Building AI Agents with Distilled Thinking Frameworks

What Are Thinking Framework Skills?

Thinking framework skills (or “distilled skills”) are executable AI agent modules that encapsulate an individual’s decision-making patterns. Unlike traditional prompt-based roleplaying, these skills:

• Represent verified mental models rather than surface-level opinions
• Include structured components: core heuristics, expression patterns, value hierarchies, and anti-patterns
• Maintain clear boundaries about what the framework can and cannot address

The open-source Nuwa Skill toolkit (repository: alchaincyf/nuwa-skill) enables this distillation process. Install via:

npx skills add alchaincyf/nuwa-skill

The Distillation Methodology

To create a valid thinking framework skill, the source material must pass three validation criteria:

1. Cross-domain consistency – The principles must recur across different contexts in the subject’s work/life
2. Generative power – The framework should enable solving novel problems, not just explain past decisions
3. Exclusive insight – The thinking pattern should offer distinct advantages over conventional approaches

Example distilled frameworks:

• Elon Musk’s “Asymptotic Limit” – Start with physics-first principles, quantify current performance gaps, then eliminate non-essential steps
• Charlie Munger’s inversion – “Tell me where I’ll die so I can avoid going there” approach to risk assessment
• Richard Feynman’s clarity test – If you can’t explain a concept simply, you don’t understand it well enough

Creating Your First Distilled Skill

To distill a new thinking framework:

1. Gather source material – Collect at least 50,000 words of primary sources (interviews, writings, decision records)
2. Identify repeating patterns – Look for:
   • Decision heuristics used across different domains
   • Consistent evaluation criteria
   • Repeated warnings or anti-patterns
3. Structure the framework – Organize into:
   • Core mental models (3-5 fundamental principles)
   • Decision-making algorithms (step-by-step heuristics)
   • Expression DNA (characteristic communication patterns)
4. Define boundaries – Explicitly document:
   • Contexts where the framework applies
   • Known limitations
   • Edge cases requiring alternative approaches
5. Package as a skill – Use Nuwa Skill’s template:

npx skills create yourname/subject-skill

Key Differences from Roleplay Prompts

Traditional “act as X” prompts differ from distilled skills in critical ways:

Dimension	Roleplay Prompt	Distilled Skill
Source	Surface impressions	Verified recurring patterns
Structure	Ad-hoc responses	Documented decision pathways
Boundaries	Unlimited claims	Scoped applicability
Output	Plausible-sounding text	Framework-driven analysis

The test: Can the output analyze a novel problem the original person never addressed?

Practical Applications

Five high-value uses for distilled thinking frameworks:

1. Industry mentor simulation – Run decisions through top performers’ frameworks (e.g., product development via distilled Jobs framework)
2. Domain expert lenses – Apply specialized thinking patterns to your problems (e.g., Taleb’s convexity strategies for risk assessment)
3. Competitor analysis – Understand rivals’ decision logic from public materials
4. Methodology distillation – Capture domain-specific approaches (e.g., “writing frameworks” rather than individual authors)
5. Comparative analysis – Contrast different frameworks on the same problem (e.g., Musk vs. Jobs on product strategy)

Implementation Notes

The Nuwa Skill ecosystem currently includes eight verified frameworks:

1. Steve Jobs (product design/strategy)
2. Elon Musk (engineering/first principles)
3. Charlie Munger (investment/inversion)
4. Richard Feynman (learning/science)
5. Naval Ravikant (wealth/leverage)
6. Nassim Taleb (risk/antifragility)
7. Xuefeng Zhang (education/information asymmetry)
8. X-mentor (social media operations)

Install any via:

npx skills add alchaincyf/[framework-name]-skill

Check the official repository for the latest additions and version updates.

(Figures and case details in this article are as reported in the source material and have not been independently verified.)

Series: AI Practice