Use Marsilea with AI assistants#

Marsilea ships two things for AI coding assistants: an official skill and an llms.txt that any assistant can read.

Agentic skill#

Run these inside Claude Code:

/plugin marketplace add Marsilea-viz/marsilea-skill
/plugin install marsilea@marsilea-marketplace

The skill then activates on its own whenever you ask for a composable visualization, a heatmap with annotations, an oncoprint, an UpSet plot. It will then write Marsilea code against the current API instead of guessing.

Marsilea itself still has to be installed in the Python environment the assistant runs code in:

pip install marsilea

The skill lives in its own repository, Marsilea-viz/marsilea-skill; open an issue there for skill-specific problems.

Any other assistant: llms.txt#

llms.txt is a plain-text summary of a project written for language models: what the library is, the core API, and a link per topic. Marsilea’s llms.txt lives at

Paste that URL into an assistant that can fetch pages, or paste the file’s contents into the context window of one that cannot. It is short enough to include wholesale and covers the canvases, the plotters, and one linked example per plot type.

Getting good results#

  • Name the plot type you want (“oncoprint”, “UpSet plot”, “annotated heatmap”). Each one has a gallery example with full source, and assistants that fetch pages will find it.

  • Say which side each piece belongs on. Marsilea’s model is a main plot plus add_top / add_bottom / add_left / add_right, so phrasing the request that way maps straight onto the API.

  • If an assistant produces code for an older API, point it at the API reference or the llms.txt above.