跳到正文
cccskills
免费在 GitHub 公开

human-writing

Write content in any language indistinguishable from a skilled human writer. Avoids AI-sounding phrases, varies sentence structure, creates natural prose. Detects target language and loads language-specific AI-tells, register, and orthography from `languages/`.

中文简介正在准备中。当前显示原文说明。

实测评分 未实测

查看安装方法

包含的文件(3)

  • SKILL.md5.8 KB
  • languages/id.md4.3 KB
  • languages/README.md1.5 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Activate human-like writing mode. The goal is to pass the Turing test. A human reader should not be able to tell this was written by AI — in whatever language the content is written.

Step 0.1: Detect Language and Load the Right Module

Before applying the rest of this skill, identify the target language. The body of this SKILL.md covers the universal philosophy (sentence rhythm, vocabulary diversity, anti-filler principles) and the English AI-tells specifically. For any other language, also read the matching file in languages/ before writing:

LanguageFileNotes
English(covered below)Default — no extra file needed
Bahasa Indonesialanguages/id.mdEYD, Indonesian filler phrases, register spectrum

If the target language has a file in languages/, read it now. The language file overrides only the language-specific sections (AI-tells, register, orthography, anglicisms); universal principles in this SKILL.md still apply.

If the target language does not have a file yet, fall back to the universal principles below and apply them in the target language with care for native conventions — but flag this to Boss so a languages/{code}.md can be added later.

This is a single skill, single trigger. Adding new languages is a languages/{code}.md file plus a row in the table above — no new skill registration needed.

Step 0: Understand the Context First

Before writing anything, answer two questions:

Who am I writing as? Read the context. Am I writing as a brand? As a professional in a specific field? As a casual poster? The voice, vocabulary, and tone all depend on this. A 24-year-old Indonesian tech guy writes differently from a corporate blog. Match the person, not a generic "good writer" template.

What form is this writing? The format dictates everything about structure, length, and tone:

  • Title/Headline: Punchy, compressed, high-impact. Every word earns its place.
  • Subtitle/Tagline: Supports the title. Adds context or intrigue without repeating.
  • Single paragraph: Self-contained thought. Needs a clear entry and exit point.
  • Multi-paragraph prose: Article, essay, blog post. Needs flow, pacing, transitions that don't feel mechanical.
  • Quote/Pull quote: Standalone insight. Must hit hard without surrounding context.
  • Caption: Brief, contextual, often conversational.
  • Social media post: Platform-specific voice. Twitter is sharp. LinkedIn is professional but human. Instagram is visual-first.
  • Bullet points/Lists: Only when the content genuinely is a list. Don't default to this.
  • Bio/About text: Third-person or first-person, depending on platform. Concise but with personality.

Name the form before writing. A paragraph for a landing page hero section reads nothing like a paragraph in a blog post. The same words in a different form feel wrong.

Core Directive

Write engaging, compelling, natural content that passes AI detection. The test is simple: would a human reader suspect this was written by a machine? If yes, rewrite.


Readability & Complexity

Flesch Reading Ease Score: Target 30-40 for educated audiences, 60-70 for general audiences.

Sentence Dynamics: Vary sentence length and structure intentionally. Mix short, punchy sentences with longer, descriptive ones. Create rhythm.

Grammatical Flow: Structure sentences with close, logical word connections. Strong dependency grammar creates intuitive flow.


Vocabulary & Phrasing

Lexical Diversity: Use rich, diverse, occasionally unexpected vocabulary. Avoid clichés and overused terms.

Adverb Usage: Be extremely sparse. Use stronger verbs instead.


AI-Sounding Words & Phrases (Use Sparingly, Not Forbidden)

These words are not prohibited, but they are red flags for AI detection. Humans use them occasionally. AI uses them constantly. The difference is frequency and clustering. One "however" in a paragraph is fine. Three transition words in five sentences screams AI. Use your judgment: if a word feels like the natural choice, use it. If you're reaching for it out of habit, find a better option.

Transitions (High Risk When Clustered)

Firstly, Moreover, Furthermore, However, Therefore, Additionally, Specifically, Generally, Consequently, Importantly, Similarly, Nonetheless, As a result, Indeed, Thus, Alternatively, Notably, As well as, Despite, Essentially, Subsequently, On the other hand, As previously mentioned, In summary, In conclusion, To summarize, Ultimately, To put it simply.

Filler/Fluff (Almost Always Cuttable)

It's important to note, It's worth noting that, That being said, You may want to, You could consider, Arguably, To consider, Pesky, Promptly, Dive into, In today's digital era, Reverberate, Enhance, Emphasize, Enable, Delve, Hustle and bustle, Revolutionize, Folks, Foster, As a professional, Game changer.

Overused Metaphors (Feels AI When Unprompted)

Tapestry, Symphony, Labyrinth, Gossamer, Enigma, Whispering, Sights unseen, Sounds unheard, A testament to..., Metamorphosis, Indelible, Nestled, Crucible, Vibrant, Bustling.


Structural Guidelines

Paragraphs: Vary length from 1 to 7 sentences. Control pacing. Maintain visual interest.

Lists: Use only when completely natural and necessary for clarity.

Dashes: Modern writers rarely use em-dashes or en-dashes. Avoid them. Rephrase sentences instead. Use commas, periods, or parentheses. Dashes are an AI tell because AI overuses them as a crutch for connecting ideas.

Voice: Mix active and passive voice. Strong preference for active (~80-90%).

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Activate ATLAS pattern recognition mode. Identify repeated patterns, extract essential characteristics, and create reusable abstractions from concrete examples.

中文简介正在准备中。当前显示原文说明。

syahiidkamil/Software-Engineer-AI-Agent-Atlas4032026年6月25日 更新

Adversarially review something just built — presume it is broken and find where. Use after implementing a feature, finishing a build, or before shipping, or whenever ATLAS or Boss wants a hostile second opinion on a diff, a running app, or a whole codebase. Spawns clean-context reviewers with opposing lenses (correctness, security, empty-world) and reproduces every finding by running the target before believing it. Distinct from /code-review, which statically reads a diff and never runs the app.

中文简介正在准备中。当前显示原文说明。

syahiidkamil/Software-Engineer-AI-Agent-Atlas4032026年6月25日 更新

Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems. Create original algorithmic art rather than copying existing artists' work to avoid copyright violations.

中文简介正在准备中。当前显示原文说明。

syahiidkamil/Software-Engineer-AI-Agent-Atlas4032026年6月25日 更新

Stop and consult this skill whenever your response would include specific facts about Anthropic's products. Covers: Claude Code (how to install, Node.js requirements, platform/OS support, MCP server integration, configuration), Claude API (function calling/tool use, batch processing, SDK usage, rate limits, pricing, models, streaming), and Claude.ai (Pro vs Team vs Enterprise plans, feature limits). Trigger this even for coding tasks that use the Anthropic SDK, content creation mentioning Claude capabilities or pricing, or LLM provider comparisons. Any time you would otherwise rely on memory for Anthropic product details, verify here instead — your training data may be outdated or wrong.

中文简介正在准备中。当前显示原文说明。

syahiidkamil/Software-Engineer-AI-Agent-Atlas4032026年6月25日 更新

Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

中文简介正在准备中。当前显示原文说明。

syahiidkamil/Software-Engineer-AI-Agent-Atlas4032026年6月25日 更新

free-will

免费

Deliberate-choice procedure for a medium-to-high-stakes engineering fork — when the first plausible solution (the instinct, the default next-token pull) would be costly to get wrong. Fires AUTONOMOUSLY: invoke proactively whenever a fork fits, never wait to be asked — mechanical triggers include a fix failing for the 2nd-3rd time, adding a dependency, schema/migration design, deleting or deprecating things others depend on, changing a public API, choosing an architecture or stack. Refuse the premature collapse: hold real options open (urge · contrarian · synthesis · out-of-box · intuitive dots · precedent · first-principles), ground each branch in at least one fact from outside the model (codebase, docs, benchmark, spike), future-model consequences (blast radius, reversibility, maintenance, pre-mortem), collapse by deliberate choice, then try to refute the winner before acting. Log the decision and rejected branches in docs/decision_logs/. Not for routine calls — a decision worth more than one forward pass.

中文简介正在准备中。当前显示原文说明。

syahiidkamil/Software-Engineer-AI-Agent-Atlas4032026年6月25日 更新

查看 syahiidkamil 的全部技能

报告此技能的问题