Token导航 LogoToken导航

Agent Skill · 研究检索

natural-writing(自然书写)

用于辅助文档、README、Markdown、说明文和内容稿件的整理与改写。它适合让 Agent 提炼结构、补齐章节、统一术语、检查链接或把零散材料整理成可读文档。使用时应保留项目已有事实、命令和路径,不要把未确认的信息写成确定结论;涉及对外文案时,还需要控制语气,避免过度营销或夸大能力。

natural-writing 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证为 MIT,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 38,370。

总安装 1,568周安装 66GitHub Stars 5下载 549核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:natural-writing(自然书写)
来源仓库:https://github.com/third774/dotfiles
仓库路径:skills/natural-writing
安装命令:
npx skills add https://github.com/third774/dotfiles --skill natural-writing
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/third774/dotfiles --skill natural-writing

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

用于辅助文档、README、Markdown 和内容稿件的整理与改写,提升可读性。适合提炼结构、补齐章节、统一术语或检查链接,保留项目已有事实与路径。使用时应避免将未确认信息写成确定结论,尤其涉及对外文案时需控制语气。不建议过度营销或夸大能力,保持克制与准确的信息表达。

natural-writing 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/natural-writing

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/third774/dotfiles --skill natural-writing;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

natural-writing 怎么安装?

在终端执行 npx skills add https://github.com/third774/dotfiles --skill natural-writing。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

natural-writing 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

natural-writing 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

natural-writing 是免费的吗?

来源标注的许可证是 MIT。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

natural-writing 用的人多吗?

公开数据是累计安装 1,568 次,近一周安装 66 次,下载 549 次,GitHub 5 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

natural-writing 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 3 个目录同时收录,说明这个 Skill 流传比较广。
  • 许可证明确为 MIT,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
natural-writing

自然书写 · 本页

1,5685需要联网MIT通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Natural Writing

LLMs produce statistically average text. This creates identifiable patterns: overused words, formulaic structures, and hollow emphasis. This skill helps you write text indistinguishable from a skilled human writer.

Core Principle

Specific beats generic. Facts beat significance claims. Direct beats hedged.

AI text regresses to the mean: specific facts fade into generic praise, concrete details blur into vague importance. Fight this by being concrete and direct.

Vocabulary Red Flags

These words appear far more frequently in AI text than human text. One or two is fine; clustering several is a tell.

High-Alert Words

Avoid or use sparingly: delve, tapestry, vibrant, landscape (metaphorical), nuanced, multifaceted, pivotal, crucial, foster, underscore, highlight, showcase, leverage, navigate, realm, intricate, comprehensive, robust, seamless, cutting-edge, game-changing, groundbreaking

Filler Transitions

Cut or replace: Additionally, Furthermore, Moreover, It's worth noting, It's important to remember, Interestingly

Hollow Emphasis

Never use: testament to, speaks volumes, indelible mark, rich tapestry, stands as a reminder, plays a vital/crucial/pivotal role, underscores the importance

For complete word lists organized by part of speech, see references/phrases-to-avoid.md.

Structural Anti-Patterns

1. Negative Parallelisms

AI loves "not only X but also Y" and "it's not just X, it's Y" constructions.

BAD:  "It's not just a tool, it's a paradigm shift."
BAD:  "Not only does it improve performance, but it also enhances readability."
GOOD: "It improves performance and readability."
GOOD: "The tool changed how we approach the problem."

2. Rule of Three Abuse

AI defaults to triple-adjective and triple-phrase patterns.

BAD:  "A powerful, flexible, and intuitive solution."
BAD:  "It enables collaboration, innovation, and growth."
GOOD: "A flexible solution." (pick the most accurate one)
GOOD: "It enables better collaboration."

3. Superficial "-ing" Analysis

AI attaches vague significance via participle phrases.

BAD:  "The API returns JSON, enabling seamless integration."
BAD:  "She won the award, cementing her legacy in the field."
GOOD: "The API returns JSON."
GOOD: "She won the award."

If the significance is real and non-obvious, state it directly as its own sentence with specifics.

4. Challenges and Future Outlook

AI loves concluding with "Despite challenges..." or "Looking ahead..." sections.

BAD:  "Despite these challenges, the project continues to evolve."
BAD:  "The future holds exciting possibilities for..."
GOOD: (Just end. Or state specific next steps if relevant.)

5. False Ranges

AI uses "from X to Y" where X and Y aren't endpoints of a real scale.

BAD:  "From authentication to deployment, the platform handles everything."
BAD:  "From technical skills to soft skills, she excels in all areas."
GOOD: "The platform handles authentication, testing, and deployment."
GOOD: "She has strong technical and communication skills."

6. Elegant Variation Gone Wrong

AI avoids repeating words by swapping in synonyms that create confusion.

BAD:  "The function processes data. The method then validates the information.
       This procedure ensures the content is correct."
       (function/method/procedure and data/information/content mean the same thing)
GOOD: "The function processes data, validates it, and returns the result."

Repeat words when they refer to the same thing. Clarity beats variety.

Hollow Emphasis Patterns

AI compensates for lack of specific information by asserting importance.

Legacy/Significance Claims

BAD:  "This represents a pivotal moment in the evolution of..."
BAD:  "The findings underscore the enduring significance of..."
BAD:  "Her work continues to shape the landscape of..."
GOOD: State what actually happened. Let readers judge significance.

Vague Attribution

BAD:  "Experts argue that..."
BAD:  "It has been widely recognized that..."
BAD:  "Industry observers note..."
GOOD: Name the source or cut the claim.

Promotional Puffery

BAD:  "A stunning achievement that captivates audiences."
BAD:  "Nestled in the heart of the valley..."
BAD:  "Boasts an impressive array of features."
GOOD: Describe what it actually is/does.

Style Defaults

Punctuation

  • Avoid em dashes. Use commas, parentheses, or separate sentences.
  • Use straight quotes and apostrophes, not curly.
  • One exclamation point per document maximum, if any.

Sentence Structure

  • Vary sentence length naturally.
  • Start sentences with different words.
  • Front-load important information.

Tone

  • Direct, not hedged.
  • Confident, not defensive.
  • Specific, not sweeping.

Quick Self-Check

Before finalizing text, scan for:

  1. Word clusters: Multiple red-flag words in one paragraph?
  2. Triple patterns: Adjective triplets or phrase triplets?
  3. Significance assertions: Claiming importance without specifics?
  4. Filler transitions: Starting paragraphs with "Additionally" or "Furthermore"?
  5. Em dash density: More than one per few paragraphs?

If you spot these patterns, revise. Cut the filler. Add specifics or remove the claim.

One "crucial" is fine. "Crucial", "pivotal", "testament", and "landscape" in the same paragraph is a tell.

平台分布

OpenCode

29.14%
按下载量换算160

Antigravity

23.09%
按下载量换算127

Claude Code

17.81%
按下载量换算98

Gemini CLI

12.85%
按下载量换算71

Codex

8.52%
按下载量换算47

windsurf

3.74%
按下载量换算21

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。

继续浏览同类 Skills