A high-performance strategy set for reducing token burn and optimizing context windows in AI agentic systems.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install lean-context
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install lean-context using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Token Optimizer?
Token Optimizer provides a systematic approach to minimizing the token footprint of AI agents like Claude Code and OpenClaw. By addressing the primary causes of token drain—such as tool output accumulation, context compounding, and oversized system prompts—this skill ensures that your Openclaw Skills remain fast and cost-effective without sacrificing reasoning quality. It focuses on precision context engineering to keep sessions lean and effective.
This skill is essential for developers running production-grade agents where long-running sessions or large codebases can lead to exponential cost increases. It provides the frameworks for selective loading, compaction, and efficient tool definitions that directly impact the bottom line of AI operations.
Token Optimizer Use Cases
- Reducing AI API costs and token burn in production environments.
- Optimizing CLAUDE.md or AGENTS.md configurations for minimal context window usage.
- Designing token-efficient multi-agent architectures that isolate research from implementation.
- Auditing existing project configurations for context bloat and unnecessary metadata.
- Managing high-turn coding sessions where conversation history compounds costs.
How Token Optimizer Works
- Analyzes the baseline system prompt to identify and remove redundant instructions, targeting a baseline under 500 tokens.
- Implements progressive disclosure by replacing inline documentation with file pointers that the agent only reads when necessary.
- Scopes tool outputs using CLI utilities like grep, jq, and head to prevent massive data dumps from entering the context history.
- Applies compaction and session hygiene strategies to summarize and reset the context window between unrelated tasks.
- Utilizes prompt caching and sub-agent architectures to ensure that heavy data exploration doesn't pollute the main agent's memory.
Token Optimizer Setup
To implement these strategies within your Openclaw Skills, start by auditing your current configuration files and scoping your tool outputs.
# Audit your current prompt size to establish a baseline
wc -w CLAUDE.md
# Implement scoped tool outputs in your workflow
grep "ERROR" logs/app.log | head -n 10
# Use jq to extract specific fields instead of loading full JSON blobs
cat data.json | jq '.items[0:5]'
Token Optimizer Data Schema & Taxonomy
The skill organizes its data management into a three-tier hierarchy to ensure maximum efficiency within Openclaw Skills.
| Tier | Storage Location | Token Strategy |
|---|---|---|
| Metadata | SKILL.md description | Always loaded for trigger recognition (minimal) |
| Core Logic | SKILL.md body | Loaded only when the skill is explicitly triggered |
| References | references/*.md | Loaded on-demand via file pointers |
| Memory | memory/*.md | Curated cross-session continuity without raw history |
Token Optimizer Advanced Features
- LLMLingua-style token-level compression for high-volume narrative documents.
- Multi-tier model routing (Haiku for lookups, Sonnet for logic) to optimize the cost-per-token ratio.
- Automated prompt caching for static prefixes to take advantage of provider-level optimizations.
- Sub-agent delegation patterns where exploration agents return only distilled summaries to the main controller.
- Hash-check caching to prevent the agent from re-reading unchanged files across different sessions.
SKILL.md
Loading
Related Openclaw Skills
Content Repurpose
An AI-powered utility to transform a single piece of long-form content into a comprehensive multi-platform distribution pack while maintaining original voice and intent.
Deep Article Analysis
A sophisticated analysis framework that applies over 10 cognitive thinking models to extract core viewpoints and deep insights from any article.
SEO Blog Writer (aim-blog-write)
A comprehensive AI skill for generating SEO-ready blog posts with automated image embedding and professional Word document export.
Live Streaming Sales Script Generator
A professional AI-driven tool designed to generate high-conversion scripts for live streaming e-commerce across multiple platforms.
Digital Human Video Generator
A sophisticated AI skill that transforms static images into synchronized talking digital human videos using audio files or text-to-speech.
Foreign Trade News Query
An AI agent skill that fetches and summarizes international trade news, import/export policies, and market dynamics from the last 1 to 3 days.
Featured*
