| name | clinic-protocol-json |
| description | Use for synthetic clinic tasks that require protocol-bound clinical decision-support JSON from a runtime environment, especially respiratory/CAP, pediatric head injury, potassium repletion, care-management routing, and observation-window retrieval cases. |
Clinic Protocol JSON
Use this skill when the user asks for a structured JSON answer for a synthetic clinic case using a runtime environment and an answer_template.json.
Core Workflow
- Read the user prompt and the full
input/payloads/answer_template.json. The template is the output contract; preserve required keys, enum spellings, nullability, numeric precision, and ordering rules.
- Read
environment_access.md for the base URL and allowed endpoints. Use only those endpoints. Do not mutate the environment or place orders.
- Extract the target
case_id and task_id. Fetch GET /api/cases/{case_id} first; it usually returns the case, patient, findings, observations, medications, allergies, problems, imaging, care registry, and SDOH facts.
- Fetch
GET /api/protocols and the protocol matching the case type or prompt. Use aggregate endpoints only when the case detail is incomplete.
- Build the answer from source data and protocol rules, not from prior examples. Include only stable identifiers that support selected facts.
- Return exactly one JSON object. No markdown, comments, prose, or extra top-level keys.
Useful protocol mapping:
| Case type / prompt | Protocol |
|---|
| adult respiratory infection, CAP | RESP-CAP-2026 |
| pediatric head injury, concussion triage | PEDS-HEAD-2026 |
| potassium replacement / repletion | K-REPLETION-2026 |
| care-management routing | CM-HIGH-RISK-2026 |
| observation window / protocol gate | OBS-WINDOW-2026 |
Source Handling Rules
- Treat observations as authoritative only when
status is final, unless the template or protocol explicitly says otherwise.
- For observation-window tasks, filter by target patient, target code, status, and
from <= effective_time < to. Sort matches by effective_time ascending, then identifier ascending when requested.
- Exclude
preliminary, entered-in-error, and canceled observations from positive protocol decisions. List them as exclusions only when the template asks for excluded distractors.
- Prefer case-specific bundled data from
/api/cases/{case_id} over broad endpoint lists because broad lists include distractors.
- Use active medications, active allergies, and active problems unless the task asks for historical context.
- Evidence IDs should be identifiers for facts actually used. Follow template ordering; if no rule exists, put the case ID first, then observations/imaging/protocol sources by relevance.
- Set safety-check booleans to
true only when the answer avoids the unsupported claim named by that check. Do not claim absent findings unless the record explicitly documents absence or a protocol permits the inference.
Respiratory / CAP
Use final vitals, respiratory tests, imaging, active allergies, active respiratory problems, and the respiratory protocol.
- Assess community-acquired pneumonia when the record has focal consolidation/infiltrate or a pneumonia-compatible final CXR impression plus respiratory symptoms.
- Assess viral URI/supportive care when protocol evidence does not support pneumonia and bacterial treatment is not indicated.
- Escalate to ED when protocol triggers are present: room-air oxygen saturation below the protocol threshold, respiratory rate at or above the protocol threshold, hypotension, confusion, sepsis concern, immunocompromise, or multilobar disease.
- Use outpatient close follow-up for pneumonia without ED triggers; use the protocol follow-up interval.
- Red flags should reflect documented facts: borderline hypoxemia, severe hypoxemia, pleuritic chest pain, respiratory distress, confusion, hemoptysis, persistent fever, or worsening shortness of breath.
- Recommended tests should be protocol-indicated and case-supported: chest x-ray, viral PCR, pulse-ox recheck, and basic CBC only when indicated by the case/template.
- Medication strategy must respect active allergy classes. Avoid beta-lactam/penicillin, sulfonamide, macrolide, or tetracycline classes when active allergies implicate them. For ED disposition, defer antibiotic selection to ED if the template supports that strategy. For outpatient CAP, use the protocol-supported outpatient antibiotic strategy and populate drug details from case/protocol/order information; if none is supplied, rely on standard adult outpatient CAP conventions without inventing an allergy conflict.
- Stabilization actions are only for immediate needs such as supplemental oxygen or urgent ED transfer.
- Return precautions should cover worsening dyspnea/shortness of breath, hypoxia, confusion, persistent fever, chest pain, hemoptysis when available in the template.
Pediatric Head Injury
Use final neurologic observations, history findings, the head-injury protocol, and the template's red-flag vocabulary.
- High risk / ED route: repeated vomiting, worsening severe headache, seizure, basilar skull signs, focal neurologic deficit, GCS below 15, prolonged loss of consciousness, or other urgent route trigger from protocol.
- Intermediate risk: concussion or mild TBI features without urgent triggers, such as head impact with nausea, headache, brief symptoms, or mild coordination abnormality without focal weakness.
- Low risk: minor head injury with normal exam and no concussion features.
- Imaging recommendation should match risk: no immediate CT for non-urgent low/intermediate clinic observation; CT/ED consideration or urgent CT for urgent triggers.
red_flags are present findings only. absent_red_flags should include only explicitly negated urgent features, such as no loss of consciousness, no repeated vomiting, no seizure, or no focal weakness.- Restrictions generally include short cognitive/physical rest, return-to-learn accommodations, no high-risk sports until cleared, and symptom-appropriate driving restriction. Use stricter clearance wording for ED/high-risk or provider-clearance scenarios.
- Use the protocol follow-up interval: later follow-up for stable outpatient mild TBI, earlier/ED route for urgent triggers.
Potassium Repletion
Use final serum potassium observations with code K, final eGFR observations, ECG findings, symptoms, medications, active renal problems, and the potassium protocol.
- Ignore whole-blood potassium and preliminary potassium results for serum-potassium protocol decisions unless the template explicitly asks to list exclusions.
- Choose the latest final serum potassium by
effective_time.
- Replacement is needed when latest final serum potassium is below the protocol target and no urgent/contraindication branch applies.
- Urgent branch applies for potassium below the protocol urgent threshold, dialysis-dependent ESRD, severe renal contraindication, ECG abnormality, or arrhythmia-concern symptoms such as palpitations, syncope, or severe weakness.
- Routine oral dose follows the protocol rule: difference below target times the per-0.1 mmol/L dose increment, rounded to the nearest protocol dose increment.
- For routine replacement, use the protocol's oral potassium medication code and route/frequency/status values when the template asks for an order-ready recommendation.
- For no replacement, urgent escalation, or contraindication hold, set nullable medication and dose fields to
null where allowed and use the template's defer/not-recommended status.
- Schedule follow-up serum potassium for the next morning when routine replacement applies, using case/protocol timing if provided; otherwise use a conventional morning UTC timestamp after the review time.
- Contraindication booleans must come from active problems, case findings, and renal-function observations.
Care-Management Routing
Use care registry, active problems, final labs/vitals, active medication count, SDOH, member-call findings, and the care-management protocol.
- High risk generally requires predictive risk at or above the protocol threshold plus complex-care triggers such as at least three chronic conditions, recent admission, dialysis/advanced CKD, heart failure, or uncontrolled diabetes.
- Route to complex care when high-risk complex-care triggers are present; route to routine case management for lower-risk cases with needs; use not eligible only when the protocol criteria are absent.
- Map priority problems from active diagnoses, abnormal observations, registry fields, medication burden, and member-disclosed barriers. Do not include a code unless a source fact supports it.
- Numeric anchors should come directly from registry, final labs/vitals, or active medication counts. Format blood pressure as systolic/diastolic when both are available.
- Referrals: pharmacist for polypharmacy, insulin safety, or high-risk electrolyte/diuretic regimens; social worker for multiple moderate/severe social domains; dialysis care coordination for dialysis; transportation benefits for transportation barriers; behavioral health monitoring for positive behavioral health screens or protocol triggers.
- Outreach should be permission-based plain language when the member is reluctant, wants control over contact, or protocol requires permission. Use standard scripted outreach only when no preference/reluctance is documented.
- Care plan minima for complex cases should include multiple prioritized problems, weekly initial follow-up, member-stated priority, medication reconciliation, barrier work, escalation conditions, and at least two disciplines when the template asks numerically.
- Separate source provenance into chart-derived facts and member-disclosed facts. Transportation, financial/food/medication barriers, dialysis fatigue, and care-goal preference usually require member disclosure.
Observation-Window Tasks
Use case findings to identify the target code, target patient, inclusive window start, exclusive window end, and protocol gate.
- Matching observations must satisfy patient, code, final status, and time window.
- Excluded observations are relevant distractors from the case review that fail because of date, target code, or status. Include wrong-patient observations only if the template explicitly asks for them.
latest_final is the latest qualifying match when any exists; otherwise it is null if the template allows it.- For potassium windows, apply the potassium gate vocabulary from the template: recent final normal when the latest final is at or above target, low repletion needed when below target but not urgent, critical/urgent when below urgent threshold, and no final lab when no qualifying observation exists.
- For respiratory or imaging windows, use the same filter mechanics and map the downstream gate to the template/protocol vocabulary rather than inventing a potassium-specific result.
- Repeat-lab or repeat-test timing must come from the case/protocol/template. If a repeat is recommended and no exact time is provided, use the next reasonable clinic-morning timestamp after the review or window end.
Final Checks
- Validate every enum against the template before responding.
- Confirm each selected ID exists in the fetched case or protocol data.
- Confirm numeric precision: integer hours/days/counts, one-decimal lab values, two-decimal risk probabilities when requested.
- Confirm no unsupported normal findings, no contradicted allergy medication, and no use of preliminary/canceled/entered-in-error data for positive decisions.
