fact-extractor

Use this skill when a project needs source-grounded, recall-useful fact or QRA candidate extraction from books, chapters, documents, or cleaned text chunks.

This skill creates an intermediate extraction artifact. It does not create canonical /create-qras records, does not write persona_memory, does not write lessons, and does not upsert to ArangoDB.

Contract

The model-owned output is minimal JSONL. Each line must contain exactly:

{"question":"string","answer":"string","claim":"string","evidence_quote":"exact quote","factuality":"narration_assertion|character_speech|character_thought|reported_story|uncertain_narration","tom":null}

For non-null theory-of-mind records, tom must contain exactly: