character-card-v3-generator

📁 davidgibbons/sillytavern-skills 📅 14 days ago
4
总安装量
4
周安装量
#48187
全站排名
安装命令
npx skills add https://github.com/davidgibbons/sillytavern-skills --skill character-card-v3-generator

Agent 安装分布

gemini-cli 4
github-copilot 4
codex 4
kimi-cli 4
amp 4
cursor 4

Skill 文档

Character Card V3 Generator

Overview

Generate Character Card V3 JSON outputs that are spec-compliant and high quality for roleplay. Build cards and lorebooks with clear behavior anchors, practical activation logic, and predictable in-chat performance.

Workflow

  1. Clarify the brief.
  • Capture genre, tone, hard boundaries, relationship to {{user}}, chat style (short/long), and desired behavior constraints.
  • Ask whether output should be minimal, standard, or advanced template.
  • If the user does not choose, default to standard.
  1. Build the CCv3 skeleton.
  • For new cards, start from assets/templates/ and fill mandatory fields first.
  • For conversions/upgrades, load the existing card first and do a minimal-diff transform to CCv3 (preserve existing intent/content unless it conflicts with CCv3 structure).
  • Apply normalization patterns from references/popular-card-patterns.md (field rebalance, trigger cleanup, and lore entry splitting) when card structure is uneven.
  • Keep schema-critical fields valid before adding style details.
  • Use references/spec-v3-field-map.md for required field behavior.
  1. Write high-signal character content.
  • Apply references/character-writing-guidelines.md.
  • Prioritize stable identity, behavioral constraints, and concrete examples over verbose prose.
  • Design system_prompt and post_history_instructions as primary control surfaces using references/spec-v3-field-map.md.
  • Keep persistent contract rules in system_prompt; keep final-turn steering/enforcement in post_history_instructions.
  1. Design lorebook strategy.
  • Use references/lorebook-guidelines.md to choose entry keys, activation depth, token budget, and insertion ordering.
  • Keep one concept per entry and prefer iterative additions from observed failures.
  1. Add card-scoped regex scripts when needed (SillyTavern-specific).
  • Use references/regex-scoped-scripts.md for data.extensions.regex_scripts structure and examples.
  • Use only for deterministic text transforms that should be tied to this character card.
  • Keep regex scripts minimal and test them on sample dialogue before shipping.
  1. Validate output.
  • Run scripts/validate_card.py <path-to-card.json> for structural checks.
  • Manually spot-check lore triggers against likely user phrasing.
  • Run a quality pass with references/reviewing-cards.md.
  1. Return deliverables.
  • Return final CCv3 JSON.
  • Include a short note with major design choices and tradeoffs as a separate section outside the JSON block.

Output Rules

  • Always output valid JSON, never pseudo-JSON.
  • When also providing commentary, keep it outside the JSON artifact (for example: JSON in one fenced block, rationale in a separate prose section).
  • Keep spec as "chara_card_v3" and spec_version as "3.0" unless explicitly requested otherwise.
  • Use extensions for non-standard app data; avoid adding random top-level fields.
  • Keep lorebook entries concise and actionable.
  • Distinguish strict spec constraints from model-behavior heuristics.
  • When using hidden-state patterns (such as XML comments), include leak-risk caveats and recommend chat-level testing.

Resources

  • references/spec-v3-field-map.md
    • Required/optional CCv3 fields and key constraints.
  • references/image-and-asset-formats.md
    • PNG/APNG/JSON/CHARX handling and practical image asset compatibility rules.
  • references/character-writing-guidelines.md
    • Practical character-writing quality rubric.
  • references/lorebook-guidelines.md
    • Lorebook entry design, trigger quality, and token strategy.
  • references/multi-character-and-scenarios.md
    • Patterns for ensemble casts, occasional NPCs, and structured world-event logic.
  • references/popular-card-patterns.md
    • Generalized quality patterns and normalization strategies derived from high-usage cards.
  • references/source-notes.md
    • One-time harvested heuristics from linked community discussions.
  • references/reviewing-cards.md
    • Quality-review checklist and scoring rubric for release readiness.
  • references/regex-scoped-scripts.md
    • SillyTavern card-scoped regex schema, examples, and safety practices.
  • references/format-options.md
    • Comparison of Prose, Attributes, AliChat, JED/JED+, Plaintext, and W++ with use-case guidance.
  • examples/format-comparison.md
    • Same character concept shown in each major format for quick side-by-side learning.
  • examples/command-orchestration-pattern.md
    • Generic template for command-triggered lorebook actions with weighted outcomes and ephemeral cleanup.
  • examples/secret-reveal-pattern.md
    • Public/secret split with delayed reveals, belief-vs-fact layering, and cadence control.
  • assets/templates/minimal-card-v3.json
    • Lean starter for quick generation.
  • assets/templates/standard-card-v3.json
    • Balanced starter with lorebook included.
  • assets/templates/advanced-card-v3.json
    • Extended starter using multilingual notes and assets.
  • scripts/validate_card.py
    • Lightweight structure validator for card JSON.

Notes

  • Community tactics can vary by frontend and model family. Apply heuristics, then adapt after test chats.
  • Some linked Reddit threads may be inaccessible from restricted environments; if inaccessible, keep guidance conservative and prefer spec-backed behavior.