How to use the Avoid AI Writing agent skill

What agent skills are and what this one does

An agent skill is a folder of instructions and reference files that an AI agent can load when a task matches the skill's description. This skill is named avoid-ai-writing and its job is to review a piece of writing, point out phrases and structures that read as machine-generated, and suggest or make fixes without removing the author's voice.

It works from reference files that grade how severe each pattern is, a tiered vocabulary list, and context profiles such as LinkedIn, docs, or investor email that change how strictly the rules apply.

• Detect mode: lists the flags only, grouped by severity, and separates clear problems from judgment calls.
• Rewrite mode (the default): returns a cleaned-up version with a summary of what changed, then re-checks its own rewrite.
• Edit mode: changes a prose file in place with minimal, targeted edits, then re-opens the file to confirm the flagged patterns are gone.

Install the avoid-ai-writing skill

Download the ZIP for this skill from our page. Then choose one of two routes depending on how you use your agent.

For a personal setup, unzip the download into ~/.claude/skills/. To make it available only inside one project, unzip it into that project's .claude/skills/ folder instead.

If you use the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills rather than unzipping it by hand.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with natural requests

You do not need a special command. The skill's description lists the kinds of requests that should activate it, such as asking to remove AI-isms, clean up AI writing, edit a draft for AI tells, or audit a README, changelog, release note, PR description, or blog post for machine-sounding prose.

To pick a mode in plain language, use words the skill recognises. For detect mode, words like "detect", "flag only", "audit only", "scan" or "what AI patterns are in this" work. Explicit options are also accepted: --mode rewrite|detect|edit, --voice, --context, --file PATH, and --iterate N for rewrite mode, where N is the total pass count including the built-in corrective pass, capped at 2.

• "Scan this README for AI patterns, flag only."
• "Clean up the AI writing in this release note."
• "Edit docs/CHANGELOG.md in place to remove AI-isms."
• "Make this blog draft sound less like AI, context blog."

Choose a context profile

A context profile sets how strict the skill is for a given audience. You can name one or let the skill infer it from the text. The available profiles are linkedin, blog (the default, with every rule at full strength), technical-blog, investor-email, docs, and casual.

Docs and technical writing get a lighter touch: the docs profile favours clarity over voice, and technical-blog lets technical terms pass. The skill says which profile it used, so check that it matches your audience.

Read the results carefully

Every flag is a writing-quality signal, not proof of how the text was produced. The skill's own documentation warns that AI detectors are unreliable, with high false-positive rates in some scenarios, so flags should not be used to decide academic integrity, hiring, or attribution questions.

The skill also forbids inventing personality, facts, or numbers during a rewrite. If a fix would require a new claim, it flags the gap and leaves it for you to fill.

• Check every rewrite against your own knowledge of the facts.
• Treat wordiness fixes as style suggestions, not evidence about authorship.
• Quoted, attributed, or third-party text is protected: tells inside it are reported but not rewritten.

What the skill needs and its safety notes

The skill lists no requirements. It does not run scripts and does not use the network, according to its listing. It is licensed under MIT.

Edit mode refuses source code, configuration files, and generated data. Point it at prose files such as Markdown documents.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no
• Licence: MIT