Prose Guard

prose-guard

You have read a review comment from an agent that nobody could digest: technically right, three clauses deep, and impossible to act on. You have seen a Slack message with six bolded section labels over four lines of content. You have been told a deploy is fixed in a paragraph that never says what you now have to run.

Nothing about those messages is badly written. They fail because they were not written for anybody in particular.

`prose-guard` is a Claude Code plugin that checks a message on its way out — a chat message, a review comment, a commit message, a pull request description, a document — against two questions:

**Who is going to read this?** **Why should they care?**

Install

/plugin marketplace add novada-tech/prose-guard
/plugin install prose-guard@novada
/prose-guard:setup

If the install summary says `Run /reload-plugins to activate`, run that first — otherwise the hooks are not loaded yet and `/prose-guard:setup` has nothing to configure.

**Run `/prose-guard:setup` before anything else.** Nothing is checked until a level is chosen, and this is where you choose one. It takes a few minutes, asks one question at a time, and does three things.

• Asks how much checking you want, with what each level costs. Pick medium or high — what separates them is what can hold a message back, not what it costs.
• Offers to install the writing rule: 250 words that reach Claude while a message is being written, rather than when it is sent. It is the only part of this that costs nothing per message.
• Reads what is on your machine and proposes which of your tools count as sending prose, rather than assuming. Chat, issue trackers and vendor command-line tools go in here.

Skip it and prose-guard says so at the start of every session, because an install that checks nothing looks exactly like one that has nothing to object to.

Then, when you have a few more minutes:

/prose-guard:audiences

Optional, and it