기상청 날씨 (KMA Weather Korea)

---

name: kma-weather description: KMA short-term forecast API - ultra-short-term observation/forecast, short-term forecast version: 2.2.0 author: chumjibot created: 2026-02-10 updated: 2026-02-19 tags: [weather, korea, openapi, data.go.kr] connectors: [~~weather, ~~air, ~~notify]

---

KMA Weather Forecast Skill

Korea Meteorological Administration short-term forecast API + AirKorea air quality integration.

Overview

| Key | Value | |-----|-------| | Provider | KMA (Korea Meteorological Administration) | | Service ID | 15084084 | | Auth | ~/.config/data-go-kr/api_key | | Endpoint | https://apis.data.go.kr/1360000/VilageFcstInfoService_2.0 | | Playbook | playbook.md |

Scripts

scripts/
├── weather.sh             → Weather query (ncst/fcst/short)
├── morning_briefing.sh    → Morning briefing (weather + air quality)
└── grid_convert.py        → Lat/lon → grid coordinate conversion

Workflow

Step 1: Identify location & time

"서울 날씨" → Seoul Jongno-gu (nx=60, ny=127)

"내일 날씨" → short-term forecast (short)

"지금 비 와?" → ultra-short-term observation (ncst)

Refer to playbook.md for default location

Step 2: Fetch weather data

Current: weather.sh ncst [nx] [ny]

6-hour: weather.sh fcst [nx] [ny]

3-day: weather.sh short [nx] [ny]

Step 3: Air quality integration (~~air)

Fetch PM data via AirKorea skill

skills/airkorea-air-quality/scripts/air_quality.sh [station]

Step 4: Structured response using output template

Output Template

## 🌤️ [Location] Weather
Current
🌡️ Temp: X°C (Feels like X°C)
💧 Humidity: X% | 🌬️ Wind: Xm/s
🌧️ Precip: [None/Rain/Snow]
Today's Forecast
☀️ High X°C / Low X°C
🌧️ Precip probability: X%
🌫️ PM2.5: [Good/Moderate/Bad/Very Bad] (X㎍/㎥)
💡 One-liner
[Bring umbrella / Wear mask / Great day to go out]

API Types

| API | Command | Description | Update Cycle | |-----|---------|-------------|-------------| | Ultra-short obs | ncst | Current observation | Hourly | | Ultra-short fcst | fcst | 6hr forecast | Every 30min | | Short-term fcst | short | 3-day hourly | 8x/day |

Key Grid Coordinates

| Location | nx | ny | |----------|----|----| | Seoul Jongno | 60 | 127 | | Seoul Gangnam | 61 | 126 | | Busan | 98 | 76 | | Daegu | 89 | 90 | | Incheon | 55 | 124 | | Daejeon | 67 | 100 | | Jeju | 52 | 38 |

Codes

Sky (SKY)

1: Clear ☀️ | 3: Mostly cloudy ⛅ | 4: Overcast ☁️

Precipitation (PTY)

0: None | 1: Rain 🌧️ | 2: Rain/Snow 🌧️❄️ | 3: Snow ❄️ | 4: Shower 🌦️

base_time Rules

| API | Release time | Queryable | |-----|-------------|-----------| | Ultra-short obs | Every hour | +10min | | Ultra-short fcst | Every 30min | +10min | | Short-term fcst | 02,05,08,11,14,17,20,23h | +10min |

Connectors

| Placeholder | Purpose | Current Tool | |-------------|---------|-------------| | ~~weather | Weather API | KMA Short-term Forecast | | ~~air | Air quality | AirKorea | | ~~notify | Notification | Telegram |

Intent Router

| Intent | Trigger Examples | Strategy | |--------|-----------------|----------| | Quick Check | "오늘 날씨 어때?", "지금 비 와?", "기온 몇 도?" | ncst (현재 관측) + fcst (3시간 예보) → 간단 요약 | | Daily Forecast | "내일 비 와?", "주말 날씨", "모레 눈 올까?" | short (단기예보 3일) → 시간대별 상세 | | Morning Briefing | "아침 날씨 브리핑" | morning_briefing.sh → 날씨+대기질 통합 |

Routing logic: 현재/지금 → Quick Check, 내일/모레/주말 → Daily Forecast, 브리핑 → Morning Briefing

Cross-Skill Integration

| Trigger | Target Skill | Integration | |---------|-------------|-------------| | 날씨 응답 시 항상 | airkorea-air-quality | PM2.5 수치를 날씨 응답에 포함 (Output Template 참조) | | 미세먼지 나쁨 이상 | airkorea-air-quality | "마스크 챙기세요" 팁 추가 | | 비/눈 예보 시 | — | "우산 챙기세요" 자동 팁 |

Notes

1. Uses grid coordinate system (not lat/lon) → use grid_convert.py 2. Query after release time + 10 minutes 3. Short-term forecast returns large data → set sufficient numOfRows

---

*Cowork architecture v2.2 — 🦞 chumjibot (2026-02-19)*

🔧 Setup (공공데이터 포털 API)

1. data.go.kr 회원가입 2. 로그인 → 마이페이지 → 일반 인증키(Decoding) 복사 3. API 키 저장:

   mkdir -p ~/.config/data-go-kr
   echo "YOUR_API_KEY" > ~/.config/data-go-kr/api_key
   

4. 아래 서비스 활용신청 후 사용 (자동승인) - 기상청 단기예보 (15084084)

1. Uses grid coordinate system (not lat/lon) → use grid_convert.py 2. Query after release time + 10 minutes 3. Short-term forecast returns large data → set sufficient numOfRows

---

*Cowork architecture v2.2 — 🦞 chumjibot (2026-02-19)*