Skip to content

SKILL.md

Latest commit

 

History

History
136 lines (102 loc) · 6.26 KB

SKILL.md

File metadata and controls

136 lines (102 loc) · 6.26 KB
name residence-risk
description Assess flood and earthquake risk for a Taiwanese residential address. Use when the user asks about flood/淹水/淹水風險, earthquake/地震/斷層/液化/土壤液化 risk for a specific Taiwan address (e.g. "台北市信義區XX路" or "幫我看這個地址會不會淹水"). Calls the public Residence Risk TW API and returns a 0-100 score per hazard. Do NOT use for non-Taiwan addresses, legal/real-estate/insurance decisions, or generic weather/earthquake forecasts.

residence-risk skill

What this skill does

Calls the Residence Risk TW public API (POST /v1/assess) to evaluate flood and earthquake risk for any Taiwan address. Returns structured scores (0–100), five-level labels (極低 / 低 / 中 / 高 / 極高), scenario breakdowns, and a geocoded {lat, lng}.

Data sources (Taiwan government open data):

  • 經濟部水利署 — 淹水潛勢圖（24h 350/500/650 mm 三種降雨情境）
  • 中央地質調查所 — 活動斷層地質敏感區
  • 經濟部地質調查及礦業管理中心 — 土壤液化潛勢圖

When to use

Use when the user asks about any of:

  • 「這個地址會不會淹水？」/ flood risk for a Taiwan address
  • 「這裡地震風險如何？」/「有沒有斷層？」/「會不會液化？」
  • Comparing two Taiwan addresses by disaster risk
  • 搬家 / 租屋 / 買房前的防災參考（搭配強調僅供參考）

When NOT to use

  • 非台灣地址 → 回傳 ADDRESS_NOT_FOUND，請改用其他工具
  • 法律、保險核保、金融授信、不動產交易決策 → 本 API 明確不適用
  • 即時天氣 / 地震預報 → 本工具只看歷史潛勢圖資

How to call

REST (recommended)

curl -sS https://residence-risk-api.workers.dev/v1/assess \
  -H 'Content-Type: application/json' \
  -d '{"address":"台北市信義區信義路五段7號"}'

Full spec: https://residence-risk-api.workers.dev/v1/openapi.json

MCP (for Claude Desktop, Cursor etc.)

The API exposes a streamable-HTTP MCP server at POST /mcp with one tool:

  • assess_residence_risk({ address: string })

Client config example:

{
  "mcpServers": {
    "residence-risk": {
      "url": "https://residence-risk-api.workers.dev/mcp"
    }
  }
}

Response shape (abridged)

{
  "address": "台北市信義區信義路五段7號",
  "location": { "lat": 25.033, "lng": 121.567, "source": "nominatim", "display_name": "..." },
  "flood": {
    "score": 95, "level": "極高", "color": "#ef4444",
    "risks": [...],
    "reasoning": ["綜合判定：**極高**風險...", "此地點在 24h 350mm 即會淹水..."],
    "disclaimer": "..."
  },
  "earthquake": {
    "score": 86, "level": "極高", "color": "#ef4444",
    "fault":        { "score": 90, "risks": [...] },
    "liquefaction": { "score": 80, "has_data": true, "risks": [...] },
    "history": {
      "available": true, "radius_km": 50, "years_back": 10,
      "events": [{
        "earthquake_no": "2022137",
        "origin_time": "2022-09-18T06:44:15Z",
        "magnitude": 6.8, "depth_km": 7.0,
        "epicenter_distance_km": 42.3,
        "location_description": "台東縣政府北方44.7公里",
        "estimated_intensity": {
          "level": "6強", "method": "nearest_station",
          "nearest_station": { "name": "池上", "distance_km": 3.2, "pga_gal": 420.1 }
        }
      }]
    },
    "reasoning": ["綜合判定：**極高**風險...", "**斷層：位於第一類活動斷層...**", "**液化：...**", "**歷史地震佐證：...**"],
    "disclaimer": "..."
  },
  "meta": { "response_ms": 1351, "api_version": "0.2.0" },
  "disclaimer": "..."
}

Key fields for agents:

  • flood.reasoning / earthquake.reasoning are arrays of natural-language explanations (contain **bold** markdown). Always present these to the user — they are the "why" behind the score.
  • earthquake.history.events[].estimated_intensity uses Method A: nearest-station measured intensity (CWB station within 15 km of the address). null when no station is close enough — don't extrapolate.
  • earthquake.history.available === false means the administrator has not yet imported CWB historical data; respond with "此功能暫不可用" or similar.

Scoring reference

Flood (max across scenarios): 24h 350mm 淹水 > 50cm → 95; 0-50cm → 80; 24h 500mm > 50cm → 65; 0-50cm → 50; only at 650mm → 30; within 100m → 20; none → 5.

Earthquake: fault = max(inside class1 → 90, inside class2 → 70, <200m class1 → 55, <200m class2 → 45, <500m class1 → 35, <500m class2 → 25, else → 10, no data → 5); liquefaction = 高/中/低 → 80/50/20, no data → 5; combined = 0.6*fault + 0.4*liq (fault-only when liquefaction data unavailable for that county).

Levels: 81–100 極高 / 61–80 高 / 41–60 中 / 21–40 低 / 0–20 極低.

Agent usage guidelines

  1. Always include the disclaimer from the response in your final answer.
  2. Rate limit: do not exceed ~1 req/s per address — geocoding upstream (Nominatim) will throttle.
  3. Language: the API expects 繁體中文 addresses. Translate English addresses to Chinese before calling (縣市 / 區 / 路 / 號).
  4. Errors: on ADDRESS_NOT_FOUND, ask the user to clarify with 縣市 + 區; on INVALID_ADDRESS, shorten to ≤ 200 chars.
  5. Presentation: prefer the level label (極低 / 低 / 中 / 高 / 極高) and include the numeric score; show scenario breakdowns on request.
  6. Batching: no batch endpoint — loop sequentially for multiple addresses.

Example prompt → call

User: "幫我看台中市西區臺灣大道二段 2 號的淹水和地震風險"

Call: POST /v1/assess with {"address":"台中市西區臺灣大道二段2號"}, then summarize:

此地址（台中市西區臺灣大道二段2號）的淹水風險為「中」(50/100)，在 24h 500mm 降雨情境下會有 0–50cm 淹水；地震風險為「低」(25/100)，距離第二類活動斷層約 430m，尚無土壤液化資料。本結果僅供防災參考，不構成不動產或保險決策依據。

Resources