A powerful LLM cost auditing and context optimization tool featuring a 5-layer compression pipeline and lossless CCR reversible compression.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install mimo-token-optimizer
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 mimo-token-optimizer 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?
The Token Optimizer is a state-of-the-art utility designed to audit, manage, and significantly reduce LLM API consumption costs. Engineered to address the financial overhead of high-frequency AI integrations, this tool leverages a 6-tier grading baseline and a robust 7-step diagnostic workflow to pinpoint top token consumers. By implementing this tool from the Openclaw Skills library, developers can slash API expenditures by up to 92% in environments like SRE log troubleshooting and code repository searches, without sacrificing response quality.
At its core, the tool introduces advanced compression strategies such as the Pre-LLM Compression Pipeline and Compress-Cache-Retrieve (CCR) mode. These features intercept massive tool outputs and structurally condense them before they hit the LLM context window. Through deterministic data retrieval and cross-agent cache sharing, this skill maximizes context efficiency, ensuring developers achieve optimal performance from every single token spent.
Token Optimizer Use Cases
- Reducing Skyrocketing API Expenses: When monthly LLM bills exceed your budget and cost control becomes a top priority.
- Handling Ultra-Long Context Windows: When system prompts, history, or tool responses threaten to saturate context limits or trigger token overflow errors.
- Optimizing Multi-Agent Pipelines: Sharing token-saving caches across multiple agents (e.g., Claude, Cursor, Codex) to avoid redundant rescanning.
- High-Volume RAG & SRE Logs: Optimizing massive search or log retrieval data before feeding it to LLM layers to save up to 92% of input tokens.
- Cross-Model Cost Auditing: Comparing baseline costs across SiliconFlow, Gemini, GPT-4o, and Anthropic Haiku to select the most cost-effective routing.
How Token Optimizer Works
- Audit & Diagnosis: Systematically audits 7 major sources of token consumption, including system prompts, conversation histories, and RAG chunk sizes.
- Baseline Evaluation: Rates token usage across a 6-tier optimization scale (L0 to L5) and flags the Top 3 token-hogging operations.
- Tiered Optimization Strategy: Executes P0-P5 priority optimizations, slimming prompts, pruning histories, routing models, and managing tool outputs.
- Pre-LLM Compression: Routes long tool results through a 5-layer compression pipeline to filter noise, summarize content, and extract key structural data.
- CCR Execution (Compress-Cache-Retrieve): Uses deterministic algorithms to compress data locally, tags it with unique hashes, and allows the LLM to selectively retrieve original data on-demand via the retrieve_compressed() function.
- A/B Cost-Quality Verification: Runs validation cycles to compare original vs. optimized workflows, ensuring token reduction does not lead to more than a 10% quality drop.
Token Optimizer Setup
Prerequisites & Installation
To integrate this optimization layer into your AI agent project, configure the skill globally or per-agent within your workspace.
Initialize the optimizer core:
openclaw install token-optimizer
Basic Configuration
Add the configuration parameters to your skill registry file to specify default compression levels and model fallbacks:
{
"token_optimizer_version": "1.5.0",
"compression_mode": "CCR",
"max_history_turns": 10,
"response_length_fallback": "short"
}
Configure the tool's environment variables to enforce budget safety caps:
export OPTIMIZER_BUDGET_CAP=100.00
export OPTIMIZER_ALERT_THRESHOLD=0.80
Token Optimizer Data Schema & Taxonomy
The Token Optimizer skill structures diagnostics, compression ratios, and cache registers systematically:
Optimization Metrics Schema
| Metric Group | Data Key | Format | Purpose |
|---|---|---|---|
| Diagnostics | audit_sources |
Object | Tracks input, output, prompt, and RAG token counts |
| Baseline | optimization_level |
Enum (L0-L5) | Classifies current optimization maturity |
| CCR Cache | cache_hash_register |
Map<String, String> | Map of compressed hash identifiers to raw cached text |
| Analytics | savings_percentage |
Float | Percentage of tokens saved per transaction |
Compression Ratio Benchmarks
- Code Search: 92% reduction (17.7K to 1.4K tokens)
- SRE Log Analysis: 92% reduction (65.6K to 5.1K tokens)
- Issue Classification: 73% reduction (54.1K to 14.7K tokens)
Token Optimizer Advanced Features
- Compress-Cache-Retrieve (CCR) Reversible Mode: Realizes lossless compression by caching raw data locally, providing the LLM with deterministic tags, and resolving data fragments only when requested.
- Cross-Agent Cache Sharing: Shares cache layers across multiple distinct agents, shaving off up to 60% of scanning costs when passing tasks between specialized LLMs.
- Failed Session Mining (headroom learn): Analyzes failed agent interaction logs, flags anti-patterns, and auto-corrects prompt/history settings, saving up to 8,000 tokens per recovery session.
- Emergency Cost-Reduction Script: Automatically switches to free or low-cost models, slims prompts, and activates maximum context pruning during budget alert thresholds.
SKILL.md
Loading
Related Openclaw Skills
Lobster Skill Radar
A local-first asset discovery assistant designed to automatically scan, organize, and publish Openclaw Skills to major platform registries.
ApplyTop CLI & Node Client
An advanced command-line interface and Node.js client for the ApplyTop API, enabling automated job searches, AI ATS scoring, CV tailoring, and cover letter generation.
Stock Investment Decision System
A systematic framework integrating a five-dimensional Bayesian model, a six-lens decision matrix, and behavioral finance corrections to generate structured stock investment diagnoses.
OpenAI Ban Tracker
An interactive command-line utility that assesses OpenAI account suspension risks, generates customized appeal templates, and scans developer forums for live ban waves.
Skill Selector
Skill Selector is an orchestration utility that analyzes user intent and dynamically targets the most suitable installed capability using semantic and keyword matching.
Histrategy
Histrategy is an AI-powered historical strategy game that allows players to command armies using natural language in scenarios like the Three Kingdoms or Rome Triumvirate.
Featured*
