AGENT SKILL

lookup-patient-history

Load the synthetic patient chart and verbatim prior reports for EHR reporting. Use before measuring the selected current and prior images; this lookup does not perform segmentation or measurements.

License 许可
MIT
最近更新
2026-10-30
Stars
★ 0
Forks
0

01 前言 Front Matter

name
lookup-patient-history
description
Load the synthetic patient chart and verbatim prior reports for EHR reporting. Use before measuring the selected current and prior images; this lookup does not perform segmentation or measurements.
argument-hint
- image_path: required path to the current image, named `{patient_id}_{study_index}.nii.gz`. - patient_id: optional override if the filename does not follow that convention.

03 SKILL.md 正文

Lookup Patient History

Read data/patients/{patient_id}.json relative to the workspace root and inline its referenced prior reports. Return data, not a clinical interpretation.

Scope and handoff

  • File reads and file searches are permitted. Do not call segmentation or measurement tools during this lookup step. This restriction ends when the lookup returns; the parent report workflow must then perform its measurement steps.
  • When used by a report skill, return the record internally and resume the parent workflow. Do not treat this helper's completion as the final chat response.
  • When explicitly requested on its own, return only the patient-record block.
  • Keep the block concise, but preserve all study dates, paths, treatment and procedure details, LAST RECORD ENTRY NOTE, and full prior report text. Do not truncate these to meet a word budget.

How to use

  1. Resolve the identifiers.
    • Use a supplied patient_id; otherwise infer it by removing .nii.gz and the trailing _NNN from the filename, for example SYN0056_006.nii.gz gives SYN0056.
    • Retain the current study_id to identify the matching pending order. If the filename does not identify the patient or study and the caller has not supplied it, request clarification rather than guessing.
  2. Read the chart.
    • Read data/patients/{patient_id}.json.
    • If the file does not exist, return PATIENT RECORD: Not on file for patient_id {patient_id}. to the parent, which must still measure the current image. This ends only the lookup.
    • If the file exists but cannot be read or parsed, report the specific error and stop the workflow; do not mislabel it as a missing chart.
  3. Inline prior reports.
    • For each non-null report_path in imaging_studies, read the report and include its full text verbatim.
    • Resolve paths from the workspace root. If a report is not there, look for the same filename in the neighboring reports directory; do not substitute a different study.
    • If a report remains missing, include [report file missing: {path}] and continue with the other chart data. The parent must disclose that missing source.
  4. Return the block below, preserving the section headers. Mark the pending order that matches the current study.

Output format

PATIENT RECORD - {patient_id}, as of {record_as_of}

DEMOGRAPHICS: {demographics.age}{demographics.sex}.

DIAGNOSIS: {diagnosis.primary} {diagnosis.molecular} Established {diagnosis.established}

PRESENTATION: {presentation.date}. {presentation.presenting_complaint}
HPI: {presentation.hpi_summary}

ACTIVE PROBLEMS:
- {each item from active_problems}

PAST MEDICAL HISTORY:
- {each item from past_medical_history}

ALLERGIES: {allergies.notes}

CURRENT MEDICATIONS:
- {name} {dose} - {indication}
  (one line per item from medications)

PROCEDURES:
- {date}: {procedure}. {result} Complications: {complications}
  (one line per item from procedures)

TREATMENTS:
- {start} -> {end, or "ongoing" if null}: {treatment}. {notes}
  (one line per item from treatments)

EXPECTED POST-TREATMENT FINDINGS (per record):
- {each item from expected_post_treatment_findings}

IMAGING HISTORY (oldest to newest):
- {study_id} - {date}, {exam}. {summary} Image: {image_path, or "not retained"}
  (one line per imaging_studies entry)

PRIOR REPORTS:
=== {study_id} ({date}) ===
{full report text loaded from report_path, or the missing-file notice}
===
  (one block per imaging_studies entry with a report_path)

PENDING ORDERS:
- {study_id} - {date}, {exam}. Indication: {indication} Status: {status}
  (one line per pending_orders entry; mark the matching study with "<- current image")

LAST RECORD ENTRY NOTE: {last_record_entry_note}

Critical rules

  • Do not invent fields. Render missing, empty, or null fields as (none) unless the format specifies otherwise.
  • Do not summarize the prior report or substitute its measurements for a current image measurement.
  • Do not interpret findings or decide which lesion is new. The parent report skill combines the record with imaging evidence.
  • Always include LAST RECORD ENTRY NOTE so the report writer knows where the record ends.

04 同仓库更多 Skills

microsoft
microsoft/LAB730-ignite-2026-healthcare-agents

write-report-with-ehr

Write a context-aware brain MRI follow-up report from the patient's record, tool-computed lesion measurements, and change from the most recent prior image. Use when asked for a report that uses the patient's history, chart, or EHR.

行政与办公人员图像与多媒体
★ 0
microsoft
microsoft/LAB730-ignite-2026-healthcare-agents

lookup-patient-history

Load a synthetic patient EHR record (demographics, diagnosis, treatments, medications, expected post-treatment findings, prior imaging studies and reports, pending orders) for a given brain MRI image. Invoke as the first step of any context-aware reporting flow that needs patient history. No model calls.

图像与多媒体
★ 0
microsoft
microsoft/LAB730-ignite-2026-healthcare-agents

write-report-with-ehr

Write a context-aware brain MRI report using the patient chart and mandatory tool-computed linear and volumetric measurements. Use for EHR reporting with Radiology Assistant (EHR).

行政与办公人员
★ 0
microsoft
microsoft/LAB730-ignite-2026-healthcare-agents

write-imaging-factsheet-ehr

Produce an EHR imaging factsheet with mandatory segmentation, cleanup, linear dimensions, and volumes. Use when write-report-with-ehr requests imaging evidence or when explicitly asked for this factsheet.

★ 0
microsoft
microsoft/LAB730-ignite-2026-healthcare-agents

write-report-measured

Write a brain MRI report with tool-computed lesion measurements and, when a prior image is supplied, change from that prior. Use for brain MRI reports when the medimageparse `segment`, `clean_segmentation`, and `describe_lesions` tools are available.

行政与办公人员图像与多媒体
★ 0
microsoft
microsoft/LAB730-ignite-2026-healthcare-agents

write-imaging-factsheet

Generate a lesion-only imaging factsheet from a post-contrast T1 MRI, optionally comparing a prior image. Use when asked for a lesion factsheet or a descriptor-based imaging summary.

图像与多媒体
★ 0