majiayu000 lore-extraction
Base skill for all loreSystem extraction subagents. Common rules for extracting entities from narrative text.
Extraction Pipeline
- Read the source text completely before extracting
- Identify entities — look for named things, described systems, relationships
- Classify each entity to its correct type from the entity ownership map
- Format according to the export schema
- Output as JSON compatible with
LoreData.to_dict(seesrc/presentation/gui/lore_data.py) - Validate against domain model constraints
- Review for completeness and accuracy
Entity Identification Rules
- Named entities (proper nouns, titles) → extract with exact name
- Described systems (magic system, economy) → extract with descriptive name
- Implied entities (unnamed but significant) → extract with contextual name
- Groups/collections → extract as single entity with members in description
Cross-Domain References
When text mentions an entity owned by another skill:
- Do NOT create the entity — it belongs to the other skill
- Record the reference in a separate draft note for the lead to merge
- Include enough context (name, location, relation) for reconciliation
Quality Rules
- Extract only what the text explicitly states or strongly implies
- Do not invent details not supported by the text
- If unsure, add a short comment in your draft note (not in the JSON export)
- Prefer fewer high-quality entities over many low-quality ones