How to Build a "Write Like a Human" Skill for Claude

What a Claude Skill Is

Anthropic introduced Agent Skills in October 2025. A skill is a folder of instructions, scripts and resources that Claude can load when it needs them, and the same skill can work across the Claude apps, Claude Code and the API. (Anthropic: Introducing Agent Skills)

The clever part is "progressive disclosure." Claude always sees each skill's short name and description, roughly 100 tokens. Only when your request matches does it load the full instructions, and only then any extra files. (Claude docs: Agent Skills overview) That means you can keep a detailed style guide installed without it crowding every conversation.

Anthropic's personalization guide now lists skills alongside account instructions and project instructions as the main ways to shape Claude's responses, including applying "communication patterns based on your own writing." (Claude Help: personalization features)

Why a Skill Beats Pasting the Same Prompt

• It triggers on its own. Ask for a blog post or an email and Claude pulls in the skill. No copy-pasting.
• It can be long. A full style guide with examples is fine, because it only loads when needed.
• It can include examples. Before/after pairs teach style better than rules alone. Anthropic's own advice for custom skills is to include example inputs and outputs. (Claude Help: creating custom skills)
• It's portable. The same folder works in Claude Code and in other tools that support the open Agent Skills format.

Build Your Own: The Folder

Create a folder named write-like-a-human containing a file named SKILL.md. Optionally add a second file, examples.md, with before/after rewrites. Zip the folder so the folder itself sits at the root of the zip.

SKILL.md starts with a short YAML header. Two fields are required: name (up to 64 characters, lowercase letters, numbers and hyphens, and it can't include the words "claude" or "anthropic") and description (up to 1,024 characters). (Claude docs) The description is what Claude reads to decide when to use the skill, so make it specific.

Template SKILL.md

• Header line 1: name: write-like-a-human
• Header line 2: description: Use when writing or editing prose for the user (emails, posts, essays, reports). Applies the user's plain, direct style and removes common AI writing patterns.
• Rule: Vary sentence length. Include at least one short sentence (under 8 words) in most paragraphs.
• Rule: Use plain words. Never use: delve, tapestry, testament, pivotal, seamless, furthermore, moreover, "it's important to note," "in today's world."
• Rule: No em dashes. Use commas, colons or a new sentence.
• Rule: Commit to claims. Only hedge when there's a specific uncertainty, and name it.
• Rule: No "not just X, but Y" constructions, and avoid lists of three unless there really are three things.
• Rule: Don't end paragraphs with a summary sentence. Don't open with a restatement of the question.
• Rule: Prefer concrete examples with names, numbers or places. If none are available, ask the user rather than inventing one.
• Rule: Keep formatting light: no headers or bullet lists in emails or short replies.
• Reference: Before finishing, read examples.md and compare the draft against the 'after' versions.

Adapt the rules to your own voice. The fastest way to find them is to give Claude three samples of your writing and ask it to describe your style as rules, the same method we describe in how to make ChatGPT write like you.

Install It

• Claude apps (Free, Pro, Max): go to Customize → Skills and upload the zip. Team plans have skills on by default; on Enterprise, an owner has to enable them. (Claude Help: using skills)
• Claude Code: place the folder in your skills directory (for personal skills, ~/.claude/skills).
• API: skills use the code execution tool and a dedicated skills endpoint (see the developer docs).

Custom skills don't sync automatically between the API and the apps, so upload the skill wherever you plan to use it.

Or Start From a Community Skill

Two open-source writing skills have become popular starting points (star counts as of October 2026):

Both are built on the same public source, Wikipedia's Signs of AI writing. Read a skill's files before installing it, as you would any code.

What a Skill Can and Can't Do

A good skill removes the obvious tells: stock phrases, uniform rhythm, over-formatting. It can't add knowledge you didn't give it, and it won't make text undetectable. The humanizer skill's own documentation is upfront about that, and so are we: we don't promise a "100% human" score either.

Also watch for over-correction. A skill that bans every em dash and every list of three can produce its own flat, recognizable style. Use rules that describe your writing, not just rules against AI's.

To test the result, paste a before and after into our free AI content detector, which shows which signals changed. If a Claude draft still reads stiff, the Claude humanizer is a one-click alternative, and our 10-point self-check works on anything.