Deep Loop Thinker

多轮深度思考技能。借鉴OpenMythos循环推理架构，每次循环都注入新输入，逐轮深化理解。适用于重要决策、复杂问题、创意生成、问题发现。

Security Scan

The skill's description, instructions, and bundled Python code are consistent with a local, instruction-only 'multi-round thinking' helper and do not request credentials, perform network I/O, or install extra software.

Purpose & Capability

Name/description (multi-round reasoning) matches the SKILL.md and the included Python script: both implement prompts and a looped thinking workflow. No unrelated binaries, credentials, or external services are requested.

Instruction Scope

SKILL.md describes workflow, triggers, and a user-feedback recording format. The included Python script implements the interactive multi-round prompts and printing only. There is a minor inconsistency: SKILL.md states that user feedback is recorded for iteration, but the script does not persist feedback to disk or transmit it anywhere — it only prints. This is a functional mismatch but not a security concern.

Install Mechanism

No install spec is present (instruction-only). The only code file is a small local Python script that runs interactively; nothing is downloaded or executed from third-party URLs.

Credentials

The skill requires no environment variables, credentials, or config paths. The code does not access environment variables, secret names, or external services.

Persistence & Privilege

The skill is not always-enabled and uses normal agent invocation settings. It does not modify other skills or system-wide configs and does not persist data or install agents/services.

Assessment

This skill appears coherent and low-risk: it runs a local Python script that prints structured multi-round prompts and requires no network access or credentials. Before installing or giving it execution rights, you may want to: (1) run the Python script locally to verify behavior, (2) confirm the SKILL.md/README expectations about recording feedback (the shipped script does not persist logs), and (3) ensure you trust the source since the package contains a code file even though it is small and benign. If you plan to let an agent invoke skills autonomously, remember autonomous invocation is platform-default — consider limiting that capability if you do not want skills to run without explicit approvals.

Deep Loop Thinker Skill v2.1

核心原理

借鉴OpenMythos的Recurrent-Depth Transformer：

• 同一层多次运行，每次有新输入注入
• 隐藏状态h_t更新：h_{t+1} = A·h_t + B·e + Transformer(h_t, e)
• 每次循环不是重复，而是递进

用户需求追踪

谁会使用这个技能？

• 创业者做重大决策
• 分析师处理复杂问题
• 开发者解决技术难题
• 研究者探索未知领域
• AI Agent增强推理能力

用户痛点

• 决策前思考不够全面
• 容易忽视风险
• 缺乏系统性反思
• 决策后不复盘

循环架构

问题输入
    ↓
[Prelude层] — 提取关键要素
    ↓
[Recurrent Block] — 多轮递进思考
    ↑____↓
    ↓ (每轮注入新洞察)
[Coda层] — 综合输出
    ↓
行动计划 + 自我反思

多轮思考模式

轮次	思考类型	核心问题
1	直觉捕捉	第一反应？情绪？
2	利益分析	谁受益？谁受损？
3	风险挖掘	最坏情况？3年后还重要吗？
4	本质洞察	根本原因？规律？
5	行动设计	第一步？Plan B？
6	反思验证	假设可靠？盲点在哪？

用户反馈收集

每次使用后记录：

【用户反馈】
- 用户是谁：{匿名/具体描述}
- 帮助程度：{1-5星}
- 改进建议：{用户的具体反馈}
- 下次需要：{用户的实际问题}

迭代记录

版本	日期	更新内容
1.0	2026-04-21	初始版本
2.0	2026-04-22	增加6轮思考
2.1	2026-04-22	增加用户追踪机制

质量标准

好的思考报告：

• ✅ 有具体数据/例子支撑
• ✅ 能回答"为什么"
• ✅ 第一步可立即执行
• ✅ 包含风险预案
• ✅ 诚实承认盲点

触发条件

• 问题影响超过3个月
• 涉及多方利益
• 情绪波动明显
• 反复纠结无法决定
• 问题反复出现

约束

• 简单问题不要过度思考（杀鸡不用牛刀）
• 不要超过6轮（防止无限循环）
• 保持诚实，不要自欺欺人
• 行动比完美计划重要
• 记录用户反馈用于迭代