My first Claude Code skill: no em-dashes, ever

I published my first Claude Code skill. It is called no-emdashes.

What is a Claude skill?

Claude Code is Anthropic's coding assistant that runs in your terminal. A skill (or plugin) is a small piece of configuration you drop into a repository that tells Claude Code what to know, how to behave, and what rules to enforce. It runs locally, loads automatically, and saves you from repeating yourself every session.

You can use skills to give Claude context about your codebase, enforce coding standards, or block behaviour you do not want. No manual prompting required.

Why no-emdashes?

The em-dash is an AI smell. It is the typographic equivalent of "I'd be happy to help": technically fine, but a clear signal that a machine wrote this. I do not want it in my output. Not in my blog posts, not in my code comments, not anywhere.

I also run an AI assistant called Dobbie at home, and it still slips one through occasionally. The problem is real enough to automate.

The skill works in two ways:

• Before every turn: injects the rule into the conversation so Claude knows upfront
• After every response: scans the output for any of 7 em-dash variants (including en-dash, horizontal bar, and their Unicode cousins); if one is found, Claude is forced to rewrite

It also adds a /no-emdashes slash command for an on-demand check.

How to install it

/plugin marketplace add woutersf/woutersf-marketplace
/plugin install no-emdashes@woutersf-marketplace

Requires Claude Code CLI and Python 3.

How I use it

Every repo I work in gets this skill. It is the first one I drop in. Once it is there, I do not think about it.

Try it and let me know

The code is on GitHub. I am curious what you think: whether the approach works for you, what other skills you would want, or if you find edge cases I have not handled.

Written with assistance from Dobbie, Frederik's AI assistant.