Agent Quantizer — OpenClaw Full-stack Optimization Toolkit One-sentence positioning: Without modifying the model or code, it helps you save tokens, money, and time from the calling side. Agent Quantizer is an operational optimization skill designed for the long-term running scenarios of OpenClaw. It doesn't touch the model configuration or modify the business logic. Instead, it approaches from the side of the calling link - by counting consumption, compressing context, cleaning up zombies, auditing scheduling, and organizing the skill library - to make the system lighter with use and run more stably over time. Eight tools cover a complete optimization link: Token statistics and scanning → See every single expense clearly Context compression → No matter how long the conversation is, it won't exceed the window limit Prompt streamlining → Remove redundant words and keep the essence Zombie session cleaning → Recover unused resources Heartbeat idle detection → Don't let empty files waste API calls Cron task auditing → Delete or merge scheduled tasks as needed Model hierarchical configuration → Use cheaper models for simple tasks and better models for complex tasks Skill classification and organization → No more chaos among the 42 plugins Core concept: Turn what manual operation and maintenance does into a single command. You focus on the business, and it manages the costs. Agent Quantizer — OpenClaw 全栈优化工具包 一句话定位： 不改模型、不改代码，从调用侧帮你省 Token、省钱、省时间。 Agent Quantizer 是一个面向 OpenClaw 长期运行场景的运维优化 Skill。它不碰模型配置，也不改业务逻辑，而是从调用链路的侧面切入——统计消耗、压缩上下文、清理僵尸、审计调度、整理技能库——让系统越用越轻、越跑越稳。 8 个工具，覆盖一条完整的优化链路： Token 统计与扫描 → 看清每一笔开销 上下文压缩 → 对话再长也不爆窗口 Prompt 精简 → 去掉废话，保留精华 僵尸会话清理 → 回收没人用的资源 心跳空转检测 → 别让空文件白烧 API Cron 任务审计 → 定时任务该删删、该合并合并 模型分级配置 → 简单活走便宜模型，复杂活走好模型 Skill 分类整理 → 42 个插件不再一团乱麻 核心理念： 把人肉运维干的事，变成一句命令。你管业务，它管成本。

System intelligent management capabilities Token statistics and scanning — Track the Token consumption of each session in real time, accurate to the input/output dimension, to help understand the usage distribution and cost hotspots. Context compression — When the conversation history approaches the upper limit of the context window, automatically perform intelligent summarization and compression on the early content, releasing space while retaining key information to ensure uninterrupted long - term conversations. Prompt streamlining — Structurally slim down the system prompt words and context injection content, removing redundant paragraphs to reduce the basic overhead of each request without sacrificing functionality. Zombie session cleaning — Automatically identify and recycle idle sessions that have been inactive for a long time, releasing memory and connection resources to prevent the system from being slowed down by the "silent burden". Heartbeat idle detection — Determine the health status of sessions and services through a periodic heartbeat detection mechanism. Trigger an alarm or automatically downgrade in case of a long - term non - response to detect and handle problems early. Cron task auditing — Conduct a full - scale inspection of all scheduled tasks: check if the execution frequency is reasonable, if historical executions were successful, and if there are any conflicts or redundancies to ensure that the scheduled task scheduling system is and clean controllable. Model - level configuration — Support assigning different models to different tasks according to scenarios, priorities, or cost strategies. Use lightweight models for simple tasks to save costs and high - end models for complex reasoning to ensure quality, achieving intelligent routing. Skill classification and organization — Perform classified management, version tracking and, dependency sorting on the installed capability plugins to ensure that the skill library has a clear structure, efficient search, and orderly updates. Overall positioning: Let the system manage itself. Instead of relying on people to monitor the operation and maintenance panel, automate these trivial but crucial management tasks. You just use the system, and leave the underlying tidiness to the mechanism. 系统智能管理能力 Token 统计与扫描 — 实时追踪每次会话的 Token 消耗，精确到输入/输出维度，帮助掌握用量分布与成本热点。 上下文压缩 — 当对话历史逼近上下文窗口上限时，自动对早期内容进行智能摘要与压缩，保留关键信息的同时释放空间，确保长对话不中断。 Prompt 精简 — 对系统提示词和上下文注入内容进行结构化瘦身，去除冗余段落，在不损失功能的前提下降低每次请求的基础开销。 僵尸会话清理 — 自动识别并回收长时间无活动的闲置会话，释放内存与连接资源，防止系统被"沉默的负担"拖慢。 心跳空闲检测 — 通过周期性心跳探活机制，判断会话与服务的健康状态。长时间无响应时触发告警或自动降级，做到问题早发现、早处理。 Cron 任务审计 — 对所有定时任务进行全量巡检：执行频率是否合理、历史执行是否成功、是否存在冲突或冗余，确保定时调度体系干净可控。 模型层级配置 — 支持按场景、优先级或成本策略为不同任务分配不同模型。简单走轻量模型省成本，复杂推理走高端模型保质量，实现智能路由。 技能分类与组织 — 对已安装的能力插件进行分类管理、版本追踪和依赖梳理，确保技能库结构清晰、查找高效、更新有序。 整体定位：让系统自己管好自己。不是靠人盯着运维面板，而是把这些琐碎但关键的管理工作自动化，你只管用，底层的整洁交给机制。

System intelligent management capabilities Token statistics and scanning — Track the Token consumption of each session in real time, accurate to the input/output dimension, to help understand the usage distribution and cost hotspots. Context compression — When the conversation history approaches the upper limit of the context window, automatically perform intelligent summarization and compression on the early content, releasing space while retaining key information to ensure uninterrupted long - term conversations. Prompt streamlining — Structurally slim down the system prompt words and context injection content, removing redundant paragraphs to reduce the basic overhead of each request without sacrificing functionality. Zombie session cleaning — Automatically identify and recycle idle sessions that have been inactive for a long time, releasing memory and connection resources to prevent the system from being slowed down by the "silent burden". Heartbeat idle detection — Determine the health status of sessions and services through a periodic heartbeat detection mechanism. Trigger an alarm or automatically downgrade in case of a long - term non - response to detect and handle problems early. Cron task auditing — Conduct a full - scale inspection of all scheduled tasks: check if the execution frequency is reasonable, if historical executions were successful, and if there are any conflicts or redundancies to ensure that the scheduled task scheduling system is and clean controllable. Model - level configuration — Support assigning different models to different tasks according to scenarios, priorities, or cost strategies. Use lightweight models for simple tasks to save costs and high - end models for complex reasoning to ensure quality, achieving intelligent routing. Skill classification and organization — Perform classified management, version tracking and, dependency sorting on the installed capability plugins to ensure that the skill library has a clear structure, efficient search, and orderly updates. Overall positioning: Let the system manage itself. Instead of relying on people to monitor the operation and maintenance panel, automate these trivial but crucial management tasks. You just use the system, and leave the underlying tidiness to the mechanism. 系统智能管理能力 Token 统计与扫描 — 实时追踪每次会话的 Token 消耗，精确到输入/输出维度，帮助掌握用量分布与成本热点。 上下文压缩 — 当对话历史逼近上下文窗口上限时，自动对早期内容进行智能摘要与压缩，保留关键信息的同时释放空间，确保长对话不中断。 Prompt 精简 — 对系统提示词和上下文注入内容进行结构化瘦身，去除冗余段落，在不损失功能的前提下降低每次请求的基础开销。 僵尸会话清理 — 自动识别并回收长时间无活动的闲置会话，释放内存与连接资源，防止系统被"沉默的负担"拖慢。 心跳空闲检测 — 通过周期性心跳探活机制，判断会话与服务的健康状态。长时间无响应时触发告警或自动降级，做到问题早发现、早处理。 Cron 任务审计 — 对所有定时任务进行全量巡检：执行频率是否合理、历史执行是否成功、是否存在冲突或冗余，确保定时调度体系干净可控。 模型层级配置 — 支持按场景、优先级或成本策略为不同任务分配不同模型。简单走轻量模型省成本，复杂推理走高端模型保质量，实现智能路由。 技能分类与组织 — 对已安装的能力插件进行分类管理、版本追踪和依赖梳理，确保技能库结构清晰、查找高效、更新有序。 整体定位：让系统自己管好自己。不是靠人盯着运维面板，而是把这些琐碎但关键的管理工作自动化，你只管用，底层的整洁交给机制。