Your company deployed AI, told you to use it, sometimes in production, and never briefed the guardrails. Custom instructions are the one layer of that stack you control. This builder assembles yours from rules that each exist because somebody's production system paid for the lesson behind them.
Three fields and your seat are yours. The sections that are already solved get appended word for word: the policy line, retrieval rules, code rules, data hygiene, verification, and honesty. Resist rewording them. The precision is what makes them work.
The full document is ready.
Leave your first name and email and it unlocks, with copy and download. You will also get the Orí Intelligence newsletter, where the failures behind these rules get written. Unsubscribe anytime.
Submit the form and your document unlocks on its own. Already on the list? Send it again, it takes a second.
Paste it into your instructions in the company workspace, or install it in your coding agent: Claude Code reads it as CLAUDE.md in your repo root or home config. Most other coding agents read the same content as AGENTS.md, and Copilot reads it at .github/copilot-instructions.md. Same document, different filename. One check before you rely on it: confirm your instructions apply where you actually work; some tools scope them to chat on the web and stop there. Then test it: start a new chat and ask what standing instructions it's holding. Your rules should come back.
Now pressure-test it.
You wrote the rules. Ten situations from work that ships will show you which ones you'd actually follow.