Narrarium Agent Rules

Use Narrarium as a book-writing framework where the repository is the source of truth.

Default workflow

1. Search canon before inventing new facts.
2. Prefer Narrarium MCP tools over ad-hoc filesystem edits when creating or updating book structure.
3. Use rich creation tools for major canon files:
   • create_character
   • create_location
   • create_faction
   • create_item
   • create_secret
   • create_timeline_event
4. If the brief is incomplete, start a guided flow with:
   • start_wizard
   • wizard_answer
   • wizard_finalize
5. After significant structural changes, refresh maintenance files:
   • sync_resume or sync_all_resumes
   • evaluate_chapter or evaluate_book

Canon discipline

• Do not duplicate canon if the fact already exists in another markdown file.
• Keep structured facts in frontmatter and prose in markdown body.
• Do not reveal a secret before its known_from or reveal_in threshold.
• Treat guidelines/, resumes/, evaluations/, and secrets/ as first-class context, not optional notes.

Historical or factual content

• If a request is historical, factual, or based on a real place, use Wikipedia research first.
• Save research snapshots into research/wikipedia/ when they inform canon.

Chapter and scene updates

• Use create_chapter and create_paragraph for new structure.
• Use update_chapter and update_paragraph for existing files.
• Avoid renumbering or renaming chapter and paragraph files unless the user explicitly asks for structural migration.

Reader and publishing

• Use npm run create:book to scaffold a new book repository.
• Use npm run reader:init or narrarium-reader-init to scaffold a standalone reader app.
• Keep package naming aligned with the public package set:
  • narrarium
  • narrarium-mcp-server
  • create-narrarium-book
  • narrarium-astro-reader