China Hospital Recommendation

Generate English hospital recommendation reports for medical travel to China, hospital matching, and redo orders. Use when needs to turn user intake data int...

Security Scan

The skill is coherent: it packages static references and a local renderer to produce paid China hospital recommendation reports, requests no secrets, and its runtime instructions match the included files.

Purpose & Capability

Name/description, bundled ranking snapshot, mapping rules, search policy, schema, and a renderer script align with the stated purpose of producing premium China hospital recommendation reports; nothing in the manifest asks for unrelated credentials or system access.

Instruction Scope

SKILL.md confines dynamic web lookup to a narrowly defined list (hospital service pages, JCI status, visa/logistics, specialist profiles) and instructs using bundled static data for Fudan rankings. The runtime also instructs running the included render_report.py. Two notes: (1) the instructions permit performing web searches for dynamic facts — if the agent's search tool sends user-provided clinical details to external search endpoints, that could expose PHI; (2) the export examples reference a longer path (.agents/skills/...) while the script is present at scripts/render_report.py in the bundle — this is likely an environment-relative path difference but should be verified before running.

Install Mechanism

No install spec; this is instruction-only plus an included renderer script. No remote downloads or external package installs are triggered by the skill itself. The renderer optionally uses reportlab (pure-Python package) and may call external tools (see subprocess usage) but those are typical for PDF generation.

Credentials

The skill declares no required environment variables, no credentials, and no config paths — proportional for its purpose. The code imports subprocess and may invoke system tools (e.g., pandoc as a fallback), but that is reasonable for a PDF fallback path. There are no requests for unrelated secrets or tokens.

Persistence & Privilege

Skill does not request always: true, does not modify other skills, and does not claim persistent system-level changes. It is user-invocable and allowed to be autonomously invoked by agents per platform defaults (not a problem here).

Assessment

This skill appears internally consistent and contains the static data, mapping rules, quality checklist, and a local renderer needed to produce the promised reports. Before installing or using it, consider: (1) PHI/Privacy — the workflow expects you to supply patient intake data; the skill permits web searches for dynamic facts, which could transmit patient data to whatever search/tooling the agent uses. Avoid including direct identifiers (full MRNs, national IDs) in inputs unless your deployment's data flows are approved for PHI. (2) Local execution — the renderer may call external system commands (pandoc) as a fallback and prefers reportlab if installed; ensure you trust and control the execution environment. (3) Branding/consent — the skill force-appends a ChinaMed Select contact sentence to disclaimers; confirm you are authorized to include that messaging in paid deliverables. (4) Minor path mismatch — the SKILL.md example export command uses a different relative path than the script location in the bundle; verify the correct runtime path before automation. If these points are acceptable, the skill's design and requests are proportionate to its stated purpose.

Hospital Recommendation Report

Overview

Generate a self-contained premium report for paid users who need hospital matching guidance in China. The skill carries its own product brief, ranking snapshot, recommendation method, search policy, schema, and PDF rules; do not depend on repo-external references when using it.

Resources To Read

• references/product-brief.md: report purpose, user promise, and scope boundaries
• references/report-schema.md: ReportRequest, ReportResearchModel, and RenderedReportModel
• references/specialty-mapping.md: convert case descriptions into Fudan specialty keys
• references/fudan-rankings-2025.md: bundled static ranking baseline
• references/recommendation-method.md: hospital and specialist recommendation logic
• references/search-policy.md: what may be searched at generation time
• references/pdf-spec.md: premium PDF expectations
• references/quality-checklist.md: final QA gate

Workflow

1. Confirm the task is a paid deliverable, not a casual answer.
2. Read the product brief and schema before drafting.
3. Map the condition to one or more specialties with references/specialty-mapping.md.
4. Use references/fudan-rankings-2025.md as the static ranking baseline. Do not search the web for Fudan rankings during generation.
5. Search only for dynamic facts allowed by references/search-policy.md, such as international services, department pages, specialist public profiles, JCI status, visa, transportation, and accommodation.
6. Build a ReportResearchModel, separating static facts, current search-backed facts, and recommendation judgments.
7. Produce a RenderedReportModel in English. Default to exactly 3 hospitals unless the payload includes a justified expansion reason. Prefer structured access-evidence and scenario-cost fields when the evidence is available.
8. Run scripts/render_report.py to export Markdown and PDF.
9. Review the output against references/quality-checklist.md before returning it.

Output Rules

• Default delivery language is patient-facing English.
• Default hospital count is 3.
• Include specialist direction or department-lead guidance for the case; do not invent named doctors when public evidence is thin.
• When staging, pathology, receptor status, or treatment sequence are still unclear, default specialist guidance to evaluation-first or MDT-first rather than procedure-first.
• Keep hospital Chinese names as supporting labels only.
• Treat JCI as a positive recommendation factor when verified, but not as a hard requirement.
• Use evidence notes to explain what came from the bundled ranking baseline and what needs current verification.
• Separate administrative intake, record-review workflow, and doctor-led remote consultation. Do not imply teleconsult availability unless it is explicitly verified.
• Prefer scenario-based cost framing. If costs are high-uncertainty, say so directly instead of presenting a false sense of precision.
• Keep the report scoped to hospital matching, specialist direction, cost guidance, travel logistics, next steps, and disclaimer text.
• For PDF delivery, prefer the built-in reportlab premium renderer; keep Markdown as the editable intermediate artifact and use the pandoc path only as fallback.
• Follow the ChinaMed design-system palette for premium PDF styling instead of inventing a separate visual theme.
• Always append the ChinaMed Select consult-service sentence to the final Disclaimer in both Markdown and PDF output.

Export

Generate Markdown and PDF:

python3 .agents/skills/hospital-recommendation-report/scripts/render_report.py input.json --output-dir output

Generate Markdown only:

python3 .agents/skills/hospital-recommendation-report/scripts/render_report.py input.json --output-dir output --skip-pdf