Agent setup

Repository skills guide a coding agent through the same Biotope CLI used in the terminal. Install the Python package in the project environment as well; installing a skill does not install its dependencies.

Choose a skill

Skill	Use it for
biotope-croissant	Source curation, typed Python mappings and Biotope graph builds
biocypher	Standalone BioCypher adapters, export schemas and database import

Start with biotope-croissant for a Biotope project. Its reference files document source generation, graph authoring and interpretation. Copy the entire skill folder, including those references, when installing it manually.

Claude Code

Add this repository as a marketplace and install its plugin:

/plugin marketplace add biocypher/biotope
/plugin install biotope@biotope

Alternatively, copy the selected folders from the repository's skills/ directory into your project's .claude/skills/. See Claude Code's plugin guide.

Cursor

For a project-local installation, copy the selected skill folders into .cursor/skills/.

Teams and Enterprise workspaces can use a team marketplace: open the Cursor Dashboard, go to Plugins → Team Marketplaces, choose Add Marketplace, and import biocypher/biotope from GitHub. See Cursor's plugin guide for access and installation settings.

Codex

From the target project, with a clone of Biotope available at ../biotope:

mkdir -p .agents/skills
cp -R ../biotope/skills/biotope-croissant .agents/skills/

Adjust the clone path to your checkout. Copy skills/biocypher in the same way when working on a standalone BioCypher project. User-wide skills can be placed in ~/.agents/skills/. See the Codex skills documentation for discovery and scope.

Work with the agent

Ask the agent to use the biotope-croissant skill and describe the intended graph, for example:

Use biotope-croissant to build a graph from the CSVs in data/. It should connect samples to donors so I can compare measurements by donor. Review the source metadata and explain any missing scientific decisions before choosing a mapping.

The agent records the purpose, reviews source descriptions, generates the source inventory and authors the graph workspace. When construction is requested, it checks and runs the selected pipeline, then reports outputs, exclusions and the independent reads it made against the sources. Existing scientific choices remain part of the project; unresolved choices need input from someone who knows the data.

Database import and querying require a separate task and environment. The tutorial covers file output, while the authoring guide explains the Python contracts.