# Nuwa Skill — Distill Anyone's Thinking into an AI Skill

> An open-source tool that extracts mental models, decision heuristics, and communication patterns from any person's content, then packages them as reusable AI agent skills.

## Install

Save as a script file and run:

# Nuwa Skill — Distill Anyone's Thinking into an AI Skill

## Quick Use
```bash
git clone https://github.com/alchaincyf/nuwa-skill.git
cd nuwa-skill
pip install -r requirements.txt
cp .env.example .env
# Add your LLM API key to .env
python main.py --source "path/to/content" --output skill.md
```

## Introduction
Nuwa Skill is an open-source framework for extracting structured thinking patterns from a person's written or spoken content and packaging them as reusable AI agent skills. It targets anyone who wants to capture expertise — their own or someone else's — in a format that LLM agents can directly consume and apply.

## What Nuwa Skill Does
- Analyzes a corpus of content (articles, transcripts, posts) to identify recurring mental models and decision heuristics
- Extracts communication patterns including tone, structure, and rhetorical preferences
- Generates a structured skill file that encodes the person's reasoning style as agent instructions
- Supports multiple input formats including text files, URLs, and transcript exports
- Produces skills compatible with Claude Code, Cursor, and other agent harnesses that accept markdown-based skill definitions

## Architecture Overview
Nuwa Skill operates as a multi-stage pipeline. The ingestion layer normalizes diverse input formats into plain text chunks. An analysis stage uses LLM calls to identify patterns across the corpus — recurring frameworks, decision criteria, writing conventions, and domain-specific vocabulary. A synthesis stage consolidates these patterns into a coherent skill definition with structured sections for context, triggers, instructions, and examples. The final output is a markdown file conforming to common agent skill formats. The pipeline is implemented in Python and uses configurable LLM backends for the analysis and synthesis steps.

## Self-Hosting & Configuration
- Clone the repository and install dependencies with pip; runs on any system with Python 3.10+
- Configure your preferred LLM provider (OpenAI, Anthropic, or compatible APIs) via environment variables
- Point the tool at a directory of source content or provide URLs for web-based content
- Adjust extraction parameters such as chunk size and analysis depth in the configuration file
- Output skill files can be placed directly into `.claude/skills/` or equivalent directories for immediate use

## Key Features
- Content-agnostic extraction works with blog posts, podcast transcripts, forum replies, and long-form writing
- Produces portable skill definitions that work across multiple agent platforms
- Iterative refinement mode lets you review and adjust extracted patterns before finalizing
- Supports batch processing of large content libraries for prolific authors
- Open architecture allows adding custom analysis modules for domain-specific pattern extraction

## Comparison with Similar Tools
- **Custom GPTs (OpenAI)** — require manual prompt writing; Nuwa Skill automates pattern extraction from existing content
- **Claude Projects** — let you upload reference docs for context; Nuwa Skill goes further by distilling actionable instructions from the content
- **Fabric (Daniel Miessler)** — provides pre-built prompt patterns; Nuwa Skill generates personalized patterns from individual content
- **MemGPT / Letta** — focus on long-term memory for agents; Nuwa Skill focuses on capturing reasoning style rather than facts
- **Prompt engineering by hand** — time-consuming and inconsistent; Nuwa Skill systematizes the extraction process

## FAQ
**Q: What kind of content works best as input?**
A: Content where the author explains their reasoning — blog posts with opinions, podcast transcripts with discussions, and advisory writing — yields the richest skill extractions.

**Q: Does it work with non-English content?**
A: Yes, as long as the underlying LLM supports the language. The extraction pipeline is language-agnostic; quality depends on the LLM's proficiency in that language.

**Q: How is this different from just uploading documents to an AI chat?**
A: Uploading documents provides raw context. Nuwa Skill distills that context into structured instructions — triggers, decision rules, and output formats — that an agent can follow without re-reading the source material.

**Q: Can I merge skills from multiple people?**
A: The tool generates one skill per corpus. You can run it on separate content sets and use multiple skills simultaneously in agents that support skill stacking.

## Sources
- https://github.com/alchaincyf/nuwa-skill
- https://github.com/alchaincyf/nuwa-skill/blob/main/README.md

---
Source: https://tokrepo.com/en/workflows/asset-c8684f62
Author: Script Depot