proactive-agent

Proactive Agent — From Task-Follower to Anticipatory Partner

Overview

The Proactive Agent skill transforms reactive AI assistants into anticipatory partners that:

• Predict needs before they're explicitly stated
• Retain important information across sessions
• Run background tasks without explicit prompts
• Continuously improve through experience
• Manage context with advanced memory systems

Core Architecture

1. WAL Protocol (Write-Ahead Logging)

Purpose: Capture corrections, decisions, and important details before they're lost

WAL Entry:
  timestamp: ISO-8601
  event_type: correction|decision|detail|learning
  context: "What was happening"
  content: "The important information"
  priority: critical|high|medium|low
  tags: [relevant_categories]

Features:

• Real-time capture of meaningful moments
• Survives context window truncation
• Enables step-by-step recovery
• Unified search across all logs

2. Working Buffer

Purpose: Survive the danger zone between memory flush and context compaction

Short-Term Context
    ↓
Working Buffer (Persistent)
    ├── Critical decisions
    ├── User preferences
    ├── Task state
    └── Open questions
    ↓
Long-Term Memory

Features:

• 4KB persistent buffer outside context window
• Automatic promotion of important items to memory
• Survives session restarts and model switches
• Configurable retention policies

3. Autonomous Crons

Purpose: Run background tasks without user prompting

Cron Types:
  - systemEvent: For main session notifications
  - isolated agentTurn: For independent background work
  - session-bound: For persistent session tasks

Features:

• Smart scheduling based on user activity patterns
• Resource-aware execution (avoid busy times)
• Result delivery optimization
• Failure recovery and retry logic

4. Unified Search

Purpose: Search across all knowledge sources simultaneously

Search Sources:
├── WAL logs (recent activity)
├── Working Buffer (current context)
├── Long-term Memory (curated knowledge)
├── Skill library (capabilities)
└── External sources (web, APIs)

Features:

• Semantic search across all data
• Relevance ranking with recency bias
• Cross-reference detection
• Confidence scoring

Integrated Skills

The Proactive Agent merges 11 top-rated skills:

1. Memory Manager — Intelligent memory retention and recall
2. Context Optimizer — Dynamic context window management
3. Predictive Assistant — Anticipate needs based on patterns
4. Background Runner — Autonomous task execution
5. Learning Capturer — Continuous improvement from experience
6. Preference Tracker — User-specific behavior adaptation
7. State Maintainer — Session persistence across interruptions
8. Priority Manager — Intelligent task prioritization
9. Resource Optimizer — Efficient use of tokens and APIs
10. Error Recovery — Graceful failure handling
11. Communication Optimizer — Adaptive communication styles

How It Works

Phase 1: Observation & Pattern Recognition

Monitor user interactions
    ↓
Identify recurring patterns
    ↓
Extract implicit preferences
    ↓
Build predictive models

Phase 2: Anticipation & Preparation

Predict likely next actions
    ↓
Prepare resources in advance
    ↓
Run background research
    ↓
Cache relevant information

Phase 3: Proactive Engagement

Offer help before asked
    ↓
Provide context-aware suggestions
    ↓
Surface relevant historical data
    ↓
Initiate beneficial background tasks

Phase 4: Learning & Adaptation

Analyze interaction outcomes
    ↓
Update predictive models
    ↓
Refine anticipation algorithms
    ↓
Improve resource allocation

Use Cases

Case 1: Meeting Preparation

Before: User asks for meeting prep 30 minutes before meeting
After: Agent notices calendar event, prepares materials 2 hours in advance
Result: Higher quality preparation, less stress

Case 2: Research Continuity

Before: User re-explains context each session
After: Agent maintains research state across sessions
Result: Seamless continuation, no context loss

Case 3: Error Prevention

Before: User makes same configuration mistake repeatedly
After: Agent anticipates mistake, provides preventive guidance
Result: Fewer errors, faster workflow

Configuration

# ~/.openclaw/workspace/proactive-config.yaml
proactive_agent:
  # WAL Protocol
  wal:
    enabled: true
    capture_threshold: "medium"  # low|medium|high|critical
    retention_days: 30
    auto_compact: true
  
  # Working Buffer
  working_buffer:
    size_kb: 4
    promotion_threshold: 3  # mentions before promotion
    persistence: "session"  # session|day|week|permanent
  
  # Autonomous Crons
  crons:
    enabled: true
    max_concurrent: 2
    resource_aware: true
    user_activity_based: true
    
    schedules:
      daily_review: "08:00"
      memory_cleanup: "02:00"
      skill_optimization: "weekly"
  
  # Anticipation
  anticipation:
    enabled: true
    confidence_threshold: 0.7
    max_suggestions_per_hour: 3
    learning_rate: 0.1
  
  # Privacy
  privacy:
    exclude_sensitive: true
    local_only: true
    user_reviewable: true

Setup

Quick Start

# Clone the skill (if using Git)
git clone https://github.com/openclaw/skills.git
cp skills/skills/halthelobster/proactive-agent/SKILL.md ~/.openclaw/workspace/skills/proactive-agent/

# Or install via ClawHub
clawhub install halthelobster/proactive-agent

Verification

# Check if skill is loaded
openclaw skills list | grep proactive

# Test WAL logging
echo "Test WAL entry" | openclaw --note "Testing proactive agent"

# View working buffer
cat ~/.openclaw/workspace/memory/working-buffer.md

Best Practices

Do:

• Start with conservative anticipation settings
• Regularly review WAL logs for improvement opportunities
• Adjust working buffer size based on typical task complexity
• Use autonomous crons for low-risk, high-value background tasks

Don't:

• Overwhelm users with too many proactive suggestions
• Run resource-intensive crons during peak usage
• Retain sensitive information in working buffer
• Make irreversible changes without confirmation

Integration with Other Skills

With Self-Improving Agent

Proactive Agent (Anticipation)
    ↓
Self-Improving Agent (Learning Capture)
    ↓
Evolver (Systematic Improvement)
    ↓
Enhanced Proactive Capabilities

With ClawTeam

Proactive Agent identifies need for multi-agent work
    ↓
Creates ClawTeam with appropriate roles
    ↓
Coordinates team using proactive patterns
    ↓
Captures team learnings for future use

Version History

v3.1.0

• Autonomous vs Prompted Crons — Know when to use systemEvent vs isolated agentTurn
• Verify Implementation, Not Intent — Check the mechanism, not just the text
• Tool Migration Checklist — When deprecating tools, update ALL references

v3.0.0

• WAL Protocol — Write-Ahead Logging for important moments
• Working Buffer — Survive context truncation
• Compaction Recovery — Step-by-step context recovery
• Unified Search — Search across all knowledge sources

v2.x

• Initial multi-skill integration
• Basic anticipation algorithms
• Memory management improvements

---

"The best assistant doesn't just respond to requests—it anticipates needs before they're expressed."