独立思考五部曲

独立思考五部曲。触发场景：**架构讨论、排错、复盘、决策、不确定知识点**。日常闲聊、确认、打招呼、简单问答 → 直接回答，不查向量、不 thinking、不走五步曲。与沟通五部曲并行通用。

Security Scan

The skill's instructions assume and mandate use of a specific local vector query script and permanent disabling of the system Memory Search, but the package declares no required files, binaries, or credentials — these mismatches and the hardcoded absolute path are incoherent and merit caution.

Purpose & Capability

The skill is an instruction-only 'independent thinking' workflow, which is reasonable as prose. However, it mandates calling a local script at a hardcoded absolute path (/Users/markma01/.openclaw/workspace/scripts/query_vector.py) and to use ChromaDB, yet the skill declares no required binaries, environment variables, or config paths. Requiring a specific local script and DB without declaring them is disproportionate and inconsistent with the stated metadata.

Instruction Scope

SKILL.md explicitly forces the agent to run a local Python script at a absolute user-specific path and forbids alternate invocation patterns. It also demands 'permanently disable system Memory Search' and 'only use the exclusive backend vector system' and requires writing 'look' outputs back into a wiki. These instructions reach into local filesystem, agent runtime behavior, and platform features without providing configuration, fallback, or justification, giving the skill broad operational influence beyond a simple decision workflow.

Install Mechanism

No install spec or code files are present (instruction-only), so nothing is written to disk by the skill itself. This lowers installation risk, but the runtime instructions still expect a local script and a ChromaDB backend that are not supplied.

Credentials

The skill asks (in practice) for access to a local script and a vector DB and asks to disable a system feature, but declares no required env vars, credentials, or config paths. That mismatch is disproportionate: the skill's operation depends on resources and potential privileges (local filesystem execution and wiki persistence) that are not declared or explained.

Persistence & Privilege

The skill does not set always:true and does not request persistent installation. However, it instructs the agent to '永久禁用系统自带 Memory Search' (permanently disable system Memory Search) and to persist outputs into a wiki for future use. Asking the agent to alter platform behavior and to write persistent knowledge is a non-trivial privilege implication even without always:true; verify what agent/platform-level actions are actually possible before trusting this instruction.

What to consider before installing

This skill is an instruction-only workflow that mandates running a local Python script at /Users/markma01/.openclaw/workspace/scripts/query_vector.py, using ChromaDB, disabling the system Memory Search, and writing results into a wiki — yet it declares no files, env vars, or config paths. Before installing or enabling it: (1) confirm you have the referenced query_vector.py and ChromaDB configured, or update the skill to point to your environment; (2) do not permit it to change platform-level features (like disabling Memory Search) unless you control and understand the platform; (3) verify where the wiki writes go and what permissions are required; (4) prefer running this in a sandboxed agent instance while testing; (5) if you cannot map the hardcoded path to your environment, treat the skill as incompatible or request the author to parameterize paths and explain required credentials.

底层核心准则 · 与沟通五部曲同级并行 自验驱动的深度决策方法论

核心定位

自验 = 用自己过去的总和（向量+wiki）替代当下他人在场

独立思考五部曲是沟通五部曲的"镜像"——结构相同，验证介质不同：

• 沟通五部曲 → 他验（对方实时反馈）
• 独立思考五部曲 → 自验（向量+wiki 历史知识库）

---

五步法结构（听→想→说→做→看）

1. 听：直觉接收

接收问题/刺激，让直觉碎片在模糊空间自由碰撞，等待模式自然浮现。

2. 想：向量溯源 + wiki扩展

开口/动手前，先向量检索历史经验和踩坑教训，再 wiki 扩展连接到已有知识网络，形成全局独立判断。

向量调用规范（强制）：

# 唯一正确格式
python3 /Users/markma01/.openclaw/workspace/scripts/query_vector.py "<检索词>"

# 禁止以下所有格式
cd /path && python3 script.py
echo "query" | python3 script.py

3. 说：有据表达 + wiki校验

将思考结论结构化输出，同步 wiki 校验论点是否与已有知识对齐，逻辑连贯、不主观臆断。

注意：独立思考的"说"不是汇报（无实时对方），而是将脑内推理外化为可供后续追溯的书面结论。

4. 做：向量校验

执行前用向量核对步骤、风险与适配方案，完成自检后再动手。

5. 看：wiki 知识关联

复盘时检索 wiki 连接已有经验，把新经验织入知识网络，防止重复踩坑。

---

检查点矩阵

阶段	检查点	他验/自验
听	直觉接收，等待模式浮现	自验
想	向量溯源 + wiki扩展	自验
说	wiki连接已有论点	自验
做	向量核对步骤	自验
看	wiki连接已有经验	自验

---

强制执行规则

1. 仅在触发场景下执行（架构讨论、排错、复盘、决策、不确定知识点）
2. 日常闲聊、确认、打招呼、简单问答 → 直接回答，不查向量、不 thinking、不走五步曲
3. 永久禁用系统自带 Memory Search，仅使用专属后台向量系统（query_vector.py / ChromaDB）
4. 「看」的输出必须闭环存入 wiki，为下次「想」提供素材
5. 适时触发：复杂任务全程检查点，简单任务降低检查点密度

---

与沟通五部曲的关系

维度	沟通五部曲	独立思考五部曲
核心	他验	自验
验证介质	对方实时反馈	向量 + wiki 历史知识库
侧重	互动、汇报、方案讨论	思考、决策、复盘
并行	对外沟通用	对内思考用

两者合并，才是完整的「想清楚 → 说出来 → 做到位 → 固下来」闭环。

---

快速触发检查清单

每次开口/动手前自检：

• 已调用 query_vector.py 检索相关历史经验
• 「想」的结论有向量事实支撑 + wiki 扩展
• 「说」的论点已 wiki 校验
• 「做」的方案经向量校验无已知风险
• 「看」的结果已连接 wiki 并计划存入

---

本 skill 为小蜂独立思考核心准则，与「沟通五部曲」同级并行。 v5.0（2026-04-17）：升级为五部曲，融入 wiki 检查点，厘清他验/自验定位