Dongbumlee/ontosmith ? reverse-engineered prompt

Build me a tool that takes a folder of documents and turns it into a clean knowledge graph that I can trust.

I want to define the schema first with a simple ontology of entity and relation types, add short definitions and a few examples, then run the pipeline so it only writes facts that are supported by the text. Every fact should keep the exact evidence quote and provenance from the source document. It should also score each candidate so I can sort or filter the weaker ones, and it should refuse to build if the graph store is not available or the corpus does not fit the schema.

After the build, I want to be able to search the results, query them through an API, and inspect the written triples from the command line. Please make it work on Python and Neo4j, and keep the build step priced and gated so it does not spend blindly. If you need to look up current docs online, feel free to do that.