Distill Jobs, Munger, or Feynman into an agent skill

What it does

女娲 (Nuwa) is a skill for Agent Skills–compatible runtimes (Claude Code, Codex, Cursor, DeepSeek Harness, and 50+ others). Give it a name — Jobs, Musk, Munger, Feynman — and it runs the full pipeline: research, extraction, verification, producing an installable skill that answers questions as that person. The repo ships 14 pre-distilled people plus one topic skill, each in its own repo with research data and sample conversations.

The interesting bit

The pitch is that this isn’t roleplay or quote-recitation. Nuwa extracts five layers — expression DNA, mental models, decision heuristics, anti-patterns, and honest limits — so a distilled Jobs answers using “focus means saying no” as a framework, not a catchphrase. Each skill also declares what it can’t do: no intuition, no post-research changes, public statements ≠ real thoughts. A skill that won’t tell you its limits isn’t trustworthy, says the README — a rare and welcome stance.

Key highlights

• Runs on any Agent Skills protocol runtime; it’s fundamentally just markdown + YAML frontmatter, so you can even paste SKILL.md into a chat directly
• 15 official skills passed an independent two-agent blind test across stance consistency, style recognizability, and edge honesty — all graded A (≥85)
• Distilled output lands in your project’s .dsh/skills/ (DeepSeek Harness) or the runtime’s skills directory and auto-loads
• Community distillations are indexed in COMMUNITY.md; contributors keep their own repos and stars
• Companion project Darwin-skill exists to keep generated skills evolving over time

Caveats

• SKILL.md is the core asset and accepts no external PRs — improvements go through issues, maintainer-implemented
• Fidelity scores are self-defined methodology; the “A grade” claims aren’t independently verifiable from the README alone
• The whole thing is only as good as public information about a person — the README is upfront about this

Verdict

Worth a look if you use a skills-compatible agent and want structured second opinions from famous thinkers’ frameworks. If you were hoping for actual insight from the real Jobs, the honest-limits section already told you that’s not on offer.