Did it work?

Ninety seconds. You run the same prompt twice, once before the file is installed and once after, and compare the two. A user who runs it once gets no proof at all.

1. Before you install anything

Paste this into your assistant and keep the answer.

Write a three-line email telling a customer their order will be a week late.

Keep the first answer. Without it there is nothing to compare and you would have to start again.

2. Install the file

Follow the section of INSTALL.md for the tool you use.

3. Run the same prompt again

In a fresh chat, not the one you have been using.

4. Put them side by side

Your file lists its checkable rules. Count how many each answer breaks. That describes the text without judging it: there is no score, no traffic light and no target you have failed to hit. The same count is computed by the style checker in your download, so you can re-run it and get the same answer.

If it did not work

1. Check it loaded. Claude Code: /context, under Memory files. GitHub Copilot Chat: expand the references list. ChatGPT, Claude and Gemini: reopen the settings screen. Cursor publishes no way to check this.
2. Try the shorter version. Long files get ignored.
3. Look for two rules that fight. When rules contradict, the assistant picks one at random.
4. Make the rule checkable. A named phrase works. A feeling does not.
5. Start a new chat. Some tools apply the file only to conversations started after it was added.
6. Say it again in your message. When your message and a file disagree, the model follows your message.

If none of those moves it, the answer may be that you are asking for something no file can deliver. What this will not do, and why sets out the limits this tool refuses to paper over.

Also from Best Served Cold.