openclaw-master-skills/skills/cognitive-forge

Cognitive Forge (认知锻造)

Your AI reads books, extracts mental models, and builds a growing library of decision frameworks. The more books processed, the more compound thinking ability your AI gains.

---

What Makes This Different

Most AI skills give you answers. Cognitive Forge builds your AI's decision framework library.

Every book processed → One new mental model → Permanently stored in thinking-patterns.md.

After 100 books, your AI has 100 reusable frameworks (Taleb's antifragility, Munger's mental models, Kahneman's dual-process theory, etc.) that it can reference across ANY domain.

Building compound thinking, one book at a time.

---

One Run, Dual Value

Each run produces two outputs simultaneously:

1. For You: F.A.C.E.T. Analysis

Sharp, actionable analysis for deep book comprehension:

• [F] Framework: Core mechanism in 50 words (not what author said, but what theory DOES)
• [A] Anchor Case: Most iconic real-world example (vivid stories stick)
• [C] Contradiction: What "common sense" does this destroy?
• [E] Edge: When does this model fail? Fragile assumptions?
• [T] Transfer: Map to YOUR reality TODAY (personalized to your job/projects)

Not a book summary — a battle-tested mental model you can apply immediately.

2. For Your AI: Knowledge Base Entry

The extracted model is permanently written to thinking-patterns.md. In future sessions, your AI can reference this framework when answering complex questions.

---

Quick Start

1. Install Dependencies

This skill requires:

• book-scout — Finds high-quality books via web search
• mental-model-forge — Applies F.A.C.E.T. analysis

# Install from ClawHub
clawhub install kedoupi/book-scout
clawhub install kedoupi/mental-model-forge

# Or manually copy to ~/.openclaw/workspace/skills/

2. Tell Your AI to Use It

Daily learning (recommended):

Ask your AI: "Run cognitive-forge daily at 8:30 AM. Topic: Business Strategy."

Your AI will:

1. Find a high-quality book on the topic (Douban ≥7.5 or Goodreads ≥3.8)
2. Extract the core mental model (F.A.C.E.T. analysis)
3. Write it to thinking-patterns.md (permanent storage)
4. Deliver a summary to you

One-time book processing:

Ask your AI: "Use cognitive-forge to analyze: 'Thinking, Fast and Slow' by Daniel Kahneman"

First run: The skill auto-creates all necessary files and directories. No manual setup needed.

---

How It Works

The Permanent Upgrade Loop

┌─────────────────────────────────────────────────────┐
│  1. Book Selection                                   │
│     ├── Priority: Direct specify > User queue > Search │
│     ├── Filter: Douban ≥7.5 OR Goodreads ≥3.8        │
│     └── Dedup: Book title + Model name (dual-layer)  │
└─────────────────────────────────────────────────────┘
                        ↓
┌─────────────────────────────────────────────────────┐
│  2. F.A.C.E.T. Analysis + Quality Verification      │
│     ├── [F] Extract core framework                   │
│     ├── [A] Find iconic case study                   │
│     ├── [C] Identify contrarian insight              │
│     ├── [E] Map failure modes                        │
│     ├── [T] Transfer to user's context               │
│     └── Self-check: Score ≥7/10 or retry             │
└─────────────────────────────────────────────────────┘
                        ↓
┌─────────────────────────────────────────────────────┐
│  3. Write to thinking-patterns.md (PERMANENT)        │
│     ├── Classify: Pattern / Principle / Concept      │
│     ├── Tag with multiple categories                 │
│     └── Verify write with date + model name check    │
└─────────────────────────────────────────────────────┘
                        ↓
┌─────────────────────────────────────────────────────┐
│  4. Your AI Can Now Reference This Model             │
│     When you ask for advice in ANY domain:           │
│     "Should I invest in X?"                          │
│     → AI checks thinking-patterns.md                 │
│     → Applies relevant frameworks                    │
│     → Gives strategically grounded insight            │
└─────────────────────────────────────────────────────┘

---

Features

Breadth vs Depth Mode

Mode	Behavior	Use When
Breadth (default)	1 book → 1 model	Daily learning, covering diverse topics
Depth	1 book → up to 5 models	Deep-diving into a rich book (e.g., Antifragile)

Depth mode extracts multiple models from one book until:

• 5 models reached, OR
• New model is a variant of an existing one, OR
• AI determines no more independent frameworks remain

Usage: "深度分析这本书" or depth_mode: depth

Multi-Source Book Selection

Three sources (priority order):

1. Direct specify: "Analyze Thinking, Fast and Slow"
2. User queue: Pre-queue books in reading-history.json
3. Web search: book-scout finds high-quality books automatically

Configurable Topic Mapping

Default weekly rotation:

Mon: Business Strategy | Tue: Psychology | Wed: Technology
Thu: Economics | Fri: Innovation | Sat: Philosophy | Sun: Biography

Override in HEARTBEAT-reading.md to match your learning goals.

Brief / Full Output

Mode	Content	Default
Full	Complete F.A.C.E.T. + scenarios + counter-example + strategic question	Yes
Brief	4-line summary with "expand" option	User requests output: brief

Knowledge Library Status

Ask: "cognitive-forge status"

Returns: total models, category distribution, weak areas, recent entries.

Weekly Review (Spaced Repetition)

Ask: "cognitive-forge review"

Quizzes you on 2-3 models from the past week. Also auto-triggers on Sundays if configured.

Error Recovery

If a run fails mid-way, the next run detects the failure and offers to resume from where it stopped.

---

How to Maximize Value

1. Enable Auto-Scanning in AGENTS.md

Add this to your AGENTS.md (optional):

## Session Startup

1. Read `SOUL.md` — this is who you are
2. Read `USER.md` — this is who you're helping
3. Read `memory/YYYY-MM-DD.md` (today + yesterday) for recent context
4. **Load `memory/knowledge-base/thinking-patterns.md`** — your decision frameworks

Why: Every session, your AI loads its accumulated mental models. When answering complex questions, it references thinking-patterns.md for relevant frameworks.

Example:

You: "Should I pivot my startup to B2B?"

Your AI:
1. Reads thinking-patterns.md
2. Finds: "Clayton Christensen's Jobs-to-be-Done"
3. Finds: "Andy Grove's Strategic Inflection Points"
4. Applies both frameworks to your question
5. Gives a strategically grounded answer

2. Schedule Daily Learning

Weekly rotation example:

Monday: Business Strategy
Tuesday: Psychology/Decision Science
Wednesday: Product/Design Thinking
Thursday: Finance/Economics
Friday: Philosophy/Systems Thinking
Saturday: Technology/Innovation
Sunday: Review week's mental models

Diverse mental models → Better cross-domain thinking.

openclaw cron create --name "Daily Book" --schedule "30 8 * * *" \
  --task "Run cognitive-forge. Topic: [based on weekday mapping]"

3. Force Your AI to Apply Models

Generic ask (doesn't use mental models):

"Should I hire more people?"

Strategic ask (forces model application):

"Apply mental models from thinking-patterns.md: Should I hire more people?
Consider: Brook's Law, Two-Pizza Teams, Scaling Challenges."

---

Sample Output

Book: Thinking, Fast and Slow by Daniel Kahneman

F.A.C.E.T. Analysis:

[F] Framework: Dual-Process Theory
Human cognition operates via two systems:
- System 1: Fast, automatic, emotional (95% of decisions)
- System 2: Slow, deliberate, rational (used only when forced)
Core: System 1 dominates by default → cognitive biases emerge when System 2 fails to override.

[A] Anchor Case: The Linda Problem
Most people choose the more specific description (conjunction fallacy).
System 1 says "the story fits!" System 2 (if engaged) knows the math.

[C] Contradiction
"People make rational decisions when given good information."
→ NO. Even experts rely on System 1 and commit systematic errors.

[E] Edge
Fails when: experts in narrow domains (chess masters' intuition IS reliable),
time-critical survival decisions, culturally variable contexts.

[T] Transfer (personalized)
For your AI product at 爱康国宾:
1. User onboarding → Design for System 1 (one-click, zero thinking)
2. Critical health decisions → Force System 2 (add friction before confirming)
3. Team decisions → Ask "What System 1 biases made us love this feature?"

---

Use Cases

1. Build a Personal Decision Framework Library

• Who: Founders, investors, strategists
• How: Process 1 book/week for 1 year = 52 mental models
• Value: Every future decision taps into 52 frameworks

2. Team Learning System

• Who: Product teams, research labs
• How: Rotate book topics weekly, share F.A.C.E.T. analyses
• Value: Shared mental model language → Better collaboration

3. AI-Powered Reading Coach

• Who: Lifelong learners
• How: Your AI reads books you don't have time for, extracts models
• Value: You get the wisdom without reading 300 pages

4. Deep-Dive Single Books

• Who: Anyone studying a specific author or domain
• How: Use depth mode to extract 3-5 models from one rich book
• Value: Thorough extraction, no framework left behind

---

File Structure

cognitive-forge/
├── SKILL.md                       # Main workflow definition
├── README.md                      # This file
└── references/
    ├── book-selection.md          # Multi-source selection + configurable mapping
    ├── example-output.md          # Full and brief mode examples
    └── knowledge-classification.md # Three-type classification with tag system

---

How This Differs from "Book Summary" Tools

Feature	Cognitive Forge	Traditional Summary Tools
Goal	Extract reusable mental model	Condense book content
Focus	Transferable frameworks	Facts and key points
Permanence	Written to thinking-patterns.md (forever)	Forgotten after reading
Application	Your AI references it across domains	You read it once
Personalization	[T] Transfer maps to YOUR context	Generic for everyone
Longevity	Models compound over time	Summaries don't interact
Quality	Self-verified (score ≥ 7/10)	No quality gate

---

Advanced Usage

Multi-Book Cross-Analysis

Ask your AI: "Process 3 books on decision-making:
1. Thinking, Fast and Slow (Kahneman)
2. Superforecasting (Tetlock)
3. The Black Swan (Taleb)

Then cross-analyze: Where do their mental models agree? Conflict? Complement?"

Domain-Specific Model Curation

For product managers:

Schedule: Process 1 product/UX book per week
Topics: Jobs-to-be-Done, Hooked Model, Lean Startup, Design Thinking
Result: thinking-patterns.md becomes your PM playbook

For investors:

Topics: Mental Models (Munger), Antifragile (Taleb), Capital (Piketty)
Result: thinking-patterns.md becomes your investment decision framework

Feishu/Notion Integration (Optional)

If you set FEISHU_APP_TOKEN or NOTION_API_KEY, the skill auto-uploads:

• Book title, F.A.C.E.T. analysis, reading date, mental model tags

Benefit: Build a searchable knowledge base across your team.

---

Troubleshooting

"Book-scout can't find books on my topic":

• Try broader topics (e.g., "Psychology" instead of "Behavioral Economics of Crypto")
• Or specify a book directly: "Analyze: [Book Title] by [Author]"
• Or add books to your queue: "Add [Book Title] to my reading queue"

"F.A.C.E.T. analysis is too shallow":

• The skill now self-verifies quality (score ≥ 7/10) and retries automatically
• For deeper analysis: "Deep F.A.C.E.T. analysis with multiple examples"
• Claude Opus recommended for best results

"How do I see what mental models my AI has learned?":

# Check your knowledge library
cat ~/.openclaw/workspace/memory/knowledge-base/thinking-patterns.md

# Or use the status command
Ask: "cognitive-forge status"

"My AI isn't applying learned models":

• Check if thinking-patterns.md is loaded in AGENTS.md
• Explicitly ask: "Apply mental models from thinking-patterns.md to [question]"

"A run failed halfway":

• Next run will auto-detect and offer to resume
• Check reading-history.json → last_attempted field for details

---

Important Notes

Data Persistence: This skill writes to thinking-patterns.md and reading-history.json in your workspace. These files persist across sessions.

Optional Integration: The "auto-load thinking-patterns.md" feature requires manual configuration in your AGENTS.md. The skill does NOT automatically modify your agent's behavior without your explicit setup.

Export Tokens: Environment variables (FEISHU_APP_TOKEN, NOTION_API_KEY) are entirely optional. The skill works perfectly fine without them (local-only mode).

First Run: All necessary files and directories are auto-created on first run. No manual setup required beyond installing dependencies.

---

License

MIT-0 (No Attribution Required)

---

Credits

Created by 汤圆 (Tangyuan) for 雯姐's learning journey.

Inspired by:

• Charlie Munger's "latticework of mental models"
• Nassim Taleb's antifragility & skin in the game
• Daniel Kahneman's dual-process theory
• Shane Parrish's Farnam Street framework

Built with:

• book-scout — Quality book discovery
• mental-model-forge — F.A.C.E.T. analysis engine

---

Feedback

Report issues or share your experience at ClawHub.

If you've processed 50+ books and want to share what your AI has learned, ping @KeDOuPi on ClawHub.