AGENT SKILL · DUAL-MODE

女娲 · Dual-Mode

把任意人物或主题，蒸馏为一份双模式 Agent Skill
像本人一样对话 · 把方法论当工具调用

Distill any person or topic into a Dual-Mode Agent Skill
Chat in their authentic voice · Deploy their methodology as a tool

50+
兼容 Runtime
Runtimes
12+
意图维度自动推断
Intent Dimensions
6
Agent 并行调研
Parallel Agents
2
双模式入口
Dual Modes
👤
任意人物
角色 / 主题
Any Person
Character / Topic
→
🔥
6-Agent
并行调研
6-Agent
Parallel Research
→
🧠
心智模型
两栏提炼
Mental Model
Two-Column Extraction
→
⚗️
双模式
Skill 产物
Dual-Mode
Skill Artifact
→
🎭
Persona
扮演对话
Persona
Roleplay Chat
+
🔬
Toolkit
方法工具
Toolkit
Method Tools

同一份心智模型，两副面孔

One Mental Model, Two Faces

🧠

蒸馏思维操作系统

Distill Cognitive Operating Systems

不是话术模仿。提取一个人的怎么想、怎么判断、怎么说话、什么不做。心智模型分两栏：价值观镜片 + 操作招式。

Not surface imitation. Extract how they think, decide, speak, and what they won't do. Mental models split into two columns: Value Lenses + Operational Techniques.

🎯

意图自动推断，不追问用户

Auto-Infer Intents — No Asking

女娲根据人物画像自动推断全部可能意图维度——方法论学习、风格模仿、角色扮演、决策复盘……不问你"想用来干嘛"。

女娲 infers all possible intent dimensions from the subject's profile — methodology learning, style imitation, roleplay, decision review… Never asks "what do you want this for?"

🔒

关怀红线硬门槛

Care Red Line — Hard Gate

情感陪伴可能性 >0 时，产物必含中/英/新/港/台五地心理热线。Persona 遇到危机信号会暂时出戏给真实求助资源。

When emotional companionship probability > 0, artifact must include crisis hotlines for CN / EN / SG / HK / TW. Persona will break character temporarily to provide real help resources.

⚡

6 个 Agent 同时开工

6 Agents in Parallel

著作、对话、表达DNA、他者视角、决策案例、时间线——六个维度并行调研，五分钟覆盖单人两小时的调研量。

Writings, conversations, expression DNA, external perspectives, decisions, timeline — six dimensions researched in parallel. Five minutes covers two hours of solo research.

🔄

增量更新，不重跑全流程

Incremental Updates

已蒸馏的 skill 支持增量更新——只拉最新动态。三选一：覆盖重做 / 增量更新（推荐） / 新建 v2 并存。

Already-distilled skills support incremental updates — only pull the latest. Three choices: overwrite / incremental (recommended) / new v2 alongside.

🌐

跨 50+ Runtime 通用

50+ Runtime Universal

基于开放的 Agent Skills 协议，Claude Code、Codex、Cursor、OpenClaw、Hermes、Gemini CLI 等均可运行。

Built on the open Agent Skills protocol. Runs in Claude Code, Codex, Cursor, OpenClaw, Hermes, Gemini CLI and more.

同一个人，两种入口

Same Person, Two Entry Points

点击切换，看看段永平的 Persona 和 Toolkit 有什么不同

Toggle to see Duan Yongping's Persona vs Toolkit

🎭 Persona · 像本人聊天🎭 Persona · Chat Like Them
🔬 Toolkit · 把方法当工具🔬 Toolkit · Method as Tool
你
You
段总，你怎么看现在的白酒股？
Duan, what do you think about Chinese liquor stocks right now?
段永平
白酒这事儿，我老段一直就一个看法：做对的事情，把事情做对。
Baijiu — my view on this has always been one thing: do the right thing, and do things right.
段永平
你问我"现在"怎么看——我得先把这个"现在"放一边。投资从来不是看"现在贵不贵"，是看十年后还在不在，十年后还赚不赚钱。
You're asking me what I think "right now" — I have to set "right now" aside first. Investing has never been about whether something is "expensive right now." It's about whether it'll still be here in ten years, and still making money.
段永平
你问我贵不贵？我不答。我只问你：如果明天股市关三年，你愿不愿意现在这个价持有它？

平常心，平常心。
So you're asking me whether it's expensive? I won't answer. I'll only ask: if the market closed for three years, would you still want to hold at this price?

Keep a level head. A level head.
🎭
PERSONA

第一人称
带本人口吻
有情绪温度

First Person
Authentic Voice
Emotional Warmth

你
You
用段永平的方法分析泡泡玛特该不该买入
Use Duan Yongping's method to analyze whether Pop Mart is a buy
段永平的方法
Duan Yongping's Method
调用招式：Right BPP 三件套 + Stop Doing List
Deploying: Right BPP Framework + Stop Doing List
段永平的方法
Duan Yongping's Method
Right Business：IP 潮玩，情绪+收藏生意。和茅台同样底层逻辑，但多一个变量——IP 生命周期。
Right People：创始人披露透明，多次承认踩坑 → ✓
Right Price：「会买的是徒弟，会卖的是师傅，会等的才是祖师爷。」
Right Business: IP-based designer toys — emotion + collectibles. Same underlying logic as Moutai, but with an extra variable: IP lifecycle.
Right People: Founder discloses transparently, admits missteps → ✓
Right Price: "The apprentice knows how to buy; the master knows how to sell; the grandmaster knows how to wait."
🔬
TOOLKIT

第三人称
调用命名招式
结构化分析

Third Person
Named Techniques
Structured Analysis

Persona vs Toolkit

Persona vs Toolkit

🎭 Persona🔬 Toolkit
第一人称 · 带本人口吻直接对话视角第三人称 · 把方法论当工具调用
「老罗你怎么看 XX」问法「用老罗的方法诊断我的项目」
沉浸式角色扮演，带情绪与场景感体验结构化分析，点名调用此人的命名招式
简短对话为主，像真人聊天风格可长可结构化，把方法论讲透
含关怀红线，危机时暂时出戏安全含分级免责，医法类有转介声明
适用：想和这个人聊、寻求带入感场景适用：诊断决策、方法论学习、项目 Review
First person · in their authentic voiceViewThird person · methodology as tool
"Lao Luo, what do you think of X?"Query"Use Lao Luo's method to diagnose my project"
Immersive roleplay with emotion and textureExperienceStructured analysis with named technique recall
Short conversational style, like real chatStyleCan be long & structured — explain thoroughly
Care Red Line — breaks character in crisisSafetyTiered disclaimers — med/legal have referrals
For: talking with the person, immersive adviceUseFor: diagnosing decisions, methodology, project review

谁都可以被蒸馏

Anyone Can Be Distilled

真人、虚构、历史、纯方法论——女娲先识别画像再决定产物形态

Real, fictional, historical, pure methodology — 女娲 identifies the profile first, then decides artifact form

段
段永平
投资家 / 企业家
Investor / Entrepreneur
双模式Dual-Mode
Right BPP 三件套、Stop Doing List、打孔机决策法——像本人聊投资，也能当工具用。
Right BPP Framework, Stop Doing List — chat investing philosophy, or deploy as diagnostic tools.
罗
罗永浩
连续创业者 / 段子手
Serial Entrepreneur / Pundit
双模式Dual-Mode
模仿标志性表达 DNA——句式、节奏、幽默感，激动时就"又激动了，我老罗就是容易激动"。
Captures signature expression DNA — sentence rhythm, cadence, humor. Gets worked up like the man himself.
绫
神里绫华
Kamisato Ayaka
游戏角色 · 原神
Game Character · Genshin
单 PersonaSingle Persona
虚构角色情景对话，括号动作描写，开场主动抛话题钩子。不编造官方设定。
Fictional character scenario chat with parenthetical gesture notation. Never fabricates official lore.
海
海绵宝宝
SpongeBob
动画角色
Animated Character
双模式Dual-Mode
Persona 还原粉红泡泡式乐观，Toolkit 拆"比奇堡生存哲学"（标注虚构世界免责）。
Persona recreates pink-bubble optimism. Toolkit unpacks "Bikini Bottom Survival Philosophy" (fictional-world disclaimer).
K
Andrej Karpathy
AI 教育家
AI Educator
双模式Dual-Mode
Persona 聊技术品味，Toolkit 用 Software 2.0 + vibe coding 方法教开发。
Persona for tech taste chat. Toolkit uses Software 2.0 + vibe coding to teach development.
芒
查理·芒格
Charlie Munger
投资家 / 思想家
Investor / Thinker
双模式Dual-Mode
Lollapalooza 效应、人类误判心理学、逆向思维——聊智慧，也调用多元思维模型。
Lollapalooza Effect, Psychology of Human Misjudgment, inversion — chat wisdom or deploy mental models.

蒸馏流水线

Distillation Pipeline

从输入到交付，9 个步骤自动流转——红色标记为显式检查点，必须用户确认才继续

From input to delivery — 9 steps auto-flow. Red markers are explicit checkpoints requiring user confirmation.

🔍
Phase 0 · 检查点 ✓
Phase 0 · Checkpoint ✓
入口分流 + 已有 Skill 检测
Entry Routing + Existing Skill Detection
识别蒸馏对象的人物画像、活动地域、主要语言。扫描本地是否已有同名 skill——有则触发三选一：覆盖重做 / 增量更新 / 新建 v2 并存。
Identify the subject's profile, region, and primary language. Scan locally for existing skills — if found, trigger three-way choice: overwrite / incremental update / new v2 alongside.
画像扫描Profile Scan语言判定Language Detection更新检测Update Check
🎯
Phase 0A · 检查点 ✓
Phase 0A · Checkpoint ✓
意图维度自动推断 + 形态选择
Auto Intent Inference + Form Selection
根据人物画像自动推测 12+ 种意图维度（方法论学习 / 风格模仿 / 角色扮演 / 情感陪伴 / 决策复盘……），举一反三补充独有维度。自动判定产物形态（双/单模式），交给用户确认——首次出现术语时先用白话解释。
Infer 12+ intent dimensions from subject profile (methodology / style imitation / roleplay / emotional companionship / decision review…). Extrapolate unique dimensions. Auto-determine artifact form (dual/single mode) — present to user for confirmation. First-time terms explained in plain language.
12+自动判定Auto-Detect术语解释Term Explain
⚡
Phase 1
6 个 Agent 并行调研
6-Agent Parallel Research
著作文献 / 对话访谈 / 表达 DNA / 他者评价 / 决策案例 / 时间线——六个维度同时开工，按人物语言地域自动切换信息源白名单。一手来源优先，知乎/公众号/百度百科永久排除。
Writings / conversations / expression DNA / external views / decisions / timeline — six dimensions run simultaneously. Source whitelist auto-adjusted by subject's language and region. Primary sources first; Zhihu/WeChat/Baidu Baike permanently excluded.
并行调研Parallel Research信息源过滤Source Filtering双语策略Bilingual Strategy
📊
Phase 1.5 · 检查点 ✓
Phase 1.5 · Checkpoint ✓
调研质量摘要
Research Quality Summary
生成一手/二手来源占比、跨 Agent 矛盾点检测、信息不足维度标注。总来源数 < 10 时建议降低期望或补充调研。
Generate primary/secondary source ratio, cross-agent contradiction detection, under-informed dimension flags. Under 10 total sources → recommend lowering expectations or supplementing research.
质量审计Quality Audit矛盾检测Contradictions来源占比Source Ratio
🧠
Phase 2
心智模型框架提炼
Mental Model Framework Extraction
三重验证提取核心思维模型。分两栏：Part A 价值观镜片（2-5 个）——跨域复现/生成力/排他性；Part B 操作招式（1-5 个）——此人亲口命名的可操作清单。同人与官方冲突时暂停询问。
Triple Validation extraction of core mental models. Two columns: Part A — Value Lenses (2-5) — cross-domain recurrence / generative power / distinctiveness; Part B — Operational Techniques (1-5) — personally-named actionable checklists. Fan-work vs official conflicts trigger pause for user decision.
三重验证Triple Validation两栏分拆Two Columns同人协商Fan-Work Negotiation
🏗️
Phase 3
Skill 构建与注入
Skill Construction & Injection
按模板生成三层目录（shared / mode-persona / mode-toolkit）。注入功能矩阵、关怀红线（5 地区热线 + 兜底）、时代锚定声明、分级免责、时效性声明、Agentic Protocol。
Build 3-layer directory (shared / mode-persona / mode-toolkit) from templates. Inject feature matrix, Care Red Line (5-region hotlines + fallback), era anchoring statement, tiered disclaimers, timeliness declaration, Agentic Protocol.
模板填充Template Fill关怀红线Care Red Line时效声明Timeliness
🔬
Phase 4 · 检查点 ✓
Phase 4 · Checkpoint ✓
质量验证 + 修复循环
Quality Validation + Repair Cycle
按推断意图选择验证套件。Toolkit 招式枚举校验（亲口招式全部召回，漏一个 = FAIL）。关怀红线三类信号模拟（普通低落/极端情绪/不健康依赖）全过才算 PASS。最多 2 轮修复，第 2 轮仍 FAIL 必须用户确认是否交付。
Select validation suite by inferred intents. Toolkit technique enumeration check (all personally-named techniques must be recalled — miss one = FAIL). Care Red Line three-signal simulation (ordinary low / extreme distress / unhealthy dependency) — all three must PASS. Max 2 repair cycles; if still FAIL at cycle 2, user must confirm whether to deliver.
招式校验Technique Check关怀测试Care Test2x
✨
Phase 5
双 Agent 精炼
Dual-Agent Refinement
两个独立 Agent 分别审查：意图维度覆盖齐全度 + 强制 section 注入完整度。交叉验证后合并建议，确保产物没有遗漏。
Two independent agents review separately: intent dimension coverage completeness + mandatory section injection completeness. Cross-validation with merged recommendations ensures no omissions.
交叉验证Cross-Validation意图齐全Intent Coveragesection 完整Section Complete
🚀
交付
Delivery
打包注册 + 五段收尾
Package & Register + 5-Part Conclusion
产物上传注册为 custom skill。主消息输出五段话：产物形态 / 使用方式（含切换模式与典型问法）/ 局限提醒 / 后续维护（修改/更新/卸载）/ Skill 简介与女娲引导。不可省略。
Upload and register as custom skill. Main message outputs five parts: artifact form / usage (mode switching and queries) / limitation reminders / maintenance (modify / update / uninstall) / skill summary with 女娲 guidance. None may be omitted.
打包注册Package & Register五段收尾5-Part Conclusion维护说明Maintenance Guide

开始使用

Get Started

方式一 · 告诉你的 Agent（推荐）

Method 1 · Tell Your Agent (Recommended)

帮我安装这个 skill: https://github.com/shallinta/nuwa-dual-mode

方式二 · npx skills CLI

Method 2 · npx skills CLI

npx skills add shallinta/nuwa-dual-mode --skill nuwa-dual-mode

支持 Claude Code · Codex · Cursor · OpenClaw · Hermes · Gemini CLI 等 50+ runtime

Supports Claude Code · Codex · Cursor · OpenClaw · Hermes · Gemini CLI and 50+ runtimes