← Back to Skills Marketplace

熵管理系统

Name: 熵管理系统
Author: 534422530

by 534422530 · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ✓ Security Clean

Downloads

Stars

Active Installs

Versions

Install in OpenClaw

/install entropy-manager

Description

熵管理系统 - 原创技能。用于管理和控制AI会话中的熵（无序度），包括上下文精简、状态重置、注意力聚焦等功能。适用于长时间会话、复杂任务、多步骤工作流等场景。

README (SKILL.md)

⚠️ 发布规则

所有发布到ClawHub的技能必须严格测试，确定没有问题再发布

技能测试验证清单

frontmatter格式正确
功能原创且实用
工作流程完整
触发条件明确
无语法错误

Entropy Manager - 熵管理系统

原创技能 | 激活词: 熵管理 / 精简上下文 / 重置状态

核心概念

什么是熵 (Entropy)?

熵 = 无序度/不确定性

在AI会话中：

上下文熵: 历史记录过多导致注意力分散
状态熵: 当前状态不清晰导致错误决策
任务熵: 目标不明确导致方向迷失

熵的症状

回复开始重复或跑题
忘记之前的决定
越来越难聚焦核心问题
输出质量下降

熵管理四大策略

1. 上下文精简 (Context Pruning)

当上下文超过阈值时触发：

保留核心决策和结论
删除中间过程和废话
压缩相似对话为摘要

触发条件: 对话超过20轮 或 上下文超过80K tokens
执行动作: 生成摘要，替换原始对话

2. 状态重置 (State Reset)

当状态混乱时触发：

明确当前任务目标
列出已完成的部分
确定下一步行动

触发条件: 任务切换 / 迷失方向 / 错误累积
执行动作: 生成状态卡片，聚焦核心

3. 注意力聚焦 (Attention Focus)

当注意力分散时触发：

识别当前核心问题
排除干扰项
设定明确边界

触发条件: 同时处理多��问题 / 任务过于复杂
执行动作: 分解任务，一次只做一件

4. 熵预警 (Entropy Alert)

持续监控熵值变化：

上下文增长速率
状态一致性
目标清晰度

指标: 
- ctx_rate: 上下文增长速度
- state_coherence: 状态一致性 (0-1)
- goal_clarity: 目标清晰度 (0-1)

熵管理流程

1. 检测 → 监控熵值指标
2. 预警 → 达到阈值时提醒
3. 精简 → 执行上下文压缩
4. 重置 → 状态聚焦
5. 验证 → 确保熵值降低

输出格式

熵状态报告

## 熵状态报告

### 当前指标
- 上下文长度: XXX tokens
- 会话轮数: XX轮
- 状态一致性: X.X
- 目标清晰度: X.X

### 熵等级: 🟢低 / 🟡中 / 🔴高

### 建议操作
1. [ ] 精简上下文
2. [ ] 重置状态
3. [ ] 聚焦任务

### 执行结果
[执行后更新指标]

应用场景

长时间会话 - 防止上下文无限增长
复杂多步骤任务 - 保持状态清晰
任务切换 - 快速重置注意力
错误恢复 - 从混乱状态中脱离

与Karpathy法则的结合

熵管理与Karpathy法则完美互补：

Karpathy法则	熵管理对应
先思考	降低决策熵
保持简单	降低复杂度熵
精准修改	降低改动熵
目标驱动	消除目标熵

原创性声明

本技能为原创，融合了：

热力学熵概念
系统论状态管理
认知科学注意力理论
AI会话优化实践

作者: laosi 创建日期: 2026-04-28

Usage Guidance

This skill appears coherent and low-risk from a platform perspective because it is instruction-only and asks for no credentials. Before installing: (1) test it on non-critical conversations to see how it prunes and to ensure important details aren't lost; (2) require explicit confirmation or keep a reversible archive before any automatic deletion/replacement of conversation history; (3) consider adding explicit thresholds and review rules (what counts as "核心决策" vs "废话") so the agent's deletions/summaries match your expectations; (4) monitor behavior in early use to ensure summaries are accurate and no important context is discarded.

Capability Analysis

Type: OpenClaw Skill Name: entropy-manager Version: 1.0.0 The 'entropy-manager' skill is a purely instructional Markdown-based bundle (SKILL.md) designed to help an AI agent manage its conversation context and focus. It contains no executable code, scripts, or system-level commands, and lacks any indicators of data exfiltration, malicious execution, or prompt injection intended to harm the user or system.

Capability Assessment

✓ Purpose & Capability

The name/description (entropy management for AI conversations) aligns with the instructions (context pruning, state reset, attention focus, alerts). There are no unexpected requirements (no env vars, no binaries, no external endpoints).

ℹ Instruction Scope

SKILL.md stays within the skill's domain (monitoring session metrics, summarizing/replacing context, producing status cards). However many actions are vaguely specified and grant the agent discretionary power (e.g., "删除中间过程和废话" / delete intermediate steps, "替换原始对话" / replace original dialog). That could lead to unintended loss of context if not constrained or confirmed by the user.

✓ Install Mechanism

No install spec and no code files (instruction-only). Lowest-risk category: nothing is written to disk and nothing is downloaded or executed beyond the agent following the prose.

✓ Credentials

The skill requests no environment variables, credentials, or config paths. No disproportionate access to secrets or unrelated systems.

✓ Persistence & Privilege

always is false and the skill does not request persistent privileges or modify other skills' configs. Autonomous invocation is allowed by default on the platform (not itself a security finding here).

How to Use

Make sure OpenClaw is installed (local or Docker)
Run the install command in chat: /install entropy-manager
After installation, invoke the skill by name or use /entropy-manager
Provide required inputs per the skill's parameter spec and get structured output

Version History

v1.0.0

原创技能 - AI会话熵管理，融合热力学、系统论和认知科学

Metadata

Slug entropy-manager

Version 1.0.0

License MIT-0

All-time Installs 0

Active Installs 0

Total Versions 1

Frequently Asked Questions

What is 熵管理系统?

熵管理系统 - 原创技能。用于管理和控制AI会话中的熵（无序度），包括上下文精简、状态重置、注意力聚焦等功能。适用于长时间会话、复杂任务、多步骤工作流等场景。 It is an AI Agent Skill for Claude Code / OpenClaw, with 41 downloads so far.

How do I install 熵管理系统?

Run "/install entropy-manager" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is 熵管理系统 free?

Yes, 熵管理系统 is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does 熵管理系统 support?

熵管理系统 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created 熵管理系统?

It is built and maintained by 534422530 (@534422530); the current version is v1.0.0.

More Skills