← Back to Skills Marketplace
kedoupi

AI Self-Evolution Engine( AI 自我进化引擎)

by KeDouPi(珂抖屁) · GitHub ↗ · v1.0.4 · MIT-0
cross-platform ✓ Security Clean
193
Downloads
0
Stars
0
Active Installs
5
Versions
Install in OpenClaw
/install cognitive-forge
Description
Dual-value learning system - extracts reusable mental models from books, writes individual pattern files (patterns/{id}.md) with YAML frontmatter for buildin...
README (SKILL.md)

Cognitive Forge (认知锻造)

One run, dual value — 每次运行同时产出两个价值:

  1. 用户获得 F.A.C.E.T. 深度分析,提取可立即应用的思维模型
  2. AI 获得 永久写入 patterns/{id}.md 的决策框架(带 YAML frontmatter),构建可复用的思维模型库

随时间积累,你的 AI 拥有一个不断增长的决策框架库(类似 Charlie Munger 的 "latticework of mental models"),在未来任何领域的提问中都可以引用。


Path Convention

所有路径均相对于 OpenClaw workspace 根目录(通常为 ~/.openclaw/workspace/)。 如用户 workspace 位于其他位置,请将文档中的路径替换为实际 workspace 路径。

用途 相对路径
阅读记录 memory/reading-history.json
思维框架库 memory/knowledge-base/patterns/*.md (每个模型一个文件)
概念库 memory/knowledge-base/concepts.md
用户画像 USER.md
调度配置 HEARTBEAT-reading.md

Routing (路由分支)

根据用户意图,选择不同的执行路径:

用户意图 路由 说明
"生成今日读书简报" / 默认 Main Workflow (breadth) 完整选书→分析→写入流程,提取 1 个模型
"深度分析《XXX》" / "depth_mode: depth" Main Workflow (depth) 对指定书籍连续提取多个模型,合并输出
"cognitive-forge status" / "认知锻造 状态" Status Branch 输出知识库统计
"cognitive-forge review" / 周日自动触发 Review Branch 间隔复习本周模型
"分析《XXX》这本书" Main Workflow (breadth, 跳过选书) 用户直接指定书籍,提取 1 个核心模型

Depth mode 触发方式

  1. 手动触发:用户说"深度分析《XXX》"或传入 depth_mode: depth
  2. 定时触发:HEARTBEAT-reading.md 中可配置 depth: true,调度时传入该参数则自动走 depth mode

Status Branch (知识库统计)

当用户请求查看知识库状态时:

  1. 统计 memory/knowledge-base/patterns/ 目录下 .md 文件数 = 模型总数
  2. 读取 memory/reading-history.json,统计:
    • 已读书籍总数(used_models 数组长度)
    • 各领域分布(按 category 分组计数)
    • 最近 5 条记录
  3. 输出格式:
## 📊 认知锻造 · 知识库状态

**模型总数**: XX 个思维框架
**已读书籍**: XX 本
**知识库大小**: XX KB

### 领域分布
| 领域 | 模型数 | 占比 |
|------|--------|------|
| Business Strategy | 5 | 25% |
| Psychology | 3 | 15% |
| ... | ... | ... |

### 最近 5 条
1. 2026-03-27 | 《反脆弱》 | 反脆弱三元组
2. ...

### 覆盖薄弱领域
⚠️ Philosophy (0), Biography (0) — 建议补充

Review Branch (间隔复习)

触发方式

  • 用户手动说 "cognitive-forge review"
  • 当 HEARTBEAT-reading.md 配置了周日时段时,自动在周日 briefing 中插入复习环节

执行逻辑

  1. 读取 memory/reading-history.json,筛选最近 7 天的 used_models 记录
  2. 随机抽取 2-3 个模型
  3. 输出复习问答:
## 🔄 本周复习:你还记得这些模型吗?

**1.「逃离机制」来自《逃离不平等》**
- 核心框架是什么?(回忆 [F])
- 什么时候会失效?(回忆 [E])

**2.「双系统理论」来自《思考,快与慢》**
- 它摧毁了什么常识?(回忆 [C])
- 你上周在工作中用到了吗?

> 回复你的答案,我帮你查漏补缺。

Main Workflow

Step 0. Environment Check (首次使用自检)

每次运行开始时执行,静默完成(不打断用户):

  1. 检查 memory/reading-history.json 是否存在

    • 不存在 → 自动创建初始文件:
      {
        "schema_version": 1,
        "last_attempted": null,
        "queue": [],
        "used_models": []
      }
      
  2. 检查 memory/knowledge-base/ 目录 是否存在

    • 不存在 → 自动创建目录
  3. 检查 memory/knowledge-base/patterns/ 目录 是否存在

    • 不存在 → 自动创建目录
  4. 检查 memory/knowledge-base/concepts.md 是否存在

    • 不存在 → 创建带标题的空文件
  5. 检查 USER.md 是否存在

    • 不存在 → 输出提示:💡 建议创建 USER.md(职业、兴趣、当前挑战),以获得个性化的 [T] Transfer 分析。
    • 存在 → 静默读取,提取用户画像
  6. 检查依赖 skill 是否可用

    • book-scoutmental-model-forge 必须可调用
    • 不可用 → 报错并停止
  7. 检查 last_attempted 字段

    • 如果 status == "failed" → 提示用户:
      ⚠️ 上次运行在「{step}」步骤失败(书籍: {book})。
      是否要恢复上次操作?回复"是"恢复,或"否"跳过。
      

Step 1. Select Book (选书)

选书来源优先级(从高到低):

来源 1: 用户直接指定

如果用户明确说了 "分析《XXX》by YYY",直接使用该书,跳过选书流程。

  • 标记 source: "user_specified"

来源 2: 预排队列

检查 memory/reading-history.jsonqueue 数组:

{
  "queue": [
    {"title": "《穷查理宝典》", "author": "彼得·考夫曼", "topic": "决策科学"}
  ]
}
  • 如果 queue 非空 → 取第一项,从 queue 中移除
  • 标记 source: "queue"

来源 3: book-scout 网络搜索

当 queue 为空且用户未指定时,调用 book-scout skill。

确定搜索主题

  1. 检查 HEARTBEAT-reading.md 是否有自定义主题映射(## 主题映射 section)
    • 如有 → 使用自定义映射
  2. 否则使用默认星期-主题映射
星期 默认主题
Monday Business Strategy
Tuesday Psychology
Wednesday Technology
Thursday Economics
Friday Innovation
Saturday Philosophy
Sunday Biography

可配置: 用户可在 HEARTBEAT-reading.md 中添加 ## 主题映射 section 覆盖默认值。 也可以按时段细分(参考 HEARTBEAT-reading.md 中的 21 主题轮转配置)。

加载去重列表(书名去重):

memory/reading-history.json 提取所有 book_title 字段值,去重后作为已读书名列表。

重要:不读取 thinking-patterns.mdpatterns/*.md。去重只需要书名,不需要模型内容。

调用 book-scout

主题: {topic}

已读书籍:
- 《精益创业》
- 《从0到1》
- 《影响力》

执行 book-scout skill,搜索符合主题的经典书籍。

重试机制

  • Attempt 1 失败 → 等 2s → 重试
  • Attempt 2 失败 → 等 3s → 重试
  • Attempt 3 失败 → 返回错误给用户:
    ⚠️ 选书失败:{error}
    已尝试 3 次。你可以直接指定书籍:"分析《书名》by 作者"
    

book-scout 成功返回

{
  "book_title": "《增长黑客》",
  "author": "肖恩·埃利斯",
  "author_nationality": "美国",
  "publish_date": "2015-04",
  "rating": 8.5,
  "review_count": 10000,
  "score": 74.4,
  "summary": "增长黑客方法论...",
  "reasoning": "评分8.5且有1万真实评价..."
}

标记 source: "web_search",进入 Step 2。

更新 last_attempted

"last_attempted": {
  "date": "YYYY-MM-DD",
  "book": "《增长黑客》",
  "step": "book_selection",
  "status": "success"
}

Step 2. Extract Mental Model (提取思维模型)

Breadth Mode (默认)

调用 mental-model-forge skill,对选中的书进行 F.A.C.E.T. 分析,提取 1 个核心思维模型

Depth Mode

当用户指定 depth_mode: depth 时,对同一本书连续提取多个思维模型

工作流

  1. 第 1 次调用 mental-model-forge,提取书中最核心的思维模型
  2. 将已提取的模型名称作为 exclude_models 参数,再次调用:
    这本书是《反脆弱》。
    exclude_models: ["反脆弱三元组"]
    请提取这本书中另一个独立的、不同的思维框架。
    
  3. 重复直到三重退出条件任一触发:
退出条件 判断方式
模型数上限 该书已提取 5 个模型 → 停止
语义去重 AI 判断新模型与已提取模型本质相同(同一思想的变体或换皮)→ 停止
AI 自评 提取后自问 "这本书还有独立的、值得提取的思维框架吗?" → No → 停止
  1. 每提取一个模型,立即执行 Step 4-5(写入知识库 + 更新记录)
  2. 所有模型提取完毕后,合并为一份报告执行 Step 3 + Step 6(生成简报 + 写入外部数据库)

Step 2.5. F.A.C.E.T. Quality Verification (结构化验证)

mental-model-forge 返回后,执行以下自检:

  • 完整性: 5 个维度 [F][A][C][E][T] 是否都有实质内容(非空、非占位符文本)
  • [F] 字数: Framework 是否 ≤ 80 字(中文)或 ≤ 50 words(英文)
  • [T] 个性化: Transfer 是否引用了 USER.md 中的具体信息(职业、项目、挑战)
    • 如果 USER.md 存在但 [T] 未引用任何用户上下文 → 验证失败
  • 质量自评: 整体分析质量 1-10 分

处理

  • 自评 ≥ 7 分且全部检查通过 → 进入 Step 3
  • 自评 \x3C 7 分或任一检查失败 → 重新调用 mental-model-forge(最多重试 1 次
  • 重试后仍不合格 → 使用当前版本但在输出中标注 ⚠️ 本次分析质量未达标,建议后续深入阅读

Step 3. Generate Briefing (生成简报)

输出模式(默认 full):

Full Mode (默认)

创建完整结构化简报,必须适配用户上下文

[强制步骤] 读取 USER.md

  • Path: USER.md(相对于 workspace 根)
  • 如果存在,提取:
    • 工作经历 / 现在 → profession
    • 兴趣 / 爱好 → interests
    • 当前焦虑 / 未来规划 → current challenges
  • 如果不存在 → 使用通用第二人称("你"),可追问用户背景

输出结构

## 📖 今日思维锚点

**书籍**: 《XXX》 - 作者
**核心一句话**: [今日思维锚点,一句话总结]

---

## 🧠 F.A.C.E.T. 认知穿透

### [F] Framework (核心框架)
[核心机制,≤80字中文]

### [A] Anchor Case (锚定案例)
[最经典的真实案例,生动讲述]

### [C] Contradiction (反共识摧毁)
❌ 被摧毁的常识: "..."
✅ 真相: ...

### [E] Edge (隐性边界)
失效条件:
1. ...
2. ...

### [T] Transfer (跨界迁移)
[映射到用户的实际上下文:职业、项目、挑战]

---

## 🎯 应用场景

| 场景 | 如何应用 | 预期效果 |
|------|---------|---------|
| [场景1:映射用户职业] | ... | ... |
| [场景2:映射用户项目] | ... | ... |
| [场景3:映射用户挑战] | ... | ... |

## 🔴 反面案例
[违反该原则的真实或假设案例]

## 🤔 战略拷问
[尖锐、具体、可行动的问题,引用用户实际上下文]
- Bad: "企业家应该怎么做?"
- Good: "你在爱康国宾的 AI 产品,是在避免失败还是利用失败?"

## 🔄 认知模式更新
**思维框架**: 看到XX → 想到XX
**决策原则**: 在XX场景下,应该XX而非XX
**盲区警告**: 小心XX情况
**反射弧**: 看到XX信号 → 联想到这个模型 → 判断/行动

---

> 💬 这个模型让你想到工作中的哪个具体场景?回复我,我帮你深入分析。

个性化规则

  • 始终用第二人称("你的"、"你在")
  • [T] Transfer 必须引用用户具体信息(职业、项目名、公司名)
  • 战略拷问必须具体到用户当前处境,不可泛泛而谈
  • 应用场景 ≥ 3 个,分别映射用户的不同维度

Brief Mode

当用户指定 output: brief 时,输出精简版:

## 📖 《书名》 - 作者

**核心框架**: [F] 一句话总结核心机制
**破除常识**: [C] 被摧毁的常识信念
**应用到你**: [T] 一个具体行动项(映射用户上下文)
**盲区**: [E] 何时失效

💡 想看完整分析?说 "展开" 即可。
  • brief 模式同样执行完整的知识库写入流程(Step 4-6),只是输出给用户的部分精简
  • 用户说 "展开" 后,输出完整 full 模式内容

Depth Mode Output (合并报告)

当 depth mode 提取了多个模型时,合并为一份报告输出:

## 📖 深度解析:《书名》 - 作者

**提取模型数**: N 个 | **模式**: Depth

---

### 💎 模型 1: [Model Name]

**[F] 核心框架**: [一句话,≤80字]
**[A] 锚定案例**: [最经典案例,2-3句]
**[C] 破除常识**: ❌ "..." → ✅ ...
**[E] 失效边界**: [何时失效]
**[T] 迁移应用**: [映射用户上下文]

---

### 💎 模型 2: [Model Name]

(同上结构)

---

### 💎 模型 3: [Model Name]

(同上结构)

---

## 🔗 模型关联分析

| 模型 | 核心逻辑 | 适用场景 | 与其他模型的关系 |
|------|---------|---------|----------------|
| 模型1 | ... | ... | 与模型2互补 / 与模型3矛盾 |
| 模型2 | ... | ... | ... |
| 模型3 | ... | ... | ... |

## 🤔 综合战略拷问

[基于所有模型的综合视角,提出一个更深层的战略问题]

关键区别

  • 每个模型的 F.A.C.E.T. 用精简格式(各维度 1-3 句,不展开)
  • 新增「模型关联分析」表格 — 展示模型间的互补/矛盾/递进关系
  • 战略拷问基于所有模型的综合视角,而非单个模型

Step 4. Update Knowledge Base (更新知识库)

分类并存储提取的模型

分类决策树

提取的知识
├─ 能否在不同领域复用为决策工具? → YES → Thinking Pattern
├─ 是否是高度抽象的通用指导原则? → YES → Principle
├─ 是否是领域特定的知识/术语? → YES → Concept
└─ 边界模糊 → 标记多个 tags

三种分类

类型 定义 示例 写入位置
Thinking Pattern 可复用决策框架 颠覆性创新框架、逃离机制 patterns/{id}.md
Principle 高度抽象指导原则 二八法则、奥卡姆剃刀 patterns/{id}.md
Concept 领域特定知识 种痘术、能量密度天花板 concepts.md

一个条目可以同时标记多个类型(如 "杠铃策略" 既是 Thinking Pattern 又有 Concept 成分)。

写入格式

For Thinking Patterns / Principles (写入 memory/knowledge-base/patterns/{id}.md):

mental-model-forge 返回的 KB_META 块提取 frontmatter 字段,从 FACET 维度映射正文字段:

---
id: {from KB_META}
name_zh: {from KB_META}
name_en: {from KB_META}
source: {book_title}, {author}
category: {from KB_META}
tags: {from KB_META}
scenarios: {from KB_META}
related_models: {from KB_META}
difficulty: {from KB_META}
date: YYYY-MM-DD
---

**核心逻辑**:
{从 [F] Core Framework 提炼的一段话,比 Framework 更完整}

**思维框架**:
{直接使用 [F] Core Framework 内容}

**决策原则**:
{从 [F] + [E] 推导,格式:在XX场景下,应该XX而非XX}

**盲区警告**:
{直接使用 [E] Hidden Boundaries 内容}

**反射弧**:
{从 scenarios 推导,格式:看到XX信号 → 联想到模型 → 判断/行动}

**锚定案例**:
{直接使用 [A] Anchor Case 内容}

**反共识**:
{from KB_META contradiction field,格式:❌ "旧常识" → ✅ 新真相}

FACET → 知识库字段映射表

FACET 维度 知识库字段 映射方式
[F] Framework 核心逻辑 + 思维框架 核心逻辑=扩展版,思维框架=原文
[A] Anchor Case 锚定案例 直接使用
[C] Contradiction 反共识 直接使用
[E] Edge 盲区警告 直接使用
[T] Transfer 不写入知识库 仅用于用户简报
决策原则 从 [F]+[E] 提炼
反射弧 从 scenarios 推导

重要:[T] Transfer 是用户简报专用维度,包含个性化上下文(职业、项目、挑战),不写入知识库。每次生成简报时根据 USER.md 实时生成。

For Concepts (写入 concepts.md):

## [Concept Name] - [Book Title]

**定义 (Definition)**:
- [简洁定义]

**上下文 (Context)**:
- 这个概念在什么领域/场景重要?

**关联理论 (Related Theories)**:
- 与哪些思维框架相关?

**来源**: [Book Title] - [Author]
**日期**: YYYY-MM-DD

更新 last_attempted

"last_attempted": {
  "date": "YYYY-MM-DD",
  "book": "《XXX》",
  "step": "knowledge_base_write",
  "status": "success"
}

Step 4.5. Verify Knowledge Base Write (写入验证,必须执行)

验证逻辑

# 验证: 检查文件是否存在
ls ~/.openclaw/workspace/memory/knowledge-base/patterns/{id}.md

自检清单

  • patterns/{id}.md 文件存在?
  • □ 文件包含完整 YAML frontmatter(--- 开头和结尾)?
  • □ frontmatter 中 date 为当天?
  • □ 正文包含所有 7 个字段(核心逻辑、思维框架、决策原则、盲区警告、反射弧、锚定案例、反共识)?

如果验证失败 → 立即重新写入,再次验证。验证通过后才能继续 Step 5。

Step 5. Update Reading Records (更新阅读记录)

memory/reading-history.jsonused_models 数组追加新条目:

{
  "date": "YYYY-MM-DD",
  "book": "书名",
  "author": "作者",
  "model": "提取的思维模型名称",
  "category": "主题分类",
  "source": "web_search | queue | user_specified",
  "applied_count": 0,
  "tags": ["thinking-pattern"]
}

同时更新 last_attempted:

"last_attempted": {
  "date": "YYYY-MM-DD",
  "book": "《XXX》",
  "step": "reading_history_update",
  "status": "success"
}

错误恢复策略

  • 如果 Step 4(写入知识库)失败 → 不更新 reading-history,下次运行会重试同一本书
  • 如果 Step 5(本步骤)失败 → 知识库已写入但记录未更新,last_attempted 标记为 failed,下次运行时提醒用户手动补录

Step 6. Write to External Database (写入外部数据库)

[检查] 读取 HEARTBEAT-reading.md 获取数据库配置

  • Path: HEARTBEAT-reading.md
  • 寻找 ## 环境配置 section
  • 提取 Feishu App Token 和 Table ID
  • 如未找到 → 跳过本步骤(local-only mode)

如果配置存在,写入 Feishu Bitable

feishu_bitable_create_record({
  app_token: "{from HEARTBEAT-reading.md}",
  table_id: "{from HEARTBEAT-reading.md}",
  fields: {
    "日期": Date.now(),
    "书名": "《反脆弱》",
    "作者": "Nassim Nicholas Taleb",
    "模型名称": "反脆弱三元组",
    "分类": "Innovation",
    "核心框架(F)": "系统分三类:脆弱、坚韧、反脆弱...",
    "应用场景": "产品迭代、技能学习、风险管理",
    "战略拷问": "你的产品是在避免失败还是利用失败?"
  }
})

Notion Database (alternative):

  • Required: NOTION_API_KEY, NOTION_DATABASE_ID
  • Map fields accordingly

如果无凭证 → 跳过(skill 仍可本地使用)


reading-history.json Schema (v1)

统一 schema 定义:

{
  "schema_version": 1,
  "last_attempted": {
    "date": "2026-03-27",
    "book": "《反脆弱》",
    "step": "knowledge_base_write",
    "status": "success"
  },
  "queue": [
    {
      "title": "《穷查理宝典》",
      "author": "彼得·考夫曼",
      "topic": "决策科学"
    }
  ],
  "used_models": [
    {
      "date": "2026-03-24",
      "book": "《上瘾》",
      "author": "尼尔·埃亚尔",
      "model": "上瘾模型(Hook Model)",
      "category": "用户增长",
      "source": "web_search",
      "applied_count": 0,
      "tags": ["thinking-pattern"]
    }
  ]
}

字段说明

  • schema_version: 当前为 1,用于未来格式升级时的迁移判断
  • last_attempted: 上次运行的状态快照,用于错误恢复
  • queue: 用户预排的待读书籍队列(FIFO)
  • used_models: 已处理的所有模型记录(追加式,不可删除)
    • source: 标记选书来源(web_search | queue | user_specified
    • applied_count: 该模型被 AI 在后续对话中引用的次数(未来追踪用,初始为 0)
    • tags: 分类标签数组(thinking-pattern | principle | concept

迁移指引: 如果你已有旧格式的 reading-history.json(只有 used_models 数组,无 schema_version),只需手动添加以下顶层字段:

{
  "schema_version": 1,
  "last_attempted": null,
  "queue": [],
  "used_models": [... 保留原有数据 ...]
}

Depth Mode Configuration

默认: breadth(每次运行处理一本新书,提取 1 个模型)

切换方式: 用户在对话中指定 depth_mode: depth 或说 "深度分析这本书"

depth 模式详细流程:

选中书籍: 《反脆弱》
│
├─ Round 1: 提取 "反脆弱三元组" → 写入 patterns/antifragility.md + reading-history
├─ Round 2: 提取 "杠铃策略" → 写入 patterns/barbell-strategy.md + reading-history
├─ Round 3: 提取 "林迪效应" → 写入 patterns/lindy-effect.md + reading-history
├─ Round 4: AI 自评 "无更多独立框架" → 停止
│
├─ 合并输出: 一份报告包含 3 个模型(精简 F/A/C/E + 🎯迁移 + 关联分析)
└─ 写入飞书: 每个模型一条记录

三重退出条件(任一触发即停止):

  1. 模型数上限: 该书已提取 ≥ 5 个模型
  2. 语义去重: AI 判断新提取的模型与该书已提取模型本质相同(同一思想的变体或换皮表达)
  3. AI 自评: 提取后自问 "这本书还有独立的、值得提取的思维框架吗?",回答 No 则停止

Quality Standards

禁止:

  • 书籍摘要或作者传记
  • 泛泛之谈("这很重要"、"值得学习")
  • 重复已提取的模型(书名和模型名双重检查)
  • 文学评论式语言

要求:

  • 尖锐、可行动的语言
  • 具体案例(不是抽象概念)
  • 直接映射到用户上下文(不可脱离 USER.md)
  • 反书评口吻(不是"推荐阅读",而是"拿走就能用")

Configuration

配置来源优先级:

1. HEARTBEAT-reading.md (推荐)

  • Path: HEARTBEAT-reading.md
  • 可配置内容:
    • 主题映射覆盖
    • Feishu/Notion 凭证
    • 调度时间段

2. 环境变量 (备选)

  • FEISHU_APP_TOKEN, FEISHU_TABLE_ID
  • NOTION_API_KEY, NOTION_DATABASE_ID

3. 默认值 (兜底)

  • 使用默认星期-主题映射
  • 无外部数据库集成(local-only mode)

用户上下文 (可选但强烈推荐):

  • Path: USER.md
  • 用于: 个性化 [T] Transfer、应用场景、战略拷问
  • 缺失时: 使用通用第二人称,可能追问用户背景

知识库路径 (自动创建):

  • Thinking Patterns: memory/knowledge-base/patterns/*.md (每个模型一个文件)
  • Concepts: memory/knowledge-base/concepts.md

References


Version: 3.0.0 Last updated: 2026-03-28 Changes: Knowledge base restructure — single-file-per-model with YAML frontmatter (patterns/{id}.md), dedup via reading-history.json only (no longer reads thinking-patterns.md), FACET→KB field mapping table, KB_META extraction from mental-model-forge output, contradiction field added

Usage Guidance
This skill is coherent with its stated purpose: it reads a small set of workspace files, calls two other skills to find and analyze books, and writes markdown model files to a knowledge-base. Before installing or enabling automatic runs: 1) Review and trust the two dependencies (book-scout, mental-model-forge) because they will be invoked and may perform web requests. 2) Only provide FEISHU/NOTION tokens if you want the optional external posting feature; treat those as sensitive credentials. 3) Be cautious about applying the README’s optional AGENTS.md edits — auto-loading SOUL.md or recent memory files each session will make more of your workspace readable to the agent. 4) Backup memory/reading-history.json and check its contents for any sensitive data you don’t want the skill to read. 5) If you need higher assurance, inspect (or request) the implementations of book-scout and mental-model-forge and run the skill in a restricted workspace first. These steps will reduce privacy/supply-chain exposure while keeping the skill’s useful behavior.
Capability Analysis
Type: OpenClaw Skill Name: cognitive-forge Version: 1.0.4 The cognitive-forge skill is a sophisticated knowledge management tool designed to extract mental models from books using F.A.C.E.T. analysis. It maintains a local knowledge base in the workspace (memory/knowledge-base/) and optionally syncs data to external platforms like Feishu or Notion using provided API keys. While it requests broad permissions—including reading user profiles (USER.md) for personalization and accessing environment variables for database integration—all behaviors are clearly documented and strictly aligned with its stated purpose of building a 'latticework of mental models.' The skill includes robust logic for deduplication, quality verification, and error recovery, with no evidence of malicious intent or unauthorized data exfiltration.
Capability Assessment
Purpose & Capability
The name/description (building a library of mental models) matches the runtime instructions: selecting books, running F.A.C.E.T. analyses, deduplicating, and writing per-model markdown files under memory/knowledge-base. Declared dependencies (book-scout and mental-model-forge) are consistent with the task.
Instruction Scope
Instructions read USER.md and memory/reading-history.json and write patterns and concept files—these are appropriate for personalization and deduplication. However the README and SKILL.md recommend optional startup behavior (edit AGENTS.md to load SOUL.md, USER.md, and recent memory files each session). That optional guidance expands what the agent will read (potentially sensitive session memory) and should be applied only intentionally.
Install Mechanism
This is an instruction-only skill with no install spec and no code files—lowest disk/write risk. Dependencies are other skills (book-scout, mental-model-forge) that must be installed separately; verify their provenance before installing.
Credentials
The skill references optional external integration tokens (FEISHU_APP_TOKEN, FEISHU_TABLE_ID, NOTION_API_KEY, NOTION_DATABASE_ID) for posting analysis to external tracking systems. Requesting those optional credentials is proportionate to the described optional feature (external tracking), but you should not provide them unless you want the skill to send data to Feishu/Notion.
Persistence & Privilege
The skill writes persistent files under memory/knowledge-base and reading-history.json as intended for a knowledge-base. It does not set always:true and does not request system-wide privileges. The only persistence concern is the README's suggestion to add auto-loading of many memory files at session startup; enabling that increases the skill's effective access surface and should be done consciously.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install cognitive-forge
  3. After installation, invoke the skill by name or use /cognitive-forge
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.4
**Cognitive Forge 1.0.4 — Key Update: Model files now use individual Markdown files with YAML frontmatter.** - Each mental model is written to its own file (`patterns/{id}.md`) with YAML frontmatter, instead of being appended to `thinking-patterns.md`. - Knowledge base status/statistics now count `.md` files in `patterns/` for model total. - Model de-duplication uses book titles only; does not check or parse `patterns/`. - Permissions and config updated for new file structure. - All workflows, review, and status logic modernized to reference the new per-model file system.
v1.0.3
No functional changes; update clarifies dependency roles and skill documentation. - Clarified the purpose and typical usage of dependencies (`book-scout`, `mental-model-forge`) in SKILL.md. - Expanded dependency comments to specify which workflow steps each is called in. - No changes to logic, features, configuration, or permissions.
v1.0.2
**Cognitive Forge 1.0.2 – Major Enhancement for Flexibility, Coverage, and Workflow Robustness** - Added support for both breadth and depth modes: analyze one or multiple models per book per user request or schedule. - Enhanced routing: new status and review branches for knowledge base stats and spaced repetition of models. - Expanded file read/write coverage: added `concepts.md`, heartbeat scheduling, and user path convention. - Improved book selection: prioritizes user-specified/queue/books, topic mapping via heartbeat file, and dual-layer deduplication by book and model. - Built-in structured quality verification for model extraction, with automatic retries for low-quality results. - Enforced first-run environment self-checks and automatic creation of needed files/folders. - All outputs now rigorously map learning to user's real context if available; fallback to generic "you" if no profile. - More robust failure, error, and retry handling across all steps. - Knowledge base output now includes statistics, recent history, and coverage recommendations.
v1.0.1
No user-facing changes detected in this version. - No modifications to functionality, documentation, or configuration. - All features and skill behaviors remain unchanged.
v1.0.0
Dual-purpose learning system - AI self-evolution: extracts reusable mental models from books, writes to thinking-patterns.md for permanent cognitive upgrades. Your AI gets smarter with every book. - Enables AI to read classic books, extract reusable mental models, and build a permanent thinking-patterns.md library. - Provides users with sharp F.A.C.E.T. analyses (Framework, Anchor Case, Contradiction, Edge, Transfer) for deep comprehension of each book. - Integrates with book-scout and mental-model-forge skills for book selection and model extraction. - Deduplicates previously read books via reading-history.json. - Personalizes insights and applications using user context from USER.md.
Metadata
Slug cognitive-forge
Version 1.0.4
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 5
Frequently Asked Questions

What is AI Self-Evolution Engine( AI 自我进化引擎)?

Dual-value learning system - extracts reusable mental models from books, writes individual pattern files (patterns/{id}.md) with YAML frontmatter for buildin... It is an AI Agent Skill for Claude Code / OpenClaw, with 193 downloads so far.

How do I install AI Self-Evolution Engine( AI 自我进化引擎)?

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

Is AI Self-Evolution Engine( AI 自我进化引擎) free?

Yes, AI Self-Evolution Engine( AI 自我进化引擎) is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does AI Self-Evolution Engine( AI 自我进化引擎) support?

AI Self-Evolution Engine( AI 自我进化引擎) is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created AI Self-Evolution Engine( AI 自我进化引擎)?

It is built and maintained by KeDouPi(珂抖屁) (@kedoupi); the current version is v1.0.4.

💬 Comments