Skip to content

soul-system.md

Latest commit

 

History

History
112 lines (84 loc) · 4.93 KB

soul-system.md

File metadata and controls

112 lines (84 loc) · 4.93 KB

Soul System

Digital twin identity scaffold management for creating and testing aligned AI personas.

Overview

LLMs can embody specific personas, but creating comprehensive identity documents and testing alignment across different models is manual and error-prone. The Soul System provides a structured approach to capturing, validating, and deploying personality models.

Architecture

  • Digital Twin Service (server/services/digitalTwin.js): Document management, testing, enrichment
  • Digital Twin Routes (server/routes/digital-twin.js): REST API endpoints mounted under /api/digital-twin/*
  • Digital Twin Page (client/src/pages/DigitalTwin.jsx): Overview, Documents, Test, Enrich, Export tabs

Features

Five-Tab Interface

  1. Overview Tab: Dashboard showing soul health score, document counts, test scores, enrichment progress, and quick actions
  2. Documents Tab: Sidebar-based document editor for managing soul markdown files by category (core, audio, behavioral, enrichment)
  3. Test Tab: Multi-model behavioral testing against 14 predefined tests, with side-by-side result comparison
  4. Enrich Tab: Guided questionnaire across 10 categories that generates soul document content from answers
  5. Export Tab: Export soul for use in external LLMs (System Prompt, CLAUDE.md, JSON, individual files)

CoS Integration

  • Soul context automatically injected into agent prompts when enabled
  • Settings control autoInjectToCoS and maxContextTokens
  • Prompt template cos-agent-briefing.md includes {{soulSection}}

Directory Structure

data/digital-twin/
├── meta.json              # Document metadata, test history, settings
├── SOUL.md                # Core identity
├── Expanded.md            # High-fidelity spec
├── BEHAVIORAL_TEST_SUITE.md  # 14 behavioral tests
├── AUDIO*.md              # Audio preferences
├── MEMORIES.md            # Generated via enrichment
├── FAVORITES.md           # Generated via enrichment
└── PREFERENCES.md         # Generated via enrichment

Enrichment Categories

Category Description
Core Memories Formative experiences
Favorite Books Books that shaped thinking
Favorite Movies Films that resonate
Music Taste Cognitive infrastructure
Communication How to give/receive info
Decision Making Approach to choices
Values Core principles
Aesthetics Visual preferences
Daily Routines Structure habits
Career/Skills Professional expertise

Additional Categories (M33.1)

  • non_negotiables: Principles and boundaries that define your limits
  • decision_heuristics: Mental models and shortcuts for making choices
  • error_intolerance: What your digital twin should never do

Validation & Analysis

  1. Completeness Validator: Checks for 6 required sections (identity, values, communication, decision making, non-negotiables, error intolerance), shows percentage complete with actionable suggestions
  2. Contradiction Detector: AI-powered analysis to find inconsistencies between soul documents, with severity levels and resolution suggestions

Dynamic Testing

  • AI generates behavioral tests based on soul content
  • Targets values, communication style, non-negotiables, and decision patterns
  • Returns structured tests with prompts, expected behaviors, and failure signals

Writing Sample Analysis

  • Paste writing samples to extract authentic voice patterns
  • Analyzes: sentence structure, vocabulary, formality, tone, distinctive markers
  • Generates WRITING_STYLE.md document content

Context Optimization

  • Document Weighting: Priority slider (1-10) on each document
  • Higher weighted documents preserved first when context limits force truncation

API Endpoints

Route Description
GET /api/digital-twin Status summary
GET /api/digital-twin/documents List documents
POST /api/digital-twin/documents Create document
PUT /api/digital-twin/documents/:id Update document
DELETE /api/digital-twin/documents/:id Delete document
GET /api/digital-twin/tests Get test suite
POST /api/digital-twin/tests/run Run single-model tests
POST /api/digital-twin/tests/run-multi Run multi-model tests
GET /api/digital-twin/enrich/categories List enrichment categories
POST /api/digital-twin/enrich/question Get next question
POST /api/digital-twin/enrich/answer Submit answer
POST /api/digital-twin/export Export soul
GET /api/digital-twin/validate/completeness Check soul completeness
POST /api/digital-twin/validate/contradictions Detect contradictions
POST /api/digital-twin/tests/generate Generate dynamic tests
POST /api/digital-twin/analyze-writing Analyze writing samples

Related Features