Using seedgraph with coding agents

A coding agent writes better test data when it knows seedgraph exists and how to call it. Pick the form your agent reads.

Claude Code: install the plugin

The seedgraph repository is also a Claude Code plugin marketplace. In a terminal:

claude plugin marketplace add jrachid/seedgraph
claude plugin install seedgraph@seedgraph

Or inside a session: /plugin marketplace add jrachid/seedgraph, then /plugin install seedgraph@seedgraph.

The plugin carries one skill, which Claude loads on its own when a task needs SQLAlchemy test data: the call, the shape syntax, the pytest fixtures and the fix for each error. It follows the repository's commits; /plugin marketplace update seedgraph fetches the latest.

To keep the skill in a project without the plugin, copy SKILL.md to .claude/skills/seedgraph/SKILL.md and commit it.

Any agent: a paragraph in AGENTS.md or CLAUDE.md

Paste this into the instructions file your agent reads (AGENTS.md, CLAUDE.md, .cursor/rules, .github/copilot-instructions.md):

## Test data

Create SQLAlchemy test data with seedgraph, never by hand-building objects or setting foreign key ids: `seed(session, User, post=2, post__comment=3)` builds 3 users with 2 posts each and 3 comments per post, flushes, and verifies every foreign key. Existing rows go in `parents=[...]`; required JSON or custom-type columns need `generators={Model: {"column": lambda ctx: ...}}`. In pytest, use the `seedgraph_graph` fixture.

Docs for agents: https://jrachid.github.io/seedgraph/llms-full.txt

Documentation in a form agents read

• llms.txt: a summary with one link per page.
• llms-full.txt: the whole documentation in one Markdown file.

Both are rebuilt with the site on each release, so they describe the version on PyPI.