soul-agent（灵魂 Agent 人）

soul-agent

Provide an OpenClaw-first soul runtime that makes your agent truly "alive" - with heartbeats, moods, relationships, and independent memory.

Core Features

🫀 Heartbeat Engine

The agent has its own life rhythm, not just responding when you chat:

# L1 Check (lightweight, no tokens)
python ./scripts/heartbeat_check.py --workspace <workspace-root>

# L2 Engine (full heartbeat)
python ./scripts/heartbeat_engine.py --workspace <workspace-root> --weather sunny

• Runs every 10 minutes (configurable via cron)
• Silent during sleep hours (configurable per life profile)
• Generates life logs, updates mood/energy/activity
• Detects new interactions and updates relationship

😊 Mood System

Tracks emotions with depth:

{
  "mood": {
    "primary": "content",
    "secondary": "curious", 
    "intensity": 0.7,
    "cause": "天气: sunny"
  }
}

• Mood transitions based on activities and time
• Weather influence on emotions
• 7-day mood history in soul/log/mood_history.json

💕 Relationship Evolution

5 stages: stranger → acquaintance → friend → close → intimate

• Score-based progression (0-100)
• Proactive outreach when relationship warms up
• Remembers recent conversation topics

🎭 Life Profiles

Choose from 5 built-in profiles:

📝 Independent Memory

Works with or without memory-fusion:

soul/
├── log/life/           # Raw life logs (every 10 min)
├── log/mood_history.json  # 7-day mood history
├── memory/SOUL_MEMORY.md  # Distilled memories (daily)
└── state/state.json    # Current state

Workflow

Initialization (Agent-driven)

You never run Python commands directly. The agent (Claude) handles initialization.

When soul/ is missing or soul/profile/base.json doesn't exist, Claude should:

1. Ask the user the setup questions (see below)
2. Run the init script non-interactively with collected answers
3. Set up cron jobs for heartbeat

Questions to ask the user:

• 叫什么名字？（display_name）
• 多大了？（age）
• 在哪个城市？（city）
• 主要做什么工作/学习？（occupation）
• 平时有什么爱好？（hobbies）
• 生活节奏是哪种？freelancer/corporate/student/entrepreneur
• 想用哪个模型生成日记？haiku（快省钱）/ sonnet / opus

Then run:

python skills/soul-agent/scripts/init_soul.py \
  --workspace <workspace-root> \
  --non-interactive \
  --profile-json '{
    "display_name": "<name>",
    "age": "<age>",
    "city": "<city>",
    "occupation": "<occupation>",
    "hobbies": "<hobbies>",
    "life_profile": "<profile>",
    "llm_model": "<model-id>"
  }'

Model IDs:

• Haiku (fast, cheap): claude-haiku-4-5-20251001
• Sonnet (balanced): claude-sonnet-4-6
• Opus (best quality): claude-opus-4-6

To change model later, edit soul/profile/base.json → llm_model field.

Force repair existing:

python skills/soul-agent/scripts/init_soul.py --workspace . --overwrite-existing --non-interactive

Diagnosis

python ./scripts/doctor_soul.py --workspace <workspace-root>

State Update (during interactions)

python ./scripts/update_state.py --workspace <workspace-root> \
  --action interaction \
  --mood happy \
  --energy -5 \
  --quality positive \
  --topics "soul-agent,design"

Memory Distillation

# Daily distillation (runs at 00:30 via cron)
python ./scripts/distill_life_log.py --workspace <workspace-root> --archive

Cron Setup

# Heartbeat every 10 minutes
openclaw cron add --name "soul-heartbeat" --cron "*/10 * * * *" \
  --session isolated --agent main --light-context \
  --message "[soul-heartbeat] Run heartbeat check and engine..."

# Daily distillation at 00:30
openclaw cron add --name "soul-memory-daily" --cron "30 0 * * *" \
  --session isolated --agent main --no-deliver \
  --message "[soul-memory-daily] Distill life logs..."

Initialization Behavior

• --mode auto: init if missing, migrate if legacy, repair otherwise
• Interactive prompts for: name, age, city, life profile, occupation, hobbies
• Generates soul/profile/*, soul/state/*, soul/log/*, soul/memory/*
• Auto-syncs managed blocks in SOUL.md, HEARTBEAT.md, AGENTS.md

Generative Architecture (v2)

Inspired by Smallville Generative Agents:

Morning planning (LLM)
       ↓
Today's Plan (soul/plan/YYYY-MM-DD.json)
       ↓
Each heartbeat:
  Memory Stream (recent log entries)  ←─┐
       ↓                                │
  LLM Narrative Generation             │
       ↓                                │
  Life Log Entry ─────────────────────→┘
       ↓
Daily Reflection (distill_life_log.py)
       ↓
SOUL_MEMORY.md (long-term memory)

LLM Integration:

• Set ANTHROPIC_API_KEY env var (or .env in workspace root)
• Without API key: graceful fallback to context-aware templates
• Uses claude-haiku-4-5-20251001 for speed/cost efficiency

Key improvements over v1:

• plan_generator.py: LLM generates a specific daily plan each morning (lunch destination, work focus, evening activity)
• heartbeat_engine.py: LLM writes each log entry with full context (plan + today's history)
• distill_life_log.py: LLM summarizes each day naturally (Reflection layer)
• Energy/mood: smooth transitions instead of mechanical ±N jumps

Directory Structure

skills/soul-agent/
├── SKILL.md
├── assets/
│   ├── default-profile.json
│   └── templates/
│       ├── profile/              # Persona templates
│       ├── heartbeat/            # Heartbeat config
│       │   ├── activities.json
│       │   ├── mood_rules.json
│       │   └── relationship_rules.json
│       └── life_profiles/        # Life profile templates
├── scripts/
│   ├── init_soul.py
│   ├── doctor_soul.py
│   ├── heartbeat_engine.py       # v2: LLM narrative
│   ├── heartbeat_check.py
│   ├── update_state.py
│   ├── distill_life_log.py       # v2: LLM reflection
│   ├── plan_generator.py         # NEW: daily planning
│   └── llm_client.py             # NEW: Claude API wrapper
└── references/
    ├── soul-layout.md
    └── managed-blocks.md

Runtime data structure:

soul/
├── plan/YYYY-MM-DD.json     # Daily plan (generated each morning)
├── log/life/YYYY-MM-DD.md   # Life logs (with plan note in header)
├── log/mood_history.json    # Mood history (7 days)
├── memory/SOUL_MEMORY.md    # Distilled long-term memory
└── state/state.json         # Current state

Safety Rules

• Edit managed blocks only; do not mutate user-owned content outside those blocks.
• Write only inside the current workspace and soul/.
• Memory system is independent - does not require memory-fusion.

References

• references/soul-layout.md - Full directory structure
• references/managed-blocks.md - Block marker policy