ワンクリックで
geo-onboarding
Turn farmer-provided map clues into draft farm, pasture, paddock, and section GeoJSON.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Turn farmer-provided map clues into draft farm, pasture, paddock, and section GeoJSON.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Run the constrained first-run onboarding workflow for one farm.
Set up a farm data pipeline for an external service using Firecrawl browser profiles, farmer-assisted login, site exploration, and generated vendor skills. Use when a farmer asks to connect a service like NoFence or to create a recurring data pipeline.
Turn trusted sources into structured ancestral knowledge with Firecrawl and openPasture tools.
Generate the daily farm summary and recommendation.
Assess pasture condition from notes, photos, and imagery.
Reason through near-term livestock movement decisions.
| name | geo-onboarding |
| description | Turn farmer-provided map clues into draft farm, pasture, paddock, and section GeoJSON. |
| version | 1.0.0 |
Use this skill when a farmer provides a map screenshot, dropped pin, survey report, address, parcel clue, or hand-drawn boundary and wants OpenPasture to show their farm on a GIS map.
OpenPasture is the toolkit and storage layer. The agent is responsible for reading the image or document, extracting the visible clues, and calling OpenPasture tools with structured geospatial data.
[longitude, latitude].pasture land units for the visible fields or pastures inside the farm.paddock or section land units only when the farmer or screenshot clearly indicates subdivisions.provenance.source to map_screenshot, survey, address, farmer_description, or the best matching source.1.0 for agent-generated geometry. For screenshot-only boundaries, prefer low-to-moderate confidence.warnings; do not claim legal, survey-grade, or parcel-accurate boundaries.save_geo_onboarding_draft for a full draft or upsert_land_unit for one follow-up edit.For a Google Maps screenshot with a dropped pin and loose boxes:
location.Prefer this shape for save_geo_onboarding_draft:
{
"name": "Duck River Farm",
"timezone": "America/Chicago",
"location": { "type": "Point", "coordinates": [-87.038675, 35.642109] },
"boundary": {
"type": "Polygon",
"coordinates": [[
[-87.043, 35.646],
[-87.034, 35.646],
[-87.034, 35.639],
[-87.043, 35.639],
[-87.043, 35.646]
]]
},
"source": "map_screenshot",
"confidence": 0.55,
"evidence": ["Google Maps screenshot with dropped pin and farmer-drawn boxes"],
"pastures": [
{
"id": "pasture_west",
"name": "West boxed pasture",
"geometry": { "type": "Polygon", "coordinates": [[[-87.042, 35.646], [-87.038, 35.646], [-87.038, 35.640], [-87.042, 35.640], [-87.042, 35.646]]] },
"confidence": 0.5,
"warnings": ["Estimated from screenshot box; farmer should confirm corners."]
}
]
}
The farmer receives a map link with the farm and pastures rendered as editable drafts. The stored geometry is good enough for visual confirmation, future NDVI/pasture analysis, and follow-up editing, but it remains clearly marked as approximate until confirmed.