Startclaw Optimizer

# OpenClaw Optimizer Skill

## Overview

The OpenClaw Optimizer is a comprehensive performance and cost optimization skill designed to enhance the efficiency of Clawdbot's subagent workflows.

## Core Components

### 1. Task Router
- Intelligent model selection based on task complexity
- Automatic routing between Haiku, Sonnet, and Opus models
- Cost prediction and optimization

### 2. Scheduler
- Robust task execution with retry mechanism
- Configurable preflight and postflight hooks
- Timeout and exponential backoff handling

### 3. Browser Governor
- Browser tab serialization
- Concurrent tab management
- Circuit breaker for preventing runaway processes

### 4. Context Compaction
- Automatic token tracking
- Summarization at 50,000 token threshold
- Preserves critical context

### 5. Real-time Dashboard
- Monitor daily budget
- Track task execution
- Visualize model distribution
- Circuit breaker status

## Expected Savings

**Before Optimization:** $90/day
**After Optimization:** $3-5/day
**Savings:** 95%

## Installation

```bash
npm install @startclaw/openclaw-optimizer
```

## Usage

```javascript
const { TaskRouter, OptimizerScheduler, BrowserGovernor } = require('@startclaw/openclaw-optimizer');

const router = new TaskRouter();
const scheduler = new OptimizerScheduler();
const browserGovernor = new BrowserGovernor();

// Automatic model and cost optimization
const modelSelection = router.selectModel(taskDescription);
await scheduler.execute(task, modelSelection);
```

## Monitoring

```bash
# Real-time dashboard
python3 scripts/dashboard.py watch
```

## License

StartClaw Internal Use License

# OpenClaw Optimizer: Context Compaction for Subagents

## Context Compaction Feature

### Key Capabilities
- Automatic token tracking across sessions
- Auto-summarization at 50,000 token threshold
- Preserves critical context and task intent
- Cost-effective summarization using Haiku model

### Usage Example

```javascript
const { SubagentContextCompactor } = require('./context-compaction');

// Initialize compactor
const compactor = new SubagentContextCompactor({
  tokenThreshold: 50000,
  summaryModel: 'claude-3-5-haiku'
});

// Before spawning subagent
const preparedContext = await compactor.prepareForSubagent(fullContext, sessionKey);
sessions_spawn({
  context: preparedContext,
  // other spawn parameters
});
```

### How It Works
1. Tracks token usage across main and subagent sessions
2. Detects when context approaches 50,000 tokens
3. Extracts and preserves critical information
4. Generates a concise summary using Haiku
5. Maintains session continuity

### Performance Impact
- Minimal overhead
- Reduces API costs by 70-90%
- Prevents runaway context growth

## Logging and Telemetry
Compaction events are logged with:
- Session Key
- Original Token Count
- Compacted Token Count
- Detected Key Themes

## Customization
- Adjust token threshold
- Modify critical information patterns
- Configure preservation of recent exchanges