What problem does it solve? Flat similarity search cannot tell you that one document overrules, distinguishes, or relies on another. This Skill builds a typed reference graph over an ingested corpus so you can traverse the actual argument structure — for example, asking which documents overrule a given authority — instead of getting back fuzzy-matched paragraphs. ## Core Features & Use Cases - Typed edge extraction: Detects inter-document references mechanically, then uses model judgment to classify each edge as overrules, distinguishes, relies_on, extends, refutes, supersedes, or cites. - Native graph storage: Writes edges via gbrain link with link_source=citation-graph provenance, making them queryable with gbrain graph-query --type and bulk-removable without touching other writers' edges. - Verified runs: The pipeline is not complete until a graph-query walk from a hub document returns the written edges; idempotent re-runs skip duplicates. - Use Case: Ingest a case-law corpus as a gbrain source, run this Skill, then ask gbrain graph-query doc-a --type overrules --direction in to learn that doc A was overruled by doc C — an answer vector search alone cannot produce. ## Quick Start Build a typed citation graph over my ingested case-law corpus so I can trace which documents overrule or rely on each other.
