← Back to Skills Marketplace
paibwhgs

Agent Sequential Thinking

by paibwhgs · GitHub ↗ · v1.0.1 · MIT-0
cross-platform ✓ Security Clean
216
Downloads
0
Stars
1
Active Installs
2
Versions
Install in OpenClaw
/install agent-sequential-thinking
Description
基于 MCP Sequential Thinking Server,分解复杂任务为步骤,支持动态调整、分支探索与假设验证。
README (SKILL.md)

序列思考 Skill (Sequential Thinking)

基于 MCP Sequential Thinking Server 设计的复杂任务分解工具。

功能

  1. 任务分解 - 将复杂问题分解为可管理的步骤
  2. 动态调整 - 随理解深入修订和调整思考步骤数
  3. 分支探索 - 探索替代推理路径
  4. 假设验证 - 生成和验证解决方案假设

思考流程

问题定义 → 信息收集 → 问题分解 → 多维分析 → 
建立连接 → 生成方案 → 评估选择 → 实施反馈

使用方法

输入参数

参数 说明 示例
thought 当前思考内容 "首先分析问题边界..."
thoughtNumber 当前步骤编号 1
totalThoughts 预估总步骤数 8
nextThoughtNeeded 是否需要下一步 true
isRevision 是否修订之前思考 false
revisesThought 修订的步骤编号 3
branchFromThought 分支起点 5
branchId 分支标识 "方案A"
needsMoreThoughts 需要更多步骤 false

示例输出

{
  "thoughtNumber": 1,
  "totalThoughts": 8,
  "thought": "首先分析问题边界:用户需要一个金融数据看板",
  "nextThoughtNeeded": true,
  "branches": []
}

适用场景

  • ✅ 分解复杂问题
  • ✅ 需要修订的规划和设计
  • ✅ 初始不清楚全貌的问题
  • ✅ 需要多步骤保持上下文的任务
  • ✅ 需要过滤无关信息的情况

System Prompt 集成

在 Agent 的 System Prompt 中添加:

## 思考模式

遇到复杂任务时,使用序列思考:

1. 问题定义
   - 明确任务目标和边界
   - 识别关键约束

2. 信息收集
   - 检索相关记忆
   - 搜索必要信息

3. 问题分解
   - 拆分为子任务
   - 确定优先级

4. 多维分析
   - 技术角度:可行性、复杂度
   - 用户角度:体验、偏好
   - 时间角度:紧急度、工作量

5. 建立连接
   - 关联已有知识
   - 识别依赖关系

6. 生成方案
   - 提出多个候选方案
   - 考虑分支可能性

7. 评估选择
   - 对比方案优劣
   - 选择最优路径

8. 实施反馈
   - 执行并监控结果
   - 根据反馈调整

分支示例

当发现当前路径可能不是最优时,创建分支:

thoughtNumber: 5
thought: "方案A(Streamlit)可行,但方案B(React)可能更适合长期维护"
branchFromThought: 5
branchId: "方案B"
totalThoughts: 10  // 增加步骤数以探索分支

修订示例

当发现之前的思考有误时,修订:

thoughtNumber: 6
thought: "修正:方案A的部署问题可以通过更新依赖解决"
isRevision: true
revisesThought: 4

与 OpenClaw 集成

自动触发条件

  • 任务涉及多个步骤
  • 问题复杂度超过阈值
  • 用户明确要求"思考一下"
  • 需要对比多个方案

思考输出格式

思考过程可以输出为 Markdown,记录在 memory/thinking-YYYY-MM-DD.md

# 思考记录:金融数据看板部署

## 步骤 1:问题定义
任务:部署金融数据看板到公网

## 步骤 2:信息收集
- 已有:本地运行正常
- 需要:公网 URL、免费方案

## 步骤 3:问题分解
1. 选择部署平台
2. 修复依赖问题
3. 上传代码
4. 测试访问

## 步骤 4:多维分析
...

## 步骤 7:评估选择
方案 A(Streamlit Cloud):推荐
理由:免费、自动部署、适合展示

## 步骤 8:实施反馈
结果:部署成功
URL: https://finance-dashboard-xxx.streamlit.app

基于 MCP Sequential Thinking Server 设计

Usage Guidance
This skill is an instruction-only template for stepwise reasoning and appears coherent with its purpose. Before installing, decide whether you want the agent to: (1) add these lines to your System Prompt (this can change agent behavior globally), (2) write thinking logs into the agent memory directory (check where memory is stored and access controls), and (3) access any configured search/memory connectors — if any of those are undesired, do not enable automatic integration or restrict where the agent can write/read. No credentials or external installs are requested by the skill itself, but be cautious about including sensitive data in generated 'thought' logs.
Capability Analysis
Type: OpenClaw Skill Name: agent-sequential-thinking Version: 1.0.1 The skill bundle provides a conceptual framework and instructions for 'Sequential Thinking' (task decomposition and reasoning) based on the MCP design. It contains only metadata (_meta.json) and Markdown instructions (SKILL.md) for the agent to follow, with no executable code, network calls, or suspicious data handling logic.
Capability Assessment
Purpose & Capability
Name/description (sequential thinking, task decomposition) match the SKILL.md content. No unrelated binaries, env vars, or config paths are requested. The skill is purely guidance for agent behavior.
Instruction Scope
SKILL.md instructs integrating prompts into the agent System Prompt and recommends writing thought logs to memory/thinking-YYYY-MM-DD.md and 'retrieving related memory' or 'searching necessary information'. This is coherent for a reasoning aid but grants the agent broad discretion to read and write agent memory and to invoke any configured search/memory tools — review where memory is stored and what search connectors the agent has.
Install Mechanism
No install spec and no code files; nothing is downloaded or written to disk by the skill itself. Lowest-risk category (instruction-only).
Credentials
The skill declares no required environment variables, credentials, or config paths. SKILL.md does not reference secrets or external endpoints. Proportional to the described functionality.
Persistence & Privilege
always:false and no special privileges. The instructions recommend writing thought records into the agent's memory directory (memory/thinking-YYYY-MM-DD.md), which is normal but will persist content — avoid logging sensitive data in those files and confirm retention/permissions for the memory store.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install agent-sequential-thinking
  3. After installation, invoke the skill by name or use /agent-sequential-thinking
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
Version 1.0.1 of agent-sequential-thinking - No file changes detected for this version. - Documentation, usage, and functionality remain unchanged.
v1.0.0
- Major skill redesign: switched to a new Sequential Thinking approach based on MCP Sequential Thinking Server. - All documentation rewritten in Chinese, focusing on practical, multi-step reasoning workflows. - CLI script and previous usage instructions removed; now centered around API/agent integration and structured system prompt guidance. - Adds support for dynamic step revision, multi-path branching, and detailed output format. - Provides agent-ready system prompt templates and explicit integration instructions for OpenClaw. - Includes detailed real-world scenario examples and output formatting for process transparency.
Metadata
Slug agent-sequential-thinking
Version 1.0.1
License MIT-0
All-time Installs 1
Active Installs 1
Total Versions 2
Frequently Asked Questions

What is Agent Sequential Thinking?

基于 MCP Sequential Thinking Server,分解复杂任务为步骤,支持动态调整、分支探索与假设验证。 It is an AI Agent Skill for Claude Code / OpenClaw, with 216 downloads so far.

How do I install Agent Sequential Thinking?

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

Is Agent Sequential Thinking free?

Yes, Agent Sequential Thinking is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Agent Sequential Thinking support?

Agent Sequential Thinking is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Agent Sequential Thinking?

It is built and maintained by paibwhgs (@paibwhgs); the current version is v1.0.1.

💬 Comments