Skip to content
Get packs

Guide · Agent skill

How to use the GeoMaster geospatial analysis agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

GeoMaster is an agent skill: a folder of instructions and reference notes that an AI agent loads when your request involves geospatial work such as satellite imagery, GIS layers, coordinate systems, terrain or spatial statistics. This guide explains what the skill changes about the agent's behaviour, how to install it, how to phrase requests so it triggers, and what it needs and risks before you rely on its output. It is aimed at readers who have never used an agent skill before.

GeoMaster Geospatial AnalysisK-Dense-AI/scientific-agent-skills · MIT

Quick start

  1. 1Download the ZIP. Get the GeoMaster ZIP from the directory page. Keep the folder structure intact, including SKILL.md and the references folder.
  2. 2Place the skill folder. Unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ inside a project. Alternatively, upload it under Settings, Capabilities, Skills in the Claude apps.
  3. 3Prepare a Python environment. Create a dedicated environment with Python 3.12 or newer and install only the packages your workflow needs, such as GeoPandas and Rasterio.
  4. 4Make a plain request. Describe your task, the files involved and any known CRS or band order. The skill triggers when the request matches its description.
  5. 5Check the output. Ask for the CRS, transform, valid-data mask and provenance to be exported, and verify numeric results on a small known sample before scaling up.

What the GeoMaster skill makes the agent do

GeoMaster supports geospatial research workflows for remote sensing, vector and raster GIS, spatial statistics, terrain and network analysis, and machine learning for Earth observation. Its main job is to make the agent follow a careful workflow: record the input provenance, verify the coordinate reference system, align rasters, preserve nodata and cloud masks, and export the CRS, transform, valid-data mask and provenance with the results.

The skill also includes a small raster helper with recipes for NDVI, terrain metrics and a simple image classifier. The agent loads only the reference notes relevant to the task rather than the whole library.

  • Checks CRS from authoritative metadata instead of guessing it from coordinate bounds
  • Aligns extent, transform, dimensions and resolution before combining layers
  • Keeps invalid cells as NaN rather than turning them into zero
  • Writes out provenance so results can be checked later

Install the GeoMaster skill from the ZIP or the Claude apps

Download the GeoMaster ZIP from the directory page, then choose one of two routes. For a personal setup, unzip the folder into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. The skill folder should contain SKILL.md, the README.md and the references/ folder.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the skill there. Either way, the agent will only use the skill when your request matches its description.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside your project
  • Claude apps: Settings → Capabilities → Skills, then upload

Set up a Python environment for the examples

The core examples need Python 3.12 or newer and packages such as GeoPandas, Rasterio, NumPy and PyProj. The skill recommends a dedicated environment and installing only the packages your workflow needs. Cloud workflows also use pystac-client, planetary-computer and odc-stac.

Native GDAL or PDAL tools are installed separately through conda-forge, not with the pip-style commands. A Rasterio wheel bundles its own GDAL library but does not install the osgeo module or the GDAL command-line programs.

  • Use a dedicated environment for each project
  • Optional runtimes such as TorchGeo, RSGISLib, Py6S, ArcPy and QGIS are installed separately
  • Earth Engine and Copernicus services need your own credentials
uv venv --python 3.13\nuv pip install geopandas==1.2.0 rasterio==1.5.2 shapely==2.1.2 pyproj==3.8.0

Trigger the skill with natural requests

You do not need special syntax. Describe the geospatial task in plain language and name the data type or method. The skill's description lists the kinds of tasks it covers, including processing satellite imagery, aligning coordinate systems and raster grids, accessing STAC catalogs, analysing geospatial time series, and implementing GIS workflows in several languages.

Give the agent what it needs to check your work: the file names, the CRS if you know it, the band order and whether the data has been masked. Being specific helps the agent avoid assumptions it cannot verify.

  • Compute NDVI from my masked four-band Sentinel-2 stack and tell me which band is red and which is near-infrared
  • Check whether these two GeoPackage layers share a CRS before I join them
  • Find Sentinel-2 scenes over this bounding box from last June with under 20 percent cloud cover
  • Derive slope and aspect from this DEM, which is in metres on a projected grid

Example: NDVI from a verified band stack

The skill's NDVI helper assumes you have already verified the band order and masks. The example below is taken from the skill and assumes a four-band B02/B03/B04/B08 stack with clouds and shadows excluded and reflectance scaled by 0.0001. The helper does not identify bands for you, so a wrong band assumption can silently produce wrong NDVI.

Do not reuse these band numbers or scaling for an arbitrary file. Single-band products from Sentinel-2 SAFE packages or STAC assets are often separate rasters.

from raster_workflows import write_ndvi\nwrite_ndvi('s2_masked_stack.tif', 'ndvi.tif', red_band=3, nir_band=4, scale=0.0001, offset=0.0)

Safety notes: scripts and network access

GeoMaster runs scripts and uses the network. The raster helper runs local Python code, and the cloud examples query remote catalogs such as Planetary Computer. Planetary Computer signing returns expiring asset URLs, so save unsigned item IDs and metadata for provenance and sign them only when you need to access the data.

Authenticated services such as Earth Engine, Copernicus CDS and CDSE require your own registration and credentials. Review scripts before running them, and treat any credentials as yours to manage. The helper writes only to new output paths.

  • Review code the agent proposes before running it on sensitive data
  • Keep service credentials out of shared project files
  • Expect cloud examples to need network access and valid accounts

Tips and limits for reliable results

Treat the skill as a checklist-driven assistant, not as an authority on your data. Hillshade is an illumination visualisation, not a flow or exposure-risk measure. The bundled classifier is a small demonstration model and does not measure accuracy; use spatial or temporal holdouts before reporting predictive performance.

Be careful with overlapping zones in spatial joins, which can duplicate observations, and with area calculations, which need a suitable metric projection. Check the skill's review notes to see which examples were executed locally and which are illustrative templates.

  • Use the skill's review notes to see which recipes were executed and which are templates
  • Choose a documented policy for points on boundaries and overlapping polygons
  • Validate results with an expert before using them for flood, fire or exposure decisions

What people use it for

  • Computing NDVI from a stack whose band order and masks you have already verified
  • Checking CRS and alignment before joining vector layers or stacking rasters
  • Deriving slope, aspect and hillshade from a projected DEM in metres
  • Finding and filtering STAC scenes by area, date and cloud cover as a starting point for analysis
  • Teaching Python geospatial workflows with GeoPandas, Rasterio and STAC catalogs

Frequently asked questions

Is the GeoMaster skill free to use?

The skill is released under the MIT License, as stated in its metadata. Some optional workflows need Earth Engine, Copernicus or other commercial services, which require your own accounts and may have their own terms.

What do I need to install GeoMaster?

You need the skill folder, unzipped into ~/.claude/skills/ or a project's .claude/skills/ folder, or uploaded in the Claude apps. The core examples need Python 3.12 or newer with GeoPandas, Rasterio, NumPy and PyProj.

Does GeoMaster run code on my computer?

Yes. The skill includes bundled Python helpers and the workflows can run scripts and make network requests, so review code before running it on sensitive data.

Can GeoMaster identify the bands in my satellite image automatically?

No. The helper does not infer band identities or masks. If you assume the wrong bands, NDVI can be wrong without any error, so verify band order from the product metadata first.

How does GeoMaster compare with running GeoPandas or Rasterio directly?

GeoMaster adds a checklist-driven workflow and reference notes for the agent to follow, such as CRS verification and provenance export. The underlying libraries still do the calculations, so you still need to understand the methods to judge the results.

Why does the classifier not report accuracy?

The bundled classifier is a small demonstration model and the skill states that it does not measure accuracy. Use spatial or temporal holdout validation before reporting predictive performance.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use GeoMaster if you are a GIS or remote sensing analyst, a student learning Python geospatial workflows, or a researcher who needs documented provenance and CRS checks. Pick something else if you need operational flood, fire or exposure risk decisions, or classifier accuracy claims without independent holdout validation.

Download the skill