EUMETSAT Agent Skills

This repository gives your coding agent seven skills for working with EUMETSAT tools. A skill is a SKILL.md file in the Agent Skills format. It holds the commands and the safety rules for one tool or service. Claude Code, Codex, OpenCode, VS Code, and any other client of the Agent Skills format can load the skills.

With the skills loaded, your agent can search and download satellite data from the EUMETSAT Data Store, run Data Tailor customisations, manage profiles, infrastructure and catalogues on the European Weather Cloud, build Firecube plugins and ingest products into datacubes, select satellite imagery around detected weather features from the Earth System Feature Database, and find the right climate data record, files and variables on the Data Store.

Installing the skills needs no account. The table below lists what each skill needs when you run its workflows.

Available Skills

Skill	What it does	What you need
eumdac	Works with the EUMETSAT Data Store and Data Tailor. Searches, lists, downloads, and orders satellite products. Checks tokens and credentials. Manages subscriptions and livefeed.	EUMDAC 3.1.1 as CLI or Python package. Data Store credentials in ~/.eumdac. Network access to api.eumetsat.int.
ewccli	Sets up the European Weather Cloud command line interface. Covers login profiles, OpenStack application credentials, Community Hub item discovery and deployment, infrastructure commands, and custom catalogues.	EWCCLI 0.3.1 on Python 3.11+. A European Weather Cloud account. For live commands, network access to the EWC and OpenStack endpoints.
ewc-community-hub	Draft, validate and submit items to the EWC Community Hub catalogue (items.yaml metadata, local schema validation, pull request).	Docker for local validation; GitHub account and EWC access for the PR.
firecube	Runs Firecube batch ingestion, which turns Earth Observation products into analysis-ready datacubes. Covers plugin development.	Firecube 0.1.7 on Python 3.12+. For execution, an appropriate plugin and source files.
esf-satellite-data	Builds reproducible satellite imagery datasets around detected weather features from the Earth System Feature database. The imagery comes from the EUMETSAT Data Store, EUMETSAT S3 buckets, or AWS Open Data.	The internal esf MCP server 0.9.0 (feature-db-api), registered with your agent. Data Store or S3 access for the imagery. Access permissions, endpoint and credentials for the esf server come from your service administrator.
feature-db-client	Fetches and saves Earth System Feature Database features as GeoDataFrames with the feature_client Python library. Runs spatial, temporal, and tag queries and nearest-feature search.	The feature-client package 0.1.0, a public package that needs no entitlement to install. Access permissions for the Feature Database service: an authorised endpoint and API key from your service administrator.
ml-menu	Works with the EUMETSAT Data Store and eumdac to find Climate Data Records, specifically. The skill is designed to improve data discoverability for long time series for training of ML models, trend analysis, anomaly detection.	Python and xarray. eumdac with Data Store credentials, as the eumdac skill needs.

Getting Started

Prerequisites

The skills do not install runtime tools or credentials. Prepare the following:

• An agent that reads SKILL.md files. The one-command installs and the prompts that ask for an explanation need no EUMETSAT runtime tool and no service credentials.
• A local checkout of this repository, only for the manual install and the agent-prompt install.
• The runtime tool and the credentials of a skill, only to run that skill's tool or service operations. The table lists them.

Installing

The repository is a plugin marketplace (.claude-plugin/marketplace.json) and a plain skills folder. Choose the install path for your agent. Each path installs all seven skills.

Claude Code

/plugin marketplace add eumetsat/skills
/plugin install eumetsat-skills@eumetsat
/reload-plugins

OpenAI Codex

codex plugin marketplace add eumetsat/skills
codex plugin add eumetsat-skills@eumetsat

OpenCode, VS Code and other skills-compatible agents

With the skills CLI:

npx skills add eumetsat/skills -g -a opencode --skill '*' -y   # all projects
npx skills add eumetsat/skills -a opencode --skill '*' -y      # this project only

Replace -a opencode with your agent. To target several agents, pass -a once for each, for example -a claude-code -a codex -a opencode. To install one skill, pass --skill eumdac. To install by hand, run:

git clone https://github.com/eumetsat/skills.git eumetsat-skills
mkdir -p ~/.agents/skills
ln -s "$PWD"/eumetsat-skills/skills/* ~/.agents/skills/

Use the skills directory that your client documents. https://agentskills.io/clients and installation list the path for each agent. For a prompt-based install, open a checkout and give your agent this prompt:

Install every skill under skills/ from this checkout for my current agent using docs/installation.md.
Keep any existing skill installation unless I authorize replacing it.
Verify that each installed skill resolves to skills/<name>/SKILL.md.

Quickstart

After the install, give your agent this prompt:

Use the eumdac skill to explain how I can search an EUMETSAT Data Store collection.
Do not install tools, configure credentials, or download data yet.

The agent names the collection and the search filters it needs. Then it explains the search procedure. The prompt asks for an explanation only, so this step needs no installed tools and no credentials. See more prompts and worked examples.

Development

Keep skill sources in skills/<name>/SKILL.md. Read AGENTS.md for the contribution rules. Read reference boundaries before you move guidance.

Validation covers tool access, CLI checks, and behavior cases. External references lists the official product and agent documentation.

Software Bill Of Materials

Generate the source inventory in SPDX JSON format at .reports/source-sbom.spdx.json:

mkdir -p .reports
trivy fs --include-dev-deps --skip-dirs .venv,.venvs,venv,env,.reports \
  --format spdx-json --output .reports/source-sbom.spdx.json .

The source scan does not resolve standalone Python imports. The tables below record the versions and licenses reviewed for the Firecube helpers.

Included Components

No third-party components are vendored.

Direct Runtime Dependencies

These dependencies are needed when running the corresponding helper scripts.

dependency	version	license	copyright	home_url	comments
firecube	0.1.7	Apache-2.0		https://github.com/eumetsat/firecube	Ingestion checks; optional API inspection in the contract checker.
numpy	2.5.2	BSD-3-Clause		https://numpy.org	Product inspection and store comparison.
xarray	2026.7.0	Apache-2.0		https://xarray.dev	Product inspection and store comparison.
zarr	3.3.0	MIT		https://github.com/zarr-developers/zarr-python	Zarr store checks.
netCDF4	1.7.4	MIT		https://github.com/Unidata/netcdf4-python	NetCDF backend; alternative to h5netcdf for supported inputs.
h5netcdf	1.8.1	BSD-3-Clause		https://h5netcdf.org	Alternative backend for NetCDF4/HDF5 inputs.
h5py	3.16.0	BSD-3-Clause		https://www.h5py.org	Fallback group walk in product inspection and store comparison when netCDF4 is absent.

Direct Build, Edit, And Test Dependencies

dependency	version	sw type	license	copyright	home_url	comments
uv	0.12.13	Development tools	MIT OR Apache-2.0		https://github.com/astral-sh/uv	Runs plugin tests and lint checks from check_done.py.
pytest	9.0.3	Development tools	MIT		https://github.com/pytest-dev/pytest	Plugin tests invoked by check_done.py.
ruff	0.15.6	Development tools	MIT		https://github.com/astral-sh/ruff	Plugin lint checks invoked by check_done.py.

Copyright and License

Copyright (c) 2026 EUMETSAT.

The skills, their references, examples, evaluation cases and documentation are licensed under the Creative Commons Attribution 4.0 International licence (LICENSE, CC BY 4.0): use, share and adapt them, including commercially, with attribution to EUMETSAT and an indication of what you changed.

The three helper scripts under skills/firecube/scripts/ are small conveniences that accompany the firecube skill. They carry the MIT licence in their headers and in LICENSE-MIT, so they can be copied into a plugin project under the usual software terms. Configuration and metadata files (plugin.json, agents/openai.yaml, skills-bench.yaml, ruff.toml) are MIT as well, stated in their headers or adjacent .ABOUT files. The plugin as a whole is therefore CC-BY-4.0 AND MIT.

See AUTHORS.md for authors and maintainers and CONTRIBUTING.md for how to contribute.