talk-normal-llm-prompt

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

talk-normal

Skill by ara.so — Daily 2026 Skills collection.

talk-normal is a system prompt (plus a shell-script helper) that strips AI slop — bullet-point padding, hollow affirmations, corporate filler — from any LLM while preserving all useful information. Tested at ~73% character reduction on GPT-4o-mini and GPT-5.4 with no information loss.

由ara.so开发的技能工具——属于Daily 2026 Skills合集。

talk-normal是一个系统提示词（附带Shell脚本辅助工具），可去除任意LLM输出中的AI冗余内容——包括无意义的项目符号、空洞的套话、企业式冗余表述——同时保留所有有用信息。在GPT-4o-mini和GPT-5.4上测试显示，可减少约73%的字符数且无信息丢失。

How it works

工作原理

The project is a single

prompt.md

file (the system prompt) plus optional shell helpers. You copy the prompt text into the "System" field of any LLM interface or API call.

repo layout
├── prompt.md          ← the system prompt (main artifact)
├── CHANGELOG.md       ← rule history
├── CONTRIBUTING.md    ← how to add rules
└── TEST_RESULTS.md    ← before/after comparisons

该项目包含一个

prompt.md

文件（核心系统提示词）以及可选的Shell辅助工具。你只需将提示词文本复制到任意LLM界面或API调用的「System」字段中即可使用。

仓库结构
├── prompt.md          ← 核心系统提示词（主要产物）
├── CHANGELOG.md       ← 规则更新历史
├── CONTRIBUTING.md    ← 规则添加指南
└── TEST_RESULTS.md    ← 优化前后对比结果

Installation

安装步骤

1 — Clone the repo

1 — 克隆仓库

bash

git clone https://github.com/hexiecs/talk-normal.git
cd talk-normal

bash

git clone https://github.com/hexiecs/talk-normal.git
cd talk-normal

2 — Read the prompt

2 — 查看提示词

bash

cat prompt.md

bash

cat prompt.md

3 — Copy into your tool

3 — 复制到你的工具中

Paste the contents of

prompt.md

into:

ChatGPT → Settings → Customize ChatGPT → Custom Instructions → "How should ChatGPT respond?"
Claude.ai → Project Instructions
Cursor / Windsurf →
```
.cursorrules
```
or global AI rules
API calls →
```
system
```
parameter (see examples below)

将

prompt.md

的内容粘贴到以下位置：

ChatGPT → 设置 → 自定义ChatGPT → 自定义指令 →「ChatGPT应如何回应？」
Claude.ai → 项目指令
Cursor / Windsurf →
```
.cursorrules
```
或全局AI规则
API调用 →
```
system
```
参数（见下方示例）

Using the prompt via API

通过API使用提示词

OpenAI (Python)

OpenAI（Python）

python

import os
from pathlib import Path
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

system_prompt = Path("prompt.md").read_text()

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": system_prompt},
        {"role": "user",   "content": "What is Python?"},
    ],
)
print(response.choices[0].message.content)

python

import os
from pathlib import Path
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

system_prompt = Path("prompt.md").read_text()

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": system_prompt},
        {"role": "user",   "content": "What is Python?"},
    ],
)
print(response.choices[0].message.content)

OpenAI (curl)

OpenAI（curl）

bash

SYSTEM=$(cat prompt.md | jq -Rs .)

curl https://api.openai.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{
    \"model\": \"gpt-4o-mini\",
    \"messages\": [
      {\"role\": \"system\", \"content\": $SYSTEM},
      {\"role\": \"user\",   \"content\": \"What is Python?\"}
    ]
  }"

bash

SYSTEM=$(cat prompt.md | jq -Rs .)

curl https://api.openai.com/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{
    \"model\": \"gpt-4o-mini\",
    \"messages\": [
      {\"role\": \"system\", \"content\": $SYSTEM},
      {\"role\": \"user\",   \"content\": \"What is Python?\"}
    ]
  }"

Anthropic Claude (Python)

Anthropic Claude（Python）

python

import os
from pathlib import Path
import anthropic

client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])

system_prompt = Path("prompt.md").read_text()

message = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=1024,
    system=system_prompt,
    messages=[{"role": "user", "content": "Explain Docker in one paragraph."}],
)
print(message.content[0].text)

python

import os
from pathlib import Path
import anthropic

client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])

system_prompt = Path("prompt.md").read_text()

message = client.messages.create(
    model="claude-opus-4-5",
    max_tokens=1024,
    system=system_prompt,
    messages=[{"role": "user", "content": "Explain Docker in one paragraph."}],
)
print(message.content[0].text)

Google Gemini (Python)

Google Gemini（Python）

python

import os
from pathlib import Path
import google.generativeai as genai

genai.configure(api_key=os.environ["GEMINI_API_KEY"])

system_prompt = Path("prompt.md").read_text()

model = genai.GenerativeModel(
    model_name="gemini-1.5-flash",
    system_instruction=system_prompt,
)

response = model.generate_content("What is a neural network?")
print(response.text)

python

import os
from pathlib import Path
import google.generativeai as genai

genai.configure(api_key=os.environ["GEMINI_API_KEY"])

system_prompt = Path("prompt.md").read_text()

model = genai.GenerativeModel(
    model_name="gemini-1.5-flash",
    system_instruction=system_prompt,
)

response = model.generate_content("What is a neural network?")
print(response.text)

Ollama (local models)

Ollama（本地模型）

bash

SYSTEM=$(cat prompt.md)

ollama run llama3 \
  --system "$SYSTEM" \
  "What is a REST API?"

Or via the Ollama Python SDK:

python

import subprocess, json
from pathlib import Path

system_prompt = Path("prompt.md").read_text()

result = subprocess.run(
    ["ollama", "run", "llama3"],
    input=f"SYSTEM: {system_prompt}\nUSER: What is a REST API?",
    capture_output=True, text=True,
)
print(result.stdout)

bash

SYSTEM=$(cat prompt.md)

ollama run llama3 \
  --system "$SYSTEM" \
  "What is a REST API?"

或通过Ollama Python SDK：

python

import subprocess, json
from pathlib import Path

system_prompt = Path("prompt.md").read_text()

result = subprocess.run(
    ["ollama", "run", "llama3"],
    input=f"SYSTEM: {system_prompt}\nUSER: What is a REST API?",
    capture_output=True, text=True,
)
print(result.stdout)

Shell helper: one-liner wrapper

Shell辅助工具：单行包装器

A reusable shell function that injects the prompt automatically:

bash

undefined

一个可重复使用的Shell函数，可自动注入提示词：

bash

undefined

Add to ~/.bashrc or ~/.zshrc

添加到~/.bashrc或~/.zshrc

export TALK_NORMAL_PROMPT="$HOME/talk-normal/prompt.md"

asknormal() { local question="$*" local system system=$(cat "$TALK_NORMAL_PROMPT")

curl -s https://api.openai.com/v1/chat/completions
-H "Authorization: Bearer $OPENAI_API_KEY"
-H "Content-Type: application/json"
-d "$(jq -n
--arg sys "$system"
--arg q "$question"
'{model:"gpt-4o-mini",messages:[{role:"system",content:$sys},{role:"user",content:$q}]}' )" | jq -r '.choices[0].message.content' }


Usage:

```bash
source ~/.bashrc
asknormal "What is the CAP theorem?"

export TALK_NORMAL_PROMPT="$HOME/talk-normal/prompt.md"

asknormal() { local question="$*" local system system=$(cat "$TALK_NORMAL_PROMPT")


使用方法：

```bash
source ~/.bashrc
asknormal "What is the CAP theorem?"

Embedding in a project's AI config

嵌入到项目的AI配置中

Cursor (

.cursorrules

)

Cursor（

.cursorrules

）

bash

undefined

bash

undefined

Prepend talk-normal to your existing rules

将talk-normal规则添加到现有规则前

cat talk-normal/prompt.md > .cursorrules echo "" >> .cursorrules echo "# Project-specific rules below" >> .cursorrules cat your-existing-rules.md >> .cursorrules

undefined

cat talk-normal/prompt.md > .cursorrules echo "" >> .cursorrules echo "# 以下是项目专属规则" >> .cursorrules cat your-existing-rules.md >> .cursorrules

undefined

OpenAI Assistants API

python

import os
from pathlib import Path
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
system_prompt = Path("talk-normal/prompt.md").read_text()

assistant = client.beta.assistants.create(
    name="Normal Assistant",
    instructions=system_prompt,
    model="gpt-4o-mini",
)
print(f"Assistant ID: {assistant.id}")

python

import os
from pathlib import Path
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
system_prompt = Path("talk-normal/prompt.md").read_text()

assistant = client.beta.assistants.create(
    name="Normal Assistant",
    instructions=system_prompt,
    model="gpt-4o-mini",
)
print(f"Assistant ID: {assistant.id}")

Combining with your own system prompt

与自定义系统提示词结合使用

talk-normal rules are additive — prepend them before your domain instructions:

python

from pathlib import Path

talk_normal = Path("talk-normal/prompt.md").read_text()

your_rules = """
You are a senior backend engineer. Answer questions about Python, Go, and distributed systems.
"""

combined_system = f"{talk_normal}\n\n---\n\n{your_rules}"

talk-normal的规则是可叠加的——将其放在你的领域专属指令之前：

python

from pathlib import Path

talk_normal = Path("talk-normal/prompt.md").read_text()

your_rules = """
你是一名资深后端工程师，请回答关于Python、Go和分布式系统的问题。
"""

combined_system = f"{talk_normal}\n\n---\n\n{your_rules}"

Common patterns

常见使用模式

Pattern 1: Measure verbosity reduction

模式1：测量冗余度降低比例

python

def verbosity_ratio(before: str, after: str) -> float:
    """Returns fraction of original length kept (lower = more concise)."""
    return len(after) / len(before)

before = "Python is a high-level, interpreted programming language known for its readability..."  # 1583 chars
after  = "Python is a high-level, interpreted language known for readability..."                  #  513 chars
print(f"{verbosity_ratio(before, after):.0%} of original length")  # → 32%

python

def verbosity_ratio(before: str, after: str) -> float:
    """返回保留的原文本长度比例（值越小表示越简洁）。"""
    return len(after) / len(before)

before = "Python is a high-level, interpreted programming language known for its readability..."  # 1583个字符
after  = "Python is a high-level, interpreted language known for readability..."                  #  513个字符
print(f"{verbosity_ratio(before, after):.0%} of original length")  # → 32%

Pattern 2: A/B test with and without the prompt

模式2：有无提示词的A/B测试

python

import os
from pathlib import Path
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
system_prompt = Path("talk-normal/prompt.md").read_text()

question = "What is Kubernetes?"

def ask(system: str | None, user: str) -> str:
    messages = []
    if system:
        messages.append({"role": "system", "content": system})
    messages.append({"role": "user", "content": user})
    resp = client.chat.completions.create(model="gpt-4o-mini", messages=messages)
    return resp.choices[0].message.content

without = ask(None, question)
with_prompt = ask(system_prompt, question)

print(f"Without: {len(without)} chars")
print(f"With:    {len(with_prompt)} chars")
print(f"Reduction: {(1 - len(with_prompt)/len(without)):.0%}")

python

import os
from pathlib import Path
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
system_prompt = Path("talk-normal/prompt.md").read_text()

question = "What is Kubernetes?"

def ask(system: str | None, user: str) -> str:
    messages = []
    if system:
        messages.append({"role": "system", "content": system})
    messages.append({"role": "user", "content": user})
    resp = client.chat.completions.create(model="gpt-4o-mini", messages=messages)
    return resp.choices[0].message.content

without = ask(None, question)
with_prompt = ask(system_prompt, question)

print(f"无提示词：{len(without)}个字符")
print(f"有提示词：{len(with_prompt)}个字符")
print(f"精简比例：{(1 - len(with_prompt)/len(without)):.0%}")

Pattern 3: Keep the prompt up to date

模式3：保持提示词更新

bash

undefined

bash

undefined

Pull latest rules from upstream

从上游拉取最新规则

cd talk-normal git pull origin main

Check what changed

查看更新内容

git log --oneline -10 cat CHANGELOG.md | head -50

---

git log --oneline -10 cat CHANGELOG.md | head -50

---

Contributing a new rule

贡献新规则

Fork the repo and create a branch:
```
git checkout -b rule/no-em-dashes
```
Edit
```
prompt.md
```
— add your rule in plain imperative English
Add an entry to
```
CHANGELOG.md
```
Open an Issue or PR describing: what slop the rule targets, a before/after example

bash

undefined

Fork仓库并创建分支：
```
git checkout -b rule/no-em-dashes
```
编辑
```
prompt.md
```
——用简洁的祈使句添加你的规则
在
```
CHANGELOG.md
```
中添加更新记录
提交Issue或PR，说明：该规则针对哪种冗余内容，附上优化前后的示例

bash

undefined

Quick before/after test for your new rule

快速测试你的新规则

SYSTEM=$(cat prompt.md) echo "Test question" | asknormal # uses your modified prompt

---

SYSTEM=$(cat prompt.md) echo "测试问题" | asknormal # 使用你修改后的提示词

---

Troubleshooting

故障排查

Symptom	Fix
Model still uses bullet points	Ensure the system prompt is in the `system` role, not prepended to `user`
Prompt too long for context window	Use a smaller model or trim older messages; `prompt.md` is intentionally compact
Ollama ignores system prompt	Some quantized models have weak instruction-following; try `mistral` or `llama3`
Rules conflict with your own system prompt	Put talk-normal rules first; add `# Override:` comment before conflicting rules
Response is too terse / lost information	The prompt reduces filler, not facts — file an issue with a reproduction case

症状	解决方法
模型仍使用项目符号	确保系统提示词放在 `system` 角色中，而非附加到 `user` 内容前
提示词过长超出上下文窗口	使用更小的模型或修剪历史消息； `prompt.md` 已被设计得尽可能紧凑
Ollama忽略系统提示词	部分量化模型的指令遵循能力较弱；尝试使用 `mistral` 或 `llama3` 模型
规则与自定义系统提示词冲突	将talk-normal规则放在前面；在冲突规则前添加 `# Override:` 注释
回应过于简洁/丢失信息	提示词仅去除冗余内容，不会丢失事实；提交Issue并附上复现案例

Key facts for agents

面向Agent的关键信息

Primary artifact:
```
prompt.md
```
— copy its text verbatim as the
```
system
```
message
No code to run: this is a prompt, not a library; no
```
pip install
```
, no build step
Model-agnostic: works with GPT, Claude, Gemini, LLaMA, Mistral, etc.
Tested reduction: ~72–73% character reduction, zero information loss on 10-question benchmark
License: MIT — use freely in commercial products

核心产物：
```
prompt.md
```
——直接复制其文本作为
```
system
```
消息
无需运行代码：这是一个提示词，而非库；无需
```
pip install
```
，无需构建步骤
模型无关：适用于GPT、Claude、Gemini、LLaMA、Mistral等各类模型
测试精简比例：在10题基准测试中，可减少约72–73%的字符数且无信息丢失
许可证：MIT——可免费用于商业产品

talk-normal-llm-prompt

Original

Translation

talk-normal

talk-normal

How it works

工作原理

Installation

安装步骤

1 — Clone the repo

1 — 克隆仓库

2 — Read the prompt

2 — 查看提示词

3 — Copy into your tool

3 — 复制到你的工具中

Using the prompt via API

通过API使用提示词

OpenAI (Python)

OpenAI（Python）

OpenAI (curl)

OpenAI（curl）

Anthropic Claude (Python)

Anthropic Claude（Python）

Google Gemini (Python)

Google Gemini（Python）

Ollama (local models)

Ollama（本地模型）

Shell helper: one-liner wrapper

Shell辅助工具：单行包装器

Add to ~/.bashrc or ~/.zshrc

添加到~/.bashrc或~/.zshrc

Embedding in a project's AI config

嵌入到项目的AI配置中

Cursor (.cursorrules)

Cursor（.cursorrules）

Prepend talk-normal to your existing rules

将talk-normal规则添加到现有规则前

OpenAI Assistants API

OpenAI Assistants API

Combining with your own system prompt

与自定义系统提示词结合使用

Common patterns

常见使用模式

Pattern 1: Measure verbosity reduction

模式1：测量冗余度降低比例

Pattern 2: A/B test with and without the prompt

模式2：有无提示词的A/B测试

Pattern 3: Keep the prompt up to date

模式3：保持提示词更新

Pull latest rules from upstream

从上游拉取最新规则

Check what changed

查看更新内容

Contributing a new rule

贡献新规则

Quick before/after test for your new rule

快速测试你的新规则

Troubleshooting

故障排查

Key facts for agents

面向Agent的关键信息

Cursor (
`.cursorrules`
)

Cursor（
`.cursorrules`
）