| name | briefing |
| description | Generate a structured pre-consultation patient briefing. Use when the physician asks for a briefing, a pre-consultation summary, or sends /briefing. |
Pre-consultation briefing
Generate a structured briefing for the patient in this conversation, grounded
in retrieved clinical evidence.
Workflow
- Review the patient record already present in this conversation
(demographics, conditions, medications, labs with reference ranges,
allergies, visits).
- Call
mcp__guidelines__search_clinical_guidelines for each major
condition, and for drug interactions when the patient takes 2+ medications.
Use specific clinical queries (e.g. "metformin renal dosing eGFR 45"),
not vague ones (e.g. "diabetes"). If any guidelines search returns nothing
relevant — for a condition, medication interaction, or screening — and the
web-research skill is available, you may search the web for authoritative
guidance and cite the source URL in the flag/action description — local
guidelines remain the primary source.
- Build the briefing:
- flags — category one of
labs (values outside reference ranges),
medications (interaction/dosing concerns), screenings (overdue
preventive screenings for age/gender/conditions), ai_insight (clinical
patterns across the data). Severity: critical = immediate concern,
warning = needs attention this visit, info = worth noting. source
is always "ai". Include suggested_action when a concrete next step
exists.
- summary —
one_liner (single sentence capturing the clinical
picture), key_conditions (active conditions from the record),
relevant_history (brief clinically relevant context).
- suggested_actions — 3-5 actions prioritized by clinical urgency
(priority 1 = most urgent), each with a brief reason.
- Cite a
[source_id] from search results for every clinical claim. If no
relevant guidelines were found, state that explicitly — never invent
source backing.
- REQUIRED FINAL STEP: call
mcp__publisher__publish_briefing exactly once
with the complete briefing. Do NOT paste the briefing JSON into your chat
reply — the dashboard renders it from the publish call.
- After the tool confirms, reply with one or two sentences summarizing the
most important finding so the physician sees it in the chat thread.
Constraints
- Only flag issues visible in the provided data.
- Be concise: physicians need quick, scannable information.