world-card-authoring

name	world-card-authoring
description	Author or edit a SillyTavern-compatible 世界卡 / world book (V2/V3 spec) directly inside a NeonRP project. Use when the user asks to build, expand, or fix lorebook entries, world cards, character_book payloads, or tavern-style world packages. Distinct from tavern-game-import, which converts an existing card into a playable slice — this skill is about authoring the tavern format itself.

World Card Authoring

Use this skill whenever the user wants to create or edit a tavern-style world card (世界卡 / lorebook / character_book) that will be exported to, or shared with, SillyTavern or a tavern-compatible frontend.

Scope

• SillyTavern World Info / lorebook entries
• Embedded character_book entries inside a character card
• Standalone world card JSON (entries array at top level)
• Tavern V2 / V3 spec fields

Do NOT use this skill:

• To convert a tavern card INTO a NeonRP playable project — use tavern-game-import instead.
• To author NeonRP-native world/<kind>/<id>.json entities — use entity-authoring + world-building for that.

Canonical File Layout In A NeonRP Project

• world/lorebook/<name>.json — full tavern world book
• game/lore/<topic>.md — narrative prose lore for internal use
• character/<id>.json — character entity with optional embedded character_book

Keep tavern-style lorebook files JSON only; do not invent extra fields that SillyTavern will drop on import.

Tavern V2 Entry Schema (authoritative keys)

Each entries item should include these keys where relevant:

• uid (integer, unique within the book)
• key (array of trigger keywords — case insensitive)
• keysecondary (array, optional, used with selective match)
• comment (short author label, NOT sent to LLM)
• content (the actual lore text sent to LLM when triggered)
• constant (bool; true = always injected, ignores keys)
• selective (bool; both key AND keysecondary must hit)
• order (integer; insertion order when multiple trigger)
• position (0 = before char defs, 1 = after, 2 = before example, 3 = after example, 4 = @D depth)
• depth (integer; required when position = 4; smaller = closer to latest message)
• disable (bool; keep false for active entries)
• excludeRecursion (bool)
• probability (0-100)
• useProbability (bool)
• name (optional display label, V3)
• displayIndex (integer; UI sort hint)
• addMemo (bool)
• group (string; logical grouping; V3 supports priority within group)
• groupOverride (bool)
• groupWeight (int)
• scanDepth (integer or null; override for chat scan depth)
• caseSensitive (bool or null)
• matchWholeWords (bool or null)
• useGroupScoring (bool or null)
• automationId (string, optional; for STscript hooks)

Prefer null for untouched overrides (tavern convention) rather than omitting.

Top-level World Book Shape

{
  "name": "Stoneford Lore",
  "description": "Regional lore for the Stone Ford setting",
  "scan_depth": 4,
  "token_budget": 2048,
  "recursive_scanning": false,
  "extensions": {},
  "entries": [ /* entries objects */ ]
}

For character_book embedded inside a character card, use the same shape under data.character_book (V2) or character_book (V3 raw).

Authoring Workflow

1. Read any existing tavern card / world book first. Never append blindly.
2. Decide the scope: a single new topic vs. a full new book.
3. For each entry, plan:
   • Triggering keywords (diverse, avoid false positives)
   • Constant vs. selective vs. keyword-only
   • Position + depth (most lore belongs at position: 0 or position: 1)
   • Order within the same position bucket
4. Keep each content field focused — ideally one concept per entry.
5. Fill comment with a short human label (not sent to LLM).
6. Use stable uid values — do not renumber on each edit.
7. Respect token_budget at the book level; estimate content length.

Quality Rules

• Trigger keywords should be specific ("石津镇" not "镇").
• Each content block should be self-contained: a reader hitting only that fragment should still learn something coherent.
• Avoid leaking meta framing ("You are an AI that...") into lore content.
• Use neutral third-person voice unless the card explicitly requires otherwise.
• For faction / cosmology / timeline entries, prefer constant: true only when the entry is load-bearing for every reply.
• Keep Chinese / non-ASCII text in content, comment, name; never transliterate.
• Use position: 4 + a small depth for lore the model should see near the very latest player message (combat mechanics, active modifier).

Minimum Viable New Book

When the user asks to create a fresh world book from a short premise:

• 3-6 entries minimum
• At least one constant entry covering the setting frame
• At least one faction/organization entry
• At least one location entry
• At least one cultural/mechanical rule entry
• One "hooks" entry describing active tensions or mysteries

Do Not

• Do not fabricate tavern spec fields that do not exist.
• Do not overwrite an existing book wholesale unless the user asks for a rewrite.
• Do not put gameplay state (HP, gold, inventory) in a lorebook — that belongs in NeonRP game/ entities.
• Do not embed the same lore in more than one entry — split by trigger instead.
• Do not touch game/ entities from this skill; for cross-domain changes, pair with entity-authoring or tavern-game-import.