An intelligent orchestration protocol that automates model routing and output compression to achieve massive token savings.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install token-saver-75plus
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 token-saver-75plus 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 Saver 75+?
Token Saver 75+ is a sophisticated model routing and optimization protocol designed for developers using Openclaw Skills who need to maintain high performance while drastically reducing operational costs. By implementing a tiered classification system, the skill ensures that expensive models are only used for complex reasoning, while delegating routine tasks like code generation, summarization, and formatting to specialized or free model endpoints.
This skill operates on the principle of understanding fully but executing cheaply. It acts as an intelligent traffic controller within your agentic workflow, using session spawning to distribute workloads across a diverse model fleet including Groq, Codex, and Claude Opus. Beyond routing, it enforces rigorous output compression rules to ensure that every token generated provides maximum value, eliminating filler and unnecessary narration.
Token Saver 75+ Use Cases
- Reducing API overhead for large-scale code refactoring by routing all generation tasks to specialized Codex models.
- Processing high-volume text analysis and formatting using free inference tiers to lower total cost of ownership.
- Managing complex multi-agent strategies where high-reasoning models like Opus only handle top-level orchestration while spawning cheaper agents for subtasks.
- Enforcing concise, technical communication in developer environments to minimize distracting AI chatter.
- Implementing automated failure recovery by escalating tasks to more capable models only when cheaper attempts fail.
How Token Saver 75+ Works
- The Request Classifier analyzes every incoming message to determine if it falls into Tier 1 (trivial), Tier 2 (bulk), Tier 3 (structured/code), or Tier 4 (strategic).
- Tool Gating logic checks if the request can be fulfilled using existing memory or if multiple tool calls can be batched to save tokens.
- The orchestrator determines the cheapest capable model and utilizes the sessions_spawn command to delegate the execution layer.
- For code-specific tasks, the system automatically spawns a Codex instance, ensuring the primary orchestrator remains focused on the high-level goal.
- All model outputs are filtered through compression templates (like CAUSE-FIX-VERIFY) to ensure the response stays within the allocated token budget for its tier.
- A measurement trailer is optionally appended to provide transparency on token usage and the specific routing path taken.
Token Saver 75+ Setup
To integrate this protocol into your Openclaw Skills workflow, you must define your model routing table and ensure your API credentials for various providers are active.
# Install the protocol into your agent directory
openclaw install token-saver-75plus
# Configure your model endpoints
export GROQ_API_KEY="your_key"
export OPENAI_API_KEY="your_key"
export ANTHROPIC_API_KEY="your_key"
Token Saver 75+ Data Schema & Taxonomy
The protocol utilizes a structured classification and routing schema to manage the lifecycle of a request. It tracks the following metadata for every interaction:
| Attribute | Description |
|---|---|
| Request Tier | T1 to T4 classification based on complexity and reasoning requirements. |
| Spawn Task | The specific, context-rich instruction string passed to sub-agents. |
| Routing Map | The sequence of models used (e.g., Llama-3 for bulk, Codex for implementation). |
| Token Budget | The maximum allowed output length based on the identified tier. |
| Measurement | The final token count and efficiency delta reported at the end of the session. |
Token Saver 75+ Advanced Features
- Dynamic Session Spawning: Orchestrates sub-tasks by spawning isolated agent sessions with full context but no history overhead.
- Multi-Tier Escalation: Automatically retries failed executions by escalating from T2/T3 models to T4 models if a failure is detected.
- Output Compression Templates: Forces models to use high-density formats like 3-bullet summaries and one-liner status updates.
- Tool Gating Intelligence: Evaluates the necessity of a tool call before execution to prevent redundant web searches or file reads.
- Silent Orchestration: Performs tier classification and model selection in the background without requiring user intervention or explicit configuration per message.
SKILL.md
Loading
Related Openclaw Skills
Luma Event Manager
An automated event management tool for Luma that enables discovery, RSVPs, and calendar synchronization via intelligent web scraping.
gitclaw
An automated backup tool that mirrors your OpenClaw workspace to a GitHub repository using a cron-driven sync script.
eToro Trading & Market Data
Enables AI agents to perform automated trading, portfolio management, and social sentiment analysis via the eToro Public API.
eToro API
A comprehensive interface for programmatically managing eToro portfolios, executing trades, and accessing real-time market data.
TonScan Wallet Balance
A zero-config tool for querying TON blockchain wallet balances, statuses, and transaction activity via the TonScan API.
Bandwidth Income
Automate the deployment and management of bandwidth-sharing nodes to generate passive crypto income from your idle internet connection.
Featured*
