kg-extract

What problem does it solve?

This skill automates turning unstructured paper text into a validated knowledge graph extraction, removing manual claim curation and reducing inconsistent or invalid KG outputs.

Core Features & Use Cases

• Single-paper extraction: fetches or reads a DOI, URL, PDF, or text and runs the v4 KG extraction prompt to produce structured JSON claims and evidence.
• Corpus batch extraction: prepares and launches batch jobs against the Anthropic Batches API with mandatory user confirmation, polls results, applies coercion maps, validates with Pydantic, and writes per-paper outputs.
• Robust post-processing: applies predicate coercion, migrates legacy evidence formats, enforces schema validation, and reports claim/evidence counts and coercion actions.

Quick Start

Use the kg-extract skill to extract knowledge graph claims from the paper with DOI 10.1038/example and return the validated KG JSON and a summary of claims and evidence.