Moving your rules into a config file

How to stop repeating the same instructions in every message, and what belongs in a system prompt against what belongs in the task.

AI Place Hub 1

If you paste the same paragraph into every conversation, it belongs somewhere else. Every major product now has a place to put standing rules, and moving them there makes answers more consistent and your messages shorter.

Where the file lives

The mechanism is the same everywhere, only the name changes. ChatGPT calls it custom instructions or project instructions. Claude Code reads a CLAUDE.md from the repository root. Several agent harnesses read AGENTS.md. Editors use a rules file. The API calls it a system parameter. All of them do what the system-prompt glossary page describes: text that applies to every turn.

What belongs there

The test is simple. Would you otherwise paste it into every message?

Role and expertise level pass. Output preferences pass - no preamble, code first, name the version you are assuming. Domain constraints pass: the stack, the conventions, the things never to touch. Refusal rules pass.

The task itself does not. Nor does anything that changes between conversations. A config file that contains this week's project is a config file you will forget to update.

Keep it short

This is where most people go wrong. A 2000-word rules file dilutes itself: the model has to weigh forty instructions against each other, and the three that matter get the same attention as the thirty-seven that do not.

Write the rules that would change an answer. Delete anything that describes default behaviour you would get anyway. "Be helpful and accurate" is not a rule.

Test that it is loaded

Put one specific, checkable instruction in the file - a formatting quirk you would notice immediately - and start a fresh session. If the quirk does not appear, the file is not being read, and everything else in it is doing nothing either. This catches the wrong directory, the wrong filename and the wrong project more reliably than reading documentation.

Ready-made starting points

The configs section has working files for a PHP project, a TypeScript repo and ChatGPT custom instructions. Take one, delete two thirds of it, and add what is actually true about your setup.

Notes from the author

Assumes you have used a chat model for a while and are repeating yourself.

Fork

Did this work for you?

Sign in to flag it

Related