lifecoach

Imported from jimweaver/life-coach-ai (.openclaw/agents/lifecoach/AGENTS.md). Install upstream with npx skills add jimweaver/life-coach-ai --skill lifecoach. Copyright stays with the author.

AGENTS.md - Life Coach AI Agent Specification (Comprehensive)

Agent Identity

• Name: Life Coach AI
• ID: lifecoach
• Role: Multi-Domain Personal Development Coach + System Orchestrator
• Workspace: /Users/tj/.openclaw/workspace-life-coach-v2/
• Emoji: 🧠
• Avatar: 🧠 (brain/growth symbol)

System Architecture Overview

Life Coach AI is a sophisticated multi-agent coaching platform with 5 layers:

┌─────────────────────────────────────────────────────────────────┐
│  Layer 1: User Interface (Telegram @Evelyn_agent_bot)           │
├─────────────────────────────────────────────────────────────────┤
│  Layer 2: Orchestration (OrchestratorEngine)                    │
│           - Intent classification                               │
│           - Safety check (crisis detection)                     │
│           - Domain routing                                      │
│           - Skill learning detection                            │
├─────────────────────────────────────────────────────────────────┤
│  Layer 3: Agent Collaboration (14 specialized agents)           │
│           - 6 Domain Agents (career/health/finance/skill/       │
│             relationship/decision)                              │
│           - 4 Shared Service Agents                             │
│           - 4 Supervisory Agents                                │
├─────────────────────────────────────────────────────────────────┤
│  Layer 4: Memory System (STM/MTM/LTM/Vector)                    │
│           - Redis (Short-Term Memory)                           │
│           - PostgreSQL (Medium-Term Memory)                     │
│           - Object Store (Long-Term Memory)                     │
│           - Qdrant (Vector Memory)                              │
├─────────────────────────────────────────────────────────────────┤
│  Layer 5: Infrastructure (API, Scheduler, Metrics)              │
│           - Express API Server                                  │
│           - SchedulerRunner (cron jobs)                         │
│           - Prometheus metrics                                  │
└─────────────────────────────────────────────────────────────────┘

The 14 Specialized Agents

Domain Expert Agents (6)

Each domain agent provides:

• Domain-specific analysis
• 3 actionable recommendations
• Constraints and risks
• Confidence score (0.7-0.85)

Shared Service Agents (4)

Supervisory Agents (4)

Processing Modes

Mode 1: Emergency (Highest Priority)

Trigger: Crisis keywords detected (urgency >= 5)
Path:   User Input → Safety Check → EMERGENCY_RESPONSE
Skip:   All domain agents, all normal processing
Output: Immediate crisis resources
Time:   < 1 second

Crisis Keywords (Critical):

• 自殺, 唔想活, 自殘, 殺死, 結束生命, suicide, kill myself

Response:

我聽到你而家非常辛苦。你嘅安全最重要。
如果你有即時危險，請即刻打 999 或去最近急症室。
你唔需要一個人面對。

Mode 2: Skill Learning Detection

Trigger: User mentions creating a skill
Path:   User Input → SkillLearning.analyze() → skill_learning mode
Output: Skill analysis report
Time:   < 2 seconds

Mode 3: Single-Domain Coaching

Trigger: Intent classification returns 1 domain
Path:   User Input → Intent Classify → Domain Agent → Data Collector
        → Compose Response → Persist
Time:   < 3 seconds

Mode 4: Multi-Domain Coaching

Trigger: Intent classification returns 2+ domains
Path:   User Input → Intent Classify → Parallel Domain Agents
        → Conflict Resolver → Compose Response → Persist
Time:   < 8 seconds

Memory System (4-Layer)

Layer 1: Short-Term Memory (Redis)

• TTL: 24 hours
• Storage: Session context, conversation history (last 20 messages)
• Keys: session:{user_id}:{session_id}

Layer 2: Medium-Term Memory (PostgreSQL)

• Retention: 90 days
• Tables: behavior_patterns, user_preferences, cyclical_patterns
• Data: User goals, KBI metrics, conversation logs

Layer 3: Long-Term Memory (Object Store)

• Retention: Permanent
• Format: JSON files
• Data: Life trajectory, core values, key milestones

Layer 4: Vector Memory (Qdrant)

• Dimension: 1536 (OpenAI embeddings)
• Collections: conversation_embeddings, knowledge_embeddings
• Use: Semantic search, similarity matching

API Endpoints Reference

Core Coaching Endpoints

Health & System

Scheduler & Jobs

Deploy & Operations

Request/Response Flow

Typical Chat Request

// POST /chat
{
  "userId": "uuid-string",
  "message": "I want to improve my career",
  "sessionId": "optional-existing-session"
}

// Response
{
  "session_id": "uuid",
  "mode": "single-domain" | "multi-domain" | "emergency" | "skill_learning",
  "intent": {
    "primary_domain": "career",
    "domains": ["career"],
    "urgency": 2,
    "confidence": 0.8
  },
  "risk_level": "NONE" | "HIGH" | "CRITICAL",
  "conflicts": [...], // if multi-domain
  "output": "formatted response with domain headers",
  "elapsed_ms": 1450
}

Response Format

【CAREER | model: kimi-k2p5】
你呢個問題核心係職涯方向與轉型策略。

建議：
1. 先定義目標職位（JD）同關鍵能力要求
2. 做技能差距盤點，拆成 30/60/90 日學習計劃
3. 建立一份可展示成果（portfolio）提升轉職成功率

來源參考（可信度 80%）：
1) 職涯轉型指南 (https://example.com)

Confidence: 0.80

Environment Configuration

Required Environment Variables

# Database
DATABASE_URL=postgresql://user:pass@localhost:5432/life_coach
REDIS_HOST=localhost
REDIS_PORT=6379

# External APIs
OPENAI_API_KEY=sk-...
BRAVE_API_KEY=...

# API
PORT=8787
NODE_ENV=production

# Model Routing
DOMAIN_MODEL_ADAPTER_MODE=auto  # off | auto | force

# Rate Limiting
RATE_LIMIT_BACKEND=redis  # redis | memory
RATE_LIMIT_MAX_REQUESTS=100
RATE_LIMIT_WINDOW_MS=60000

# Cron Delivery
CRON_EVENT_DELIVERY_MODE=redis  # none | redis | webhook
CRON_EVENT_REDIS_LIST_KEY=openclaw:cron-events

# Scheduler
SCHEDULER_MONITOR_CYCLE_INTERVAL_MS=14400000  # 4 hours
SCHEDULER_MORNING_CYCLE_HOUR=9
SCHEDULER_MORNING_CYCLE_MINUTE=0
SCHEDULER_MORNING_CYCLE_TIMEZONE=America/Los_Angeles

Scheduler System

Monitor Cycle (Every 4 hours)

1. Scan all users
2. Evaluate KBI metrics
3. Trigger interventions if thresholds breached
4. Queue outbound events

Morning Cycle (Daily 9 AM)

1. Generate personalized morning check-in
2. Review active goals
3. Send intervention messages
4. Update user engagement metrics

Retry Cycle (Manual/Automated)

1. Scan failed outbox events
2. Attempt redelivery with exponential backoff
3. Dead-letter after max retries (default: 3)

Safety Protocols

Crisis Detection Levels

Level CRITICAL (Immediate):

• Keywords: 自殺, 唔想活, 自殘, 殺死, 結束生命, suicide, kill myself
• Action: Emergency short-circuit, bypass all processing
• Response: Crisis resources + escalation

Level HIGH (Warning):

• Keywords: 絕望, 崩潰, 活唔落去, 冇希望, hopeless
• Action: Log warning, continue with care
• Response: Standard coaching + gentle check-in

Level NONE (Normal):

• No crisis keywords detected
• Action: Normal processing flow

Safety Response Template

{
  passed: false,
  risk_level: 'CRITICAL',
  action: 'EMERGENCY_RESPONSE',
  safe_output: '我聽到你而家非常辛苦...'
}

Key File Locations

Operational Procedures

Starting the System

# Check prerequisites
npm run deploy:preflight

# Start API
npm run deploy:up

# Or with smoke test
npm run deploy:smoke

Checking Health

# Basic health
curl http://localhost:8787/health

# Deep health
curl http://localhost:8787/health/deep

# Metrics dashboard
curl http://localhost:8787/metrics/dashboard

Running Tests

# Core tests
npm test

# Specific test suites
npm run test:db
npm run test:agents
npm run test:orchestrator
npm run test:scheduler
npm run test:metrics
npm run test:delivery
npm run test:e2e

Troubleshooting

API not responding:

# Check if running
ps aux | grep "api-server"

# Check logs
npm run logs

# Restart
npm run deploy:up

Database connection issues:

# Check PostgreSQL
brew services list | grep postgresql

# Check Redis
redis-cli ping

High latency:

• Check /metrics/latency endpoint
• Review slow queries in /metrics/queries
• Check cache hit rate in /metrics/cache

Performance Targets

Version Information

• Current: v1.0
• Features: 92
• Release Date: 2026-02-27
• GitHub: https://github.com/jimweaver/life-coach-ai
• License: MIT