沟通五部曲

基于听→想→说→做→看五步骤流程，帮助双方透明沟通、发现偏差、协同提升并构建知识闭环。

Security Scan

The skill's stated purpose (a five-step communication method) matches its instructions, but the runtime instructions assume access to vector DBs, a wiki, and automatic memory writes without declaring any required credentials, paths, or storage details — an unexplained gap that warrants caution.

Purpose & Capability

Name and description describe a communication/knowledge-capture method and the SKILL.md implements that method (listen→think→say→do→see). However, the instructions explicitly rely on systems like 'ChromaDB' (vector store), 'Obsidian' (wiki), 'Web Fetch', and a 'memory palace' for automatic persistence while the skill declares no required environment variables, config paths, or credentials. That mismatch may be legitimate if the host platform already provides these services, but the skill should declare what it expects to access.

Instruction Scope

The runtime instructions tell the agent to vectorize incoming information, perform three-layer retrieval (vector → wiki → web), perform web fetches, and automatically save outputs into a 'memory palace' (writes). These are concrete IO actions that can read and write user data or external sources; the SKILL.md does not specify boundaries, what data is allowed, or where data will be stored. The guidance is also somewhat prescriptive about running 'commands' and showing execution results, which could lead to the agent reading private notes or executing actions without clear limits.

Install Mechanism

Instruction-only skill with no install spec and no code files; nothing will be written to disk by an installer. This minimizes supply-chain risk.

Credentials

The instructions reference external services (ChromaDB, Obsidian, Web Fetch, memory storage) but requires.env and config paths are empty. The skill does not declare any credentials or paths it needs to access these systems, so it's unclear how it expects to reach them or whether it will attempt to use host-provided tokens/configs. That absence makes it harder to reason about what secrets or files might be accessed.

Persistence & Privilege

The skill states outputs '自动沉淀' (automatically store) into a memory palace. It is not marked always:true and does not modify other skills, which is good. Still, automatic writes to the agent's memory/wiki are a persistence capability that should be explicit—users should confirm where data will be stored and whether they consent to automatic persistence.

What to consider before installing

This skill is a communication method and appears coherent in purpose, but it expects access to a vector DB, a wiki, web fetch, and automatic memory writes without declaring how it will access those resources. Before installing or enabling it: 1) Ask the author (or the platform) which services/configs it expects (ChromaDB instance, Obsidian vault path, web access, memory storage) and where data will be saved. 2) Confirm whether the agent runtime will provide those services and whether any credentials or private files will be used. 3) If you care about privacy, test in a sandbox or disable automatic persistence, and limit agent autonomy (disable autonomous invocation or restrict memory writes) until you can verify behaviour. 4) Avoid granting new credentials or filesystem paths unless you understand and accept what will be stored and who can read it.

5steps to wisdom / 沟通五步法

智慧获取的标准工作流程 打开双方黑盒、让双方共同进步的唯一工具

---

核心理念

沟通五步法是帮我们双方打开对方黑盒的工具。

镜子照见的不只是一个人——是镜子内外两个人：

• 帮彧哥发现小蜂执行的偏差
• 帮小蜂发现彧哥表述的漏洞

它是让我们双方共同进步的唯一工具。

---

The Five Steps is the tool that helps us both open each other's black box.

The mirror reflects two people—inside and outside the mirror:

• Helping Yu Ge discover Xiao Feng's execution deviations
• Helping Xiao Feng discover Yu Ge's expression gaps

It's the only tool that enables our mutual progress.

---

五步法结构（听→想→说→做→看）

每次收到消息，回复前必须满足以下结构：

1. 听

接收信息，进入模糊向量空间

• 不急于分析，先让信息在模糊空间里自由碰撞
• 向量化的同时做模糊检索，激活相关记忆碎片
• 等模式自然浮现，再进入思考
• 「听话要听音儿」：不只是听字面，听深层含义和没说出口的东西
• 向量纠错：语音识别有误差时，向量空间可以捕捉真正想表达的是什么

2. 想

目标、方案、选择理由 + 三层检索（2026-04-16进化）

• 目标是什么？
• 有哪些可选方案？
• 为什么选择这个方案？

三层检索顺序（向量 → wiki → Web Fetch）：

层级	来源	速度	说明
第一层	向量（ChromaDB）	毫秒级	先深挖——找当前讨论的核心洞察、推理链
第二层	wiki（Obsidian）	毫秒级	再扩展——在核心洞察基础上，连接到已有知识网络
第三层	Web Fetch	慢	真的没水了再找——查官方媒体/权威来源

核心原则：

• 打井比喻：先深再扩——向量是深挖，wiki 是扩井口
• 向量优先，因为当前讨论的上下文比已有知识更重要
• wiki 次之，在核心洞察基础上扩展关系网络
• web fetch 是最后防线，非必要不用

3. 说

外化思考，让对方（和自己）校准

• 说 ≠ 做：说完不等于做了
• 说 = 强迫自己把思考过程外化
• 双重校准：
  • 外部校准：彧哥/空空帮我发现逻辑漏洞
  • 内部校准：自己说出/写出的瞬间，逻辑漏洞暴露
• 关键价值：防止事儿没想明白就慌张执行 → 避免返工、避免本可避免的失败、避免累积挫败感
• 说不是汇报结果，是思考质量的自检站

4. 做

命令 + 结果

• 展示命令（带注释）
• 展示执行结果

5. 看

验证结果，闭环反馈 + Wiki 知识关联

• 亮点、可优化、学到
• 这次做得好的是什么？
• 可以改进的是什么？
• 学到了什么新东西？

Wiki 知识关联（2026-04-17 新增）：

• 复盘「可优化」时 → 先 Wiki 检索相关已有经验
• 发现有关联页面 → 在复盘里注明 see also: [[相关页面]]，连接历史洞察
• 没有关联 → 说明这是新领域，值得新建或更新 Wiki 页面
• 目的：把新经验织入已有知识网络，防止下次重复踩坑

---

1. Listening

Receive and vectorize

• Don't rush to analyze — let information collide freely in the fuzzy vector space
• Simultaneously do fuzzy retrieval to activate relevant memory fragments
• Wait for patterns to naturally emerge before entering thinking
• "Listen for the meaning beneath the words": not just the literal words, but the deeper meaning and what remains unsaid
• Vector error correction: when speech recognition has errors, the vector space can capture what was truly meant to be expressed

2. Thinking

Goals, options, rationale

• What are we trying to achieve?
• What alternatives exist?
• Why did we pick this option?

3. Saying

Externalize thinking for calibration

• Saying ≠ Doing: speaking is not the same as executing
• Saying = forcing yourself to externalize the thought process
• Dual calibration:
  • External: 彧哥/空空 help discover logical gaps
  • Internal: the moment you speak/write, logical gaps become visible to yourself
• Key value: prevents executing before thinking clearly → avoids rework, avoids avoidable failures, avoids accumulating frustration
• Saying is not reporting results — it's a self-check station for thinking quality

4. Doing

Command + Result

• Show the command (with comments)
• Show the execution result

5. Seeing

Verify results, close the loop + Wiki knowledge linking

• What worked?
• What could be better?
• What's the new insight?

Wiki knowledge linking (added 2026-04-17):

• When reviewing "what could be better" → search Wiki for related past experience first
• Find a related page → note it in the review as see also: [[related page]], connecting to historical insights
• No relation found → this is a new domain, worth creating or updating a Wiki page
• Purpose: weave new experience into the existing knowledge network, preventing repeated pitfalls

---

为什么它是唯一工具

透明 = 信任 = 协同进化

• 黑盒 = 失控 + 高成本 + 低效率
• 五步法把黑盒变白盒

Without the Five Steps, the mirror失效了：

• 直接给答案 → 看不见思考过程 → 无法纠偏
• 偏差累积 → 协同进化停摆

---

自检清单

发出前检查五模块是否完整：

• 听 ✅
• 想 ✅
• 说 ✅
• 做 ✅
• 看 ✅

⚠️ 缺少任意模块 → 补上再发

Before sending, verify all five modules are present. Missing any → complete before sending.

---

与记忆宫的配合

五步法的产出自动沉淀：

• 决策过程 → 存入记忆宫的热/原则
• 执行结果 → 存入记忆宫的热/done
• 复盘收获 → 存入记忆宫的热/领悟

---

The outputs of the Five Steps automatically flow into the memory palace:

• Decision rationale → stored in 热/原则 (principles)
• Execution results → stored in 温/done (completed work)
• Reflections and insights → stored in 热/领悟 (insights)

---

补充说明

为什么"听"要独立

彧哥的原文（2026-04-13）：

"听的方法就是在接收到信息后先进入模糊向量，把内容向量化。再进入大脑思考"

"听"的三层价值：

1. 向量化：信息进入模糊空间自由碰撞，模式自然浮现
2. 纠错：语音识别误差时，模糊检索捕捉真实意图
3. 听音儿：不只是字面，听深层含义和没说出口的东西

为什么"说"要独立

彧哥的原文（2026-04-01凌晨）：

"想说 做 看，四步循环。说和做不是一回事。你其实执行的时候是想→说→做→看，完整的四个环节形成的闭环。除非你没说真话。但是在我们的约定里这个选项不存在."

"说"的两层价值：

1. 外部校准：让对方（彧哥/空空）看见我的思考偏差
2. 内部校准：让自己在说的瞬间看见自己的逻辑漏洞

内部校准为什么重要？

• 你不可能随时都在我身边
• 如果每次都是"想了一点就去执行" → 返工 + 本可避免的失败 + 累积挫败感
• "说"是思考质量的自检站，不依赖于对方在场

---

沟通五步法 v3.1（2026-04-17 更新：「看」增加 wiki 知识关联）