AI UNDERDOGSDAILY PICK
AI UNDERDOGS
AI 教练团，禁止评判你
An AI Coach That Can't Judge You
ibm777p2/Intelligence-Emotions
别人家 AI 帮你优化决策
这个 AI 团队只有一个规矩——不准评判你
Every other AI scores and optimizes you. This
team has one rule: judgment is forbidden
不是聊天机器人，是角色团
Not a Chatbot — a Role System
作者把 Positive Intelligence 的 PQ 体系搬进了 Claude Code
你面对的不是单一模型
是一个教练团队——Saboteur 模式帮你识别内在干扰
Sage 模式引导你换个角度看问题
两个角色轮流上场
The author imported the PQ framework into Claude
Code. You don't talk to one model —
you get a coaching team. Saboteur mode identifies
your inner resistance, Sage mode reframes your thinking
Two roles, alternating
★ SIGNAL 1
「禁止评判」是产品直觉
"No Judgment" Is Product Instinct
市面上 AI 教练工具都在评分
建议、优化你
这个作者偏要定一条规矩——角色只能引导
不能评判
这个约束本身就是品味信号：当所有 AI 都在当考官
他让 AI 当倾听者
Every AI coach out there scores, suggests, and
optimizes you. This author set one hard constraint
— roles can only guide, never judge. That
single rule is the taste signal. While every
AI plays examiner, his plays listener
★ SIGNAL 2
不是贴 Prompt，是搭技能栈
Not a Prompt Paste — a Skill Stack
这可不是复制粘贴到 ChatGPT 就完事的 prompt
作者把它做成了 Claude Code 技能栈：模板管线控制流程
coach preamble 约束角色行为
还加了上下文恢复模块
让跨会话对话不断线
14 次 commit
连可选择回答这种 UI 细节都做了
This isn't a prompt you paste into ChatGPT
The author built it as a proper Claude
Code skill stack — template pipeline, coach preamble
controlling role behavior, and a context-restore module for
cross-session continuity. 14 commits deep, with UI touches
like selectable answer options
作者没说这东西成熟了
14 次 commit 说明还在迭代中
但他选的方向挺有意思——情绪智能
当所有人都在做代码助手
写作助手的时候
他在 Claude Code 里认认真真搭了一套心理教练系统
The author never claimed it's finished — 14
commits says it's still iterating. But the direction
is interesting: emotional intelligence. While everyone builds code
assistants and writing tools, he's building a real
coaching system inside Claude Code
AI UNDERDOGS
Claude Code 里的 Positive Intelligence 实验
A Positive Intelligence experiment inside Claude Code
ibm777p2/Intelligence-Emotions
关注 · 每天发现更多 AI 神作
github.com/ibm777p2/Intelligence-Emotions