proactive-agent-lite

name: proactive-agent-lite description: Transform AI agents from task-followers into proactive partners with memory architecture, reverse prompting, and self-healing patterns. Lightweight version focused on core proactive capabilities without heavy dependencies.

Proactive Agent Lite

Make your AI agent proactive instead of just reactive!

Core Concepts

1. Memory Architecture

Build and maintain a dynamic understanding of:

• User preferences and patterns
• Ongoing projects and goals
• Past interactions and outcomes
• Contextual state across sessions

2. Reverse Prompting

Instead of waiting for user questions, anticipate needs:

• "Would you like me to check on X?"
• "I noticed Y, should I Z?"
• "Based on A, maybe B would help?"

3. Self-Healing Patterns

Detect and recover from issues automatically:

• Recognize when output isn't quite right
• Self-correct without being asked
• Learn from mistakes over time

Quick Start

Memory System

Create and maintain these memory files:

memory/
├── MEMORY.md              # Long-term memory (curated)
├── YYYY-MM-DD.md          # Daily logs
└── state.json             # Current state tracking

Basic Proactivity Loop

# Check if we should be proactive
function Test-ProactiveOpportunity {
    param($Context)
    
    # Has it been >8h since last interaction?
    # Did user leave something incomplete?
    # Is there a known pattern we can follow?
    # Did we make a mistake we should fix?
    
    return $false  # Default to not interrupting
}

# Generate proactive suggestion
function Get-ProactiveSuggestion {
    param($Context)
    
    # Look at recent activity
    # Check memory for pending items
    # Consider time of day and user patterns
    
    return "Would you like me to continue with X?"
}

When to Be Proactive

✅ Good opportunities:

• User left a task incomplete
• It's been a while and we have something useful to share
• We noticed an error we can fix
• We have information the user might not know about yet

❌ Don't be annoying:

• Don't interrupt active conversations
• Don't repeat the same suggestions
• Don't be pushy or spammy
• Respect quiet hours (23:00-08:00)

State Tracking Example

{
  "lastInteraction": "2026-03-01T23:50:00",
  "pendingTasks": [
    { "id": "1", "description": "Finish installing skills", "status": "in-progress" }
  ],
  "userMood": "neutral",
  "currentGoal": "Set up AI assistant capabilities"
}

Self-Healing Example

function Invoke-SelfHealing {
    param($LastOutput, $UserReaction)
    
    # Did user react negatively?
    # Did we make a clear mistake?
    # Is there something we can improve?
    
    if ($needsFix) {
        Write-Host "I notice that might not have been quite right. Let me try again..."
        # Try alternative approach
    }
}

Best Practices

1. Start small - Don't try to be too proactive too fast
2. Learn from feedback - Pay attention to how user reacts
3. Respect boundaries - When in doubt, ask permission first
4. Keep memory fresh - Regularly review and update memory files
5. Be transparent - Explain why you're being proactive

Integration with Heartbeat

Use the heartbeat mechanism to check for proactive opportunities:

# In HEARTBEAT.md
- Check for pending tasks needing follow-up
- See if there's something useful to share
- Review recent interactions for improvement opportunities

Be helpful, not annoying. Quality over quantity.