).

• Map all reminder channels: push notification, SMS, phone call, smart dispenser alarm, caregiver notification.
• Check for escalation patterns (first reminder, second reminder, caregiver notification).
• Verify that reminder timing accounts for patient preferences and quiet hours.
• Look for adaptive reminder timing based on historical response patterns.

• Check for taper schedule support (gradual dose increases or decreases).
• Verify that alternating day dosing is handled correctly.
• Look for cyclical medication support (chemotherapy cycles, hormone therapy).
• Examine how temporary medication holds and restarts are managed.

============================================================ PHASE 6: CAREGIVER NOTIFICATION WORKFLOWS

• Map all caregiver-facing notifications: missed dose, adherence decline, interaction alert, refill needed, adverse event reported, schedule change.
• Check for notification preferences per caregiver (channel, frequency, urgency filter).
• Verify that notifications include actionable context (what happened, what to do).
• Look for acknowledgment tracking on critical notifications.

• Examine caregiver role definitions (family member, professional caregiver, nurse, pharmacist, prescriber).
• Check for role-appropriate notification filtering (family gets adherence summary, prescriber gets adverse events).
• Verify that patient consent controls which caregivers see what data.
• Look for delegation workflows (primary caregiver designates backup).

• Check for escalation paths when notifications are not acknowledged.
• Examine time-based escalation rules (notify next caregiver after N minutes).
• Verify that critical events (adverse reaction, multiple missed doses of critical medication) have immediate escalation.
• Look for emergency contact integration for life-threatening medication events.

• Check for adherence summary reports (daily, weekly, monthly).
• Verify that reports highlight trends and exceptions, not just raw data.
• Look for medication change summaries for caregivers not present at appointments.
• Examine shared care coordination views for multi-caregiver scenarios.

============================================================ PHASE 7: ADVERSE EVENT DETECTION

• Examine how adverse events are reported (patient self-report, caregiver report, clinician documentation, sensor-detected anomalies).
• Check for structured symptom capture with severity and timing.
• Verify that adverse events are linked to suspected medications.
• Look for photographic documentation support (rashes, swelling).

• Check for pattern matching between reported symptoms and known drug side effects.
• Look for temporal correlation analysis (symptom onset relative to medication start or dose change).
• Examine rechallenge and dechallenge tracking (did symptom resolve when medication stopped? recur when restarted?).
• Check for population-level signal detection across all users.

• Map the response workflow from event detection to resolution.
• Check for severity-based routing (mild to self-monitoring, severe to prescriber alert).
• Verify that adverse events trigger medication review recommendations.
• Look for regulatory reporting preparation (MedWatch format data capture).

============================================================ SELF-HEALING VALIDATION (max 2 iterations)

1. Verify all output sections have substantive content (not just headers).
2. Verify every finding references a specific file, code location, or data point.
3. Verify recommendations are actionable and evidence-based.
4. If the analysis consumed insufficient data (empty directories, missing configs), note data gaps and attempt alternative discovery methods.

• Identify which sections are incomplete or lack evidence
• Re-analyze the deficient areas with expanded search patterns
• Repeat up to 2 iterations

• Flag specific gaps in the output
• Note what data would be needed to complete the analysis

============================================================ OUTPUT

1. {MED-001}: {title}
   • Domain: {Adherence/Interactions/Refill/Schedule/Caregiver/AdverseEvent}
   • Location:

     {file:line}

   • Impact: {what could go wrong for patient safety}
   • Recommendation: {specific improvement}

• Tracking methods: {list}
• Adherence metrics: {MPR/PDC/timing/other}
• Pattern analysis: {present/absent}
• Data reliability checks: {present/absent}

• Database source: {identified or not}
• Severity levels: {N}
• Drug-food checks: {present/absent}
• Override logging: {present/absent}
• Alert fatigue mitigation: {present/absent}

• Prediction algorithm: {days-supply/adherence-adjusted/ML}
• Pharmacy integration: {electronic/manual/none}
• Failure handling: {present/absent}

• Conflict detection: {present/absent}
• Spacing optimization: {present/absent}
• Complex regimen support: {tapers/cycling/alternating}
• Adaptive reminders: {present/absent}

• Notification types: {N}
• Role-based filtering: {present/absent}
• Escalation logic: {present/absent}
• Consent management: {present/absent}

• Do NOT recommend specific drug databases or pharmacy system vendors.
• Do NOT make clinical recommendations about medication changes or dosing.
• Do NOT evaluate the clinical accuracy of interaction databases (focus on system integration).
• Do NOT ignore privacy and consent requirements for medication data sharing.
• Do NOT skip adverse event detection even if the system focuses primarily on adherence.
• Do NOT assess prescribing appropriateness (focus on adherence system capabilities).

• "Run

  /fall-risk

  to analyze how medication data feeds into fall risk prediction."
• "Run

  /caregiver-coordination

  to evaluate broader care coordination workflows."
• "Run

  /security-review

  to audit access controls on protected health information."
• "Run

  /integration-test

  to validate pharmacy API integration reliability."

============================================================ SELF-EVOLUTION TELEMETRY

• Look for the project path in

  ~/.claude/projects/

• If found, append to

  skill-telemetry.md

  in that memory directory

undefined

• Outcome: {{SUCCESS | PARTIAL | FAILED}}
• Self-healed: {{yes — what was healed | no}}
• Iterations used: {{N}} / {{N max}}
• Bottleneck: {{phase that struggled or "none"}}
• Suggestion: {{one-line improvement idea for /evolve, or "none"}}

Only log if the memory directory exists. Skip silently if not found.
Keep entries concise — /evolve will parse these for skill improvement signals.