EvoOntology

What it does

• Versioned Ontology Layer over heterogeneous tables, files and databases — metric definitions, entity relationships and business constraints queried on demand instead of re-inferred every session
• Self-evolving by design: grounded in workload evidence and adapted from execution trajectories under controlled evaluation, not hand-maintained
• The Claude Code plugin ships three commands (/evo-build, /evo-evolve, /evo-visualize), builder and evolver skills, an auto-wired semantic MCP server and a session-start reminder
• Academic grounding with practical packaging: arXiv 2609.15779, bilingual English/Chinese docs, MIT, plugin v1.0.2

When to use it

Data agents working over a messy warehouse where table and column names do not explain metrics or joins — the agent queries the ontology instead of guessing semantics from names alone.

Getting started

git clone https://github.com/ruc-datalab/EvoOntology.git
# the plugin bundles its core package — no venv, no pip install:
claude --plugin-dir EvoOntology/plugins/claude-code
# then in a session: /evo-build, /evo-evolve, /evo-visualize

Things to know

Research-grade academic project — expect academic ergonomics. The install docs still cite MeiduoChong/EvoOntology as the marketplace source, a repository that no longer exists, and the ruc-datalab repository ships no marketplace.json — hence the local-checkout install above. The ontology self-evolves from execution trajectories, so review evolved versions before trusting them downstream. MIT; plugin v1.0.2.