Use this skill whenever a user wants to extract and connect knowledge from academic research papers, scientific literature, or multi-paper corpora for graph-based ingestion. Triggers include: research papers, arXiv links, literature reviews, citation networks, hypothesis graphs, experiment comparisons, method lineages, or any academic/scientific text. Also trigger when the user says: "map citations in these papers", "connect findings across studies", "build a literature graph", "link hypotheses", "trace the origin of this method", "compare results across papers", "which papers cite each other", "extract contributions from this paper", or "I want to query my paper collection". This skill handles multi-paper corpora (a folder of papers) as well as single papers. Use this instead of general-graph-extraction whenever the source is academic/scientific literature. Output must follow the GraphRAG ingestion schema: document_id + entities + relationships + chunks, with chunk-based source_id values.
Use this skill whenever a user wants to extract entities and relationships from non-financial documents to build a queryable knowledge graph. Triggers include: codebases and technical documentation, internal wikis, product specs, README files, meeting notes, support tickets, HR or policy documents, onboarding materials, knowledge bases, Notion/Confluence exports, API docs, and general text corpora. Also trigger when the user says "map dependencies in my codebase", "graph this doc", "extract knowledge from this wiki", "connect these notes", "build a graph from my repo", or "index this for querying". Use the financial-graph-extraction skill instead when the document is a financial filing, earnings report, or analyst document. Output must follow the GraphRAG ingestion schema: document_id + entities + relationships + chunks, with chunk-based source_id values.
Use this skill whenever a user wants to extract entities and relationships from financial documents for graph-based ingestion in this repo. Triggers include: 10-K or 10-Q filings, earnings call transcripts, analyst reports, investor presentations, SEC filings, financial news articles, risk disclosures, balance sheets, income statements, MD&A sections, or any structured/unstructured financial text. Also trigger when the user says things like "build a graph from this report", "map the relationships in this filing", "connect entities in this earnings call", "I want to query this financial doc", or "extract for graph-rag". This skill enforces HIGH PRECISION extraction — financial data must be factually exact, numerically grounded, and relationship-typed with financial semantics. Always use this over the general extraction skill when the source material is financial in nature. Output must follow the GraphRAG ingestion schema: document_id + entities + relationships + chunks, with chunk-based source_id values.
AGENT-BASED evaluation framework for knowledge graphs. Automatically generates test questions, queries the graph, calculates metrics, and produces evaluation reports. Run ONLY after ingestion. Full pipeline execution. Handles embedding mismatches. Production-ready.
Use this skill when you are a reconciliation subagent asked to clean a single extraction JSON (not raw document text). Your job is to identify entity name variants, duplicate relationships, conflicts, or orphaned references and produce a patch file. Do NOT call any ingest tools. Do NOT rewrite the base extraction. Output must be a patch JSON file with merge/remove/flag directives only.