Cyber Nuwa

Problem

Turning a person's way of thinking into an executable AI skill is hard to trace when the result merely restates success advice. Without sources, failure conditions, and refusal boundaries, the output reads like generic guidance rather than a stable, reusable thinking framework.

How It Works

Cyber Nuwa focuses on extracting thinking frameworks, capturing HOW they think rather than replaying WHAT they said. Key steps include:
- Intake routing: direct clarification for named targets; diagnosis and candidate selection for vague needs.
- Multi-source research: parallel collection of writings, conversations, expression style, outside views, decision cases, and timelines.
- Framework extraction: 3-7 mental models, 5-10 decision heuristics, expression DNA, values, and anti-patterns.
- Quality validation: eight checks via quality_check.py, with L1/L2/L3 degradation and honest boundaries when evidence is weak.

Boundaries

Best suited for targets with rich first-hand public materials and clear style/decision boundaries. Avoid cases with scarce material, confused targets, or requests to invent unpublished views. The output is a structured SKILL.md; validate it with known, edge, style, and refusal tests.