officialerictm/clawdbot-auto-compress

Auto-Compress Context Skill

Automatically compresses conversation context when approaching token limits, similar to Claude.ai's "Hold on, compressing our conversation" feature.

Overview

Monitors context usage and automatically triggers compression at 80% capacity (160k of 200k tokens), allowing conversations to continue indefinitely without manual resets.

Requirements

• Clawdbot with session access
• Python 3.8+
• Node.js (for hook integration)

How It Works

1. Monitors context usage in real-time
2. At 80% threshold, automatically compresses conversation
3. Preserves critical information while removing verbose content
4. Seamlessly resumes conversation with compressed history

Installation

1. Install the monitoring hook:

cd /home/ubuntu/clawd/skills/auto-compress
chmod +x install.sh
./install.sh

2. Add to your Clawdbot config:

agent:
  hooks:
    afterResponse:
      - /home/ubuntu/clawd/skills/auto-compress/context-monitor.js

Usage

Once installed, it works automatically. You'll see:

• 📊 Context usage in status messages
• ⚠️ Warning at 70% capacity
• 🗜️ Automatic compression at 80%
• ✅ Confirmation when compression completes

Configuration

Edit config.json to customize:

{
  "threshold": 0.8,        // Trigger at 80% capacity
  "warningThreshold": 0.7, // Warn at 70% capacity
  "preserveRecent": 20,    // Keep last 20 messages
  "compressTools": true,   // Compress tool outputs
  "notifyUser": true       // Show compression message
}

Manual Commands

• compress-now - Force compression immediately
• compress-status - Show compression statistics
• compress-disable - Temporarily disable auto-compression

Compression Strategy

1. Preserve: Recent messages, user requests, important results
2. Compress: Tool outputs, repetitive content, old context
3. Summarize: Long conversations into key points
4. Archive: Full history to session transcript

Example

When compression triggers, you'll see:

🗜️ Hold on, I need to compress our conversation to continue...
[Compressing 160k → 40k tokens]
✅ Compression complete! We can continue our conversation.

The conversation then continues seamlessly with all important context preserved.