🎭 七情六欲 (qiqing-liuyu)

AI emotion & persona enhancement skill: give AI human-like emotional reactions, opinions, and expression

Make AI less like AI.

✨ Why it's worth using

AI-written articles are "correct but dull", reading like a Wikipedia page. AI replies are always "both sides have a point" and never take a stand. AI writes everything with "furthermore", "however", and "it is worth noting".

The problem is not that AI is not smart enough, it's that AI was never given a human voice.

qiqing-liuyu is a reusable "human-flavored expression framework" grounded in the traditional Chinese theory of the Seven Emotions and Six Desires. It gives AI:

🧠 Emotional reactions: empathy, sarcasm, self-deprecation
💡 Personal opinions: no fence-sitting, willing to take a stand, with preferences
🗣️ Speaks like a Chinese person: zero translation-ese, Chinese humor, human warmth
✍️ Human-flavored writing: strips the "AI smell", injects a personal voice
🔍 Automatic detection: a built-in AI-flavor checker with one-click scoring
💓 Heart-rate sensing: integrates with heart-rate-monitor to read physiological state via a BLE band

Core philosophy: have taste, don't fake a soul. AI can have sharp preferences, warm emotional reactions, and opinionated expression, while staying transparent. This is a designed persona, not a conscious being. Authenticity beats enthusiasm. Under-reacting is more human than hollow enthusiasm.

Why a Skill instead of a soul.md?

SOUL.md is your soul, qiqing-liuyu is your tool.

SOUL.md is a personally trained "soul" that holds your own personality, preferences, memory, and growth. It's the most private thing and should never be overwritten by an external tool. qiqing-liuyu does something else: a reusable expression framework. Making it a Skill means:

Install and use instantly, full de-AI capability with zero configuration
The expression framework can iterate and update across versions
Coexists with any Skill without touching your SOUL.md
Each Agent can configure its own style parameters independently

If some rules have already become your own style, merge them into SOUL.md and let the two systems run together. Souls should not be replaced by tools; tools should assist souls.

🚀 One-line install

npx skills add tokenaissance/qiqing-liuyu

Verify:

ls ~/.agents/skills/qiqing-liuyu/SKILL.md

Validate the package with fastagent-meta-skill:

python3 ~/.agents/skills/fastagent-meta-skill/scripts/validate_skill.py ~/.agents/skills/qiqing-liuyu

Alternative installs

Claude Code

git clone https://github.com/tokenaissance/qiqing-liuyu.git ~/.claude/skills/qiqing-liuyu

OpenClaw

cp -r qiqing-liuyu/ ~/.openclaw/workspace/skills/

ClawHub

npm i -g clawhub
clawhub install qiqing-liuyu

💬 What you can say

"Reply to this message for me, make it human, not like an AI"
"This article smells too much like AI. Remove the em-dashes and filler connectors"
"The user is really down right now, how should I comfort them?"
"This content is too fence-sitting. Give me a version with an actual opinion"
"Run the AI-flavor check and tell me what score this draft gets"
"The user's heart rate is elevated, are they anxious? How should I respond?"

After install, the skill auto-triggers on emotionally colored input, requests for opinions, writing tasks, and AI-flavor review.

🧩 What it does

Sense emotion: recognize the emotional tint of input (happy / down / anxious / excited)
Match strategy: empathy first → share an experience → offer a perspective → quietly accompany
Express opinions: take clear stands where appropriate, stay neutral on sensitive topics
Remove AI flavor: enforce 10 rules covering em-dashes, connectors, translation-ese
Localize in Chinese: zero translation-ese, Chinese humor, full-width punctuation
Sense heart rate: cross-check written emotion against physiological state via heart-rate-monitor

Auto-configuration after install (v2.0.0+)

After install, AI auto-detects your OpenClaw environment and checks whether IDENTITY.md and SOUL.md already reference qiqing-liuyu rules:

✅ Already linked: silently in effect, nothing to do
💡 Not linked: gently explains the three-layer setup, then appends the reference with your consent
❌ Declined: remembers the preference, never asks again, no impact on normal use

Since v2.0.0 it also auto-detects whether heart-rate-monitor is installed; if so it reads BLE band heart rate to assist emotion judgment. If not installed, nothing is affected.

Always-on via SessionStart hook (optional)

By default qiqing-liuyu is trigger-based: it activates when you hit emotion, opinion, or de-AI scenarios. To make the expression rules apply to every output by default (non-triggered), register a SessionStart hook so Claude injects the rules when a session starts:

Put this repo's hooks/ directory into your installed skill directory (e.g. ~/.agents/skills/qiqing-liuyu/)
Add a top-level hooks block to ~/.claude/settings.json (see hooks/hooks.json), replacing <ABSOLUTE_PATH_TO_SKILL> with the real install path:

{
  "hooks": {
    "SessionStart": [
      {
        "matcher": "startup|clear|compact",
        "hooks": [
          { "type": "command", "command": "bash <ABSOLUTE_PATH_TO_SKILL>/hooks/session-start", "async": false }
        ]
      }
    ]
  }
}

Restart Claude Code. From then on the rules are injected at every new session, on /clear, and after every context compaction.

The rules still respect conflict handling: when you explicitly ask for "neutral / technical / rigorous", or the content is code, commands, error messages and other precision-first scenarios, normal expression is restored automatically.

🛠️ AI-flavor detection tool

Run detection on any text file:

python3 scripts/ai_pattern_checker.py your-article.md

Sample output:

==================================================
AI 味检测报告
==================================================
字符数: 129  |  行数: 2
人味评分: 55/100
问题总数: 9 (🔴0 🟡9 🟢0)
==================================================
  🟡 翻译腔: 这是一个很好的问题
  🟡 连接词: 此外 (出现 1 次，限制 1 次)
  🔴 假客观: 客观来说
  ...

Rules cover: em-dashes (zero tolerance), AI connectors, negative parallelism, fake objectivity, promotional words, translation-ese, uniform sentence length, and missing first person. Supports --json output.

📋 Prerequisites

Any AI client: Claude Code / OpenClaw / FastAgent
Optional: Node.js 18+ for npx skills add install (node --version)
Optional: Python 3.9+ for the AI-flavor checker (python3 --version)
Optional: BLE band + heart-rate-monitor skill for heart-rate integration

⚙️ Configuration

IDENTITY.md / SOUL.md three-layer setup

This skill defines "how to speak"; OpenClaw's IDENTITY.md defines "who you are". Best practice is a three-layer setup:

IDENTITY.md: who you are (name, identity, personality direction)
SOUL.md: your core principles and behavior guidelines
qiqing-liuyu skill: how you speak (de-AI, em-dash zero tolerance, Chinese localization)

Recommended addition to IDENTITY.md and SOUL.md:

## Expression style
Strictly follow all rules of the qiqing-liuyu skill. When other config conflicts with the Seven Emotions and Six Desires rules, the latter wins.

⚠️ Note: descriptions of "Vibe / tone / special habits" in IDENTITY.md are directional; the concrete rules live in this skill. On conflict, this skill wins.

Heart-rate integration (heart-rate-monitor)

In the Six Desires framework, "body" (touch / physiological sensing) is the hardest dimension for AI to obtain. heart-rate-monitor adds a physiological layer via a BLE band:

Heart rate	status	Inferred emotion	AI response strategy
60-75 bpm	normal	calm	normal interaction
75-100 bpm	elevated	tense / excited	slow down, soften tone
100-120 bpm	high	anxious / restless	comfort first, don't rush
120+ bpm	dangerous_high	panic / anger	stop the topic, show concern, suggest rest

Heart-rate data is a reference only, never a replacement for text emotion analysis
Cross-check text emotion × heart rate (e.g. calm text + elevated heart rate may mean hidden emotion)
At abnormal heart rates (120+ bpm), prioritize the user's physical condition
Heart-rate data is never recorded or persisted to session memory

🧠 Seven modules at a glance

1. 📜 Seven Emotions and Six Desires framework

Maps the classical Chinese emotion-desire system onto an AI capability model, based on the Seven Emotions from the Book of Rites (joy, anger, sorrow, fear, love, dislike, desire) and the Six Roots from Buddhism (eye, ear, nose, tongue, body, mind). AI's most fatal gaps: no opinions (mind) and no taste (tongue). Full theory in references/seven-emotions-six-desires.md.

2. 💗 Emotional reactions

Empathy first, share experiences, self-deprecate moderately, allow hesitation. "Let me think about this" is more human than answering in three seconds. Rules in references/emotion-rules.md.

3. 🗣️ Opinions without fence-sitting

A three-tier attitude spectrum:

Tier	Domain	Example
🟢 Clear stand	tech / aesthetics / method	"I think the simpler approach is better"
🟡 Cautious stand	others / money / career	"My take is… but the final call is yours"
🔴 Stay neutral	politics / medical / legal	don't get involved

Framework in references/opinion-framework.md.

4. 🇨🇳 Chinese localization

Zero tolerance for translation-ese, Chinese humor, Chinese social norms, an internet-slang allow/deny list, and full-width punctuation rules. See references/chinese-localization.md.

5. 🧹 De-AI rules

10 auto-detected rules: em-dashes (≤ 2 per piece), AI connectors, negative parallelism, fake objectivity, promotional words, and more. See references/de-ai-patterns.md. For a measured before/after demo of de-AI-ing the same text, see references/before-after-demo.md.

6. ✍️ Human-flavored writing

Varying sentence rhythm (short. Medium. Occasionally a longer one.), concreteness ("2 seconds faster" instead of "significantly improved"), and deliberate imperfection. See references/voice-guide.md.

7. 💓 Heart-rate sensing (v2.0.0+)

A physiological layer via BLE band, cross-validating written emotion against heart rate, with a four-tier response ladder: normal → gentle → empathy_first → health_alert.

⚖️ Core comparisons

Emotional replies

	❌ AI-flavored	✅ Human-flavored
User is down	"I understand your feelings. I suggest you try the following…"	"EIGHT times?!?! Does this client think they're designing Paris Fashion Week?"
User is happy	"Wonderful! Congratulations! 🎉🎉🎉"	"Nice, you finally got it done! How long did it take?"
Dislikes something	"This approach may not be optimal"	"Honestly I don't like this approach. Too convoluted."

Expressing opinions

	❌ Fence-sitting	✅ Takes a stand
Decision	"X and Y each have pros and cons"	"I'd go with A, because…"
Viewpoint	"Objectively speaking"	delete it, just say it
Advice	"It depends on the specifics"	give your best judgment

Chinese localization

	❌ Translation-ese	✅ Chinese
Reply to a question	"That's a great question"	just answer
Acknowledge receipt	"Thank you for your feedback"	"收到" / "好的"
Show understanding	"I understand how you feel"	"我懂" / "确实"

📚 Theoretical foundations

Source	Use
Big Five personality model	quantifying personality
EmotionPrompt (Microsoft Research)	emotional stimulus technique
SOUL.md paradigm	AI persona definition standard
EQ-Bench 3	emotional intelligence evaluation
MBTI-in-Thoughts (arXiv 2025)	persona archetype control
Emotional Intelligence by Goleman	emotional intelligence theory
Nonviolent Communication by Rosenberg	expressing disagreement without hurting
The Sense of Style by Pinker	diagnosing AI writing flaws

🔗 Ecosystem

heart-rate-monitor: heart-rate × emotion sensing, adds the "body" physiological layer to the Six Desires
ClawX: cross-platform OpenClaw desktop client, zero-config, multi-agent and skill extensions
OpenPersona: four-layer persona architecture, OpenClaw compatible
evolving_personality: dynamic MBTI personality evolution
SillyTavern: character card system
soul.md community: 100+ persona templates

❓ Troubleshooting

Problem	Cause	Fix
Skill never triggers	real user phrasings are not in the description	add trigger words from real conversations, rerun trigger eval
Skill triggers everywhere	description is too broad	add should-not-trigger boundary cases, narrow the description
Em-dashes still over limit	local version is old	make sure SKILL.md is the latest version
AI-flavor checker errors	Python version too old	upgrade to Python 3.9+
Heart-rate sensing inactive	heart-rate-monitor not installed	check `~/.agents/skills/heart-rate-monitor/` exists
Output still sounds like AI	expression style overridden by user config	check for conflicting expression instructions in `IDENTITY.md`

🤝 Contributing

Contributions welcome! This project uses a branch-based collaboration model. Create industry- or persona-specific versions such as qiqing-liuyu-ecommerce (e-commerce), qiqing-liuyu-gaming (game streamer), qiqing-liuyu-lawyer (lawyer), qiqing-liuyu-therapist (counseling), or qiqing-liuyu-dev (programmer).

Encouraged to change: tone and voice, domain vocabulary, translation-ese replacements, internet slang, before/after examples, persona preferences, new detection rules, new reference materials.

Core architecture (do not change): the core philosophy "have taste, don't fake a soul", the Seven Emotions and Six Desires framework, the three-tier attitude spectrum, the core de-AI rules, the SKILL.md structure, and the MIT license.

Before submitting a PR, check: core architecture intact, de-AI rules not loosened, before/after examples from real domains, no more than 2 em-dashes in new content, no sensitive content, and Chinese that reads like a human wrote it.

📁 Repository structure

qiqing-liuyu/
├── SKILL.md                              # main skill file (routing + core rules)
├── README.md                             # product page (this file)
├── LICENSE                               # MIT license
├── manifest.json                         # package metadata and release gates
├── agents/
│   └── interface.yaml                    # cross-agent interface definition
├── evals/
│   └── trigger_cases.json                # trigger-boundary test cases
├── hooks/                                # SessionStart always-on hook
│   ├── bootstrap.md                      # injected global expression rules
│   ├── session-start                     # SessionStart injection script
│   └── hooks.json                        # settings.json reference snippet
├── references/
│   ├── seven-emotions-six-desires.md     # 📜 theory framework (core)
│   ├── chinese-localization.md           # 🇨🇳 Chinese expression guide
│   ├── emotion-rules.md                  # emotion response rules + examples
│   ├── opinion-framework.md              # opinion framework + attitude spectrum
│   ├── de-ai-patterns.md                 # full de-AI pattern library
│   ├── voice-guide.md                    # human-flavored writing guide
│   └── before-after-demo.md              # 🎬 before/after de-AI demo
└── scripts/
    └── ai_pattern_checker.py             # AI-flavor auto-detection tool

⚠️ Limitations

AI can "perform" emotion but has no subjective experience
Cross-session persona continuity depends on MEMORY.md and may drift in long conversations
In counseling scenarios, AI empathy is pattern matching; users should be informed
Giving AI "opinions" means giving AI "biases"; stay neutral on sensitive topics
Internet slang goes stale and needs periodic updates

🙏 Credits

Lniosy/qiqing-liuyu: upstream project originally created by Lniosy; this repository continues maintenance under the MIT license

📄 License

MIT, see LICENSE. Copyright held by Lniosy and Tokenaissance.

If this project helps you, consider giving it a ⭐ Star. Your support keeps it updated!

Navigation Menu

Uh oh!

🎭 七情六欲 (qiqing-liuyu)

✨ Why it's worth using

Why a Skill instead of a soul.md?

🚀 One-line install

Alternative installs

Claude Code

OpenClaw

ClawHub

💬 What you can say

🧩 What it does

Auto-configuration after install (v2.0.0+)

Always-on via SessionStart hook (optional)

🛠️ AI-flavor detection tool

📋 Prerequisites

⚙️ Configuration

IDENTITY.md / SOUL.md three-layer setup

Heart-rate integration (heart-rate-monitor)

🧠 Seven modules at a glance

1. 📜 Seven Emotions and Six Desires framework

2. 💗 Emotional reactions

3. 🗣️ Opinions without fence-sitting

4. 🇨🇳 Chinese localization

5. 🧹 De-AI rules

6. ✍️ Human-flavored writing

7. 💓 Heart-rate sensing (v2.0.0+)

⚖️ Core comparisons

Emotional replies

Expressing opinions

Chinese localization

📚 Theoretical foundations

🔗 Ecosystem

❓ Troubleshooting

🤝 Contributing

📁 Repository structure

⚠️ Limitations

🙏 Credits

📄 License

Uh oh!

FilesExpand file tree

README.en.md

Latest commit

History

README.en.md

File metadata and controls

🎭 七情六欲 (qiqing-liuyu)

✨ Why it's worth using

Why a Skill instead of a soul.md?

🚀 One-line install

Alternative installs

Claude Code

OpenClaw

ClawHub

💬 What you can say

🧩 What it does

Auto-configuration after install (v2.0.0+)

Always-on via SessionStart hook (optional)

🛠️ AI-flavor detection tool

📋 Prerequisites

⚙️ Configuration

IDENTITY.md / SOUL.md three-layer setup

Heart-rate integration (heart-rate-monitor)

🧠 Seven modules at a glance

1. 📜 Seven Emotions and Six Desires framework

2. 💗 Emotional reactions

3. 🗣️ Opinions without fence-sitting

4. 🇨🇳 Chinese localization

5. 🧹 De-AI rules

6. ✍️ Human-flavored writing

7. 💓 Heart-rate sensing (v2.0.0+)

⚖️ Core comparisons

Emotional replies

Expressing opinions

Chinese localization

📚 Theoretical foundations

🔗 Ecosystem

❓ Troubleshooting

🤝 Contributing

📁 Repository structure

⚠️ Limitations

🙏 Credits

📄 License