rp-character-crafter-skill
v1.4.0
Published
SillyTavern immersive RP character card crafting & reverse distillation Agent Skill — COT requirement gathering, 3-layer character build, V1/V2/V3 JSON output, Mode A/B chat distillation, Mode C subtitle (.ass/.srt/.ssa) distillation, Lorebook world book
Maintainers
Readme
RP Character Crafter
Immersive character card crafting + reverse distillation + subtitle extraction + evolutionary iteration Agent Skill for SillyTavern
Not filling forms — creating souls. Every character card is a breathing personality.
Capabilities
| Capability | Description | |---|---| | Character Card Creation | COT requirement gathering → 10-question Character Core + psychological defense mechanisms → 3-layer build → Multi-format output | | Multi-Format Output | Natural Language / W++ / Boostyle / JSON (V1/V2/V3/CHARX) | | Reverse Distillation A | Extract personality from chat logs with D3.5 deep psychological analysis (defense mechanisms / attachment styles / trigger mapping) | | Reverse Distillation B | Only a first message? Design probe conversations, sample responses, then distill | | Subtitle Distillation C | Extract characters from .ass/.srt/.ssa subtitle files — 6-dimension analysis + 7-level confidence + subtitle lorebook | | Character Archetype Library | 12 complete archetype templates (Yandere/Tsundere/Kuudere/Dandere/Devoted/Domineering/Sharp-tongued/Genki/Mysterious/Lazy/Healing Two-faced) | | Evolutionary Iteration | Darwin-style ratchet optimization — 3 feedback entry modes → 4-phase evolution → keep only improvements | | Reflection Protocol | Structured self-critique: 30+ interrogation questions + 3 key tests + RLHF compensation + hallucination safety boundaries | | Advanced Techniques | 8 expert techniques (multi-card synergy / scene anchoring / token budget / character arcs / dialogue rhythm / flaws / first messages / trigger words) | | Troubleshooting | 6 diagnostic workflows (OOC / card activation / world book / robotic dialogue / format compatibility / distillation accuracy) | | Lorebook / World Book | Structured CharacterBook entries triggered by keywords | | Reply Format Enforcement | Parenthesized actions + bare dialogue + 1-3 paragraphs → auto-embedded into JSON fields | | Anti-Hallucination | Asks user when info is missing. Every distilled trait has a confidence level. Reflection has safety boundaries against drift. |
Installation
via npm (recommended)
npm install -g rp-character-crafter-skillThe postinstall script automatically copies SKILL.md and templates to ~/.agents/skills/rp-character-crafter/.
via Git
git clone https://github.com/w32394045-dotcom/rp-character-crafter-skill.git \
~/.agents/skills/rp-character-craftervia curl (SKILL.md only)
mkdir -p ~/.agents/skills/rp-character-crafter
curl -o ~/.agents/skills/rp-character-crafter/SKILL.md \
https://raw.githubusercontent.com/w32394045-dotcom/rp-character-crafter-skill/main/SKILL.mdTemplate Files
templates/sample-character.json— Standard SillyTavern JSON templatetemplates/wplusplus-example.txt— W++ format exampletemplates/boostyle-example.txt— Boostyle format example
Quick Start
Trigger keywords: create a character card lorebook world book distill reverse engineer SillyTavern character setting subtitle .ass .srt .ssa anime character optimize character iterate reflect
Three core workflows
| Scenario | Flow | |---|---| | Create a new character | Step 1 COT → Step 2 Core (10 questions + defense mechanisms) → Step 3 Layer build → Step 4 Format → Step 5 Iterate | | Distill from data | Chat logs → Mode A / Subtitles → Mode C / First message only → Mode B probe sampling | | Optimize existing card | Feedback/chats/subtitles → Darwin ratchet loop (diagnose → fix → verify → keep improvements) + Reflection Protocol |
Core rules (not optional)
- Ask when unsure — missing info = must ask (at least 3 questions at once)
- CHECKPOINT before proceeding — confirm at every stage
- Reply format embedded —
(action) dialogueformat goes intosystem_prompt+mes_example - Never fabricate in distillation — every trait gets a confidence label
- Subtitle distinction — separate scripted dialogue from improvisation, mark translation artifacts
Full Documentation
See SKILL.md — 3982 lines covering:
| Chapter | Content | |---------|---------| | 1-2 | Anti-hallucination rules + Step 1-5 workflow (COT / 10-question core / defense mechanisms / 3-layer build / CHECKPOINTs) | | 3-5 | Complete format guides (V1/V2/V3 JSON schemas) + reply format spec + packaging | | 6 | Lorebook CharacterBook structure & writing principles | | 7-8 | Depth building templates + quality checklist + anti-patterns | | 9 | Reverse distillation Mode A/B (D3.5 psychological analysis: defense mechanisms / attachment styles / trigger mapping / non-verbal cues) | | 10 | Creation examples | | 11 | Subtitle Distillation Mode C (.ass/.srt/.ssa → C2 analysis techniques + full example) | | 12 | Character Archetype Library (12 archetypes: 5-question core + 3-layer build + sample lines + variants + traps) | | 13 | Advanced Techniques (multi-card synergy / scene anchoring / token budgets / character arcs / dialogue rhythm / flaws / first messages / trigger words) | | 14 | Troubleshooting (OOC / card activation / world book / robotic dialogue / format compatibility / distillation accuracy) | | 15 | Evolutionary Iteration (Darwin ratchet: 3 feedback modes → 4 phases → independent verification → loop exit) | | 16 | Reflection Protocol (30+ interrogation questions → 3 key tests → RLHF compensation → hallucination safety boundaries → drift detection) |
Darwin Baseline Score: 87/100
Evaluated with Darwin Skill 2.0 9-dimension rubric. Runtime neutrality: PASS (0 red flags).
License
MIT
