Agent skill¶
GeoLibre ships an agent skill — a SKILL.md that teaches an external AI
coding agent (Claude Code, Claude Desktop, or any runtime that loads skills) how
to build GeoLibre maps on your behalf.
It lives in the repository at
skills/geolibre/.
What it is for¶
GeoLibre already has an in-app AI Assistant, which chats with your data inside the running app and applies changes through the app's own store. The skill covers the other direction: an agent working outside GeoLibre — in your terminal, your editor, or a notebook — that should produce a GeoLibre map as its output.
| AI Assistant | Agent skill | |
|---|---|---|
| Runs | Inside the app | In your agent, outside the app |
| Acts on | The live map you are looking at | A .geolibre.json file on disk |
| Needs | A configured AI provider | An agent that loads skills |
| Output | Edits you can undo with Ctrl/Cmd + Z | A project file and/or a standalone HTML page |
They compose: an agent authors the project, you open it in GeoLibre and keep working with the Assistant.
What it teaches¶
- Which entry point to use — the MCP server for a chat session, the Python package for notebooks and scripts, hand-written project JSON when nothing is installed, and the embed API for a live instance in a page.
- The authoring workflow — create a project, add layers, frame the camera, style or classify, add a legend or colorbar, export a shareable HTML page.
- The rules that are easy to get wrong — that
classify_layerneeds inlined GeoJSON, that a local file path does not travel to a recipient, that abboxresolves to a camera approximately, that basemap and color-ramp names should come fromlist_catalograther than a guess, and thatexport_html'sapp_urlis a trust boundary.
Install¶
git clone --depth 1 https://github.com/opengeos/GeoLibre /tmp/geolibre
mkdir -p ~/.claude/skills
cp -r /tmp/geolibre/skills/geolibre ~/.claude/skills/
Use .claude/skills/ inside a project instead of ~/.claude/skills/ to scope
it to one repository. Start a new agent session afterwards; the skill loads
itself when a request matches.
For the shortest path, install the MCP server the skill prefers:
pip install "geolibre[mcp]"
claude mcp add geolibre -- geolibre-mcp --root ~/maps
The skill still works without it — it falls back to the Python API and to writing project JSON directly.
Other agent runtimes¶
SKILL.md is plain Markdown with YAML frontmatter, so most agent runtimes that
support skills can load it as-is. If yours has no skill loader, paste the file
into your system prompt or AGENTS.md — it is written to stand on its own, and
its references/ files are only read when a task needs them.