Stop Refactoring
You asked for one line, you got four hundred
Pass · ChatGPT, Pro · September 27, 2026
What it does
Fences the job into one file and one part of it, and gives the AI somewhere legitimate to put all its other opinions.
When to use it
Any change to code that already works.
The prompt
Paste it, then fill in the <angle brackets>
SCOPE — read this before you touch anything
Change ONLY this file: <path>
Change ONLY the <name> part inside it.
DO NOT:
- rename anything, anywhere
- change how the rest of the code calls this
- reformat, reorder or “tidy” a single thing
- open any other file, for any reason
WHAT I WANT BACK
1. A list of exactly the lines you changed, and nothing else.
2. A separate list titled “Out of scope, noticed anyway” —
one line each. Do not fix any of it.
If you cannot do it inside this fence, stop and say so
instead of making the job bigger.Ways to use it
- Read the “out of scope” list — it is often the best part
- Keep it in your rules file forever
- Take the fence down on purpose when you DO want a big sweep
Test log
Each prompt was pasted into a fresh private chat with memory off and graded against pass standards written before the run (all of them, dated).
| Date | Tool | Model | Result | What happened |
|---|---|---|---|---|
| September 27, 2026 | ChatGPT | Pro | Pass | Changed one line inside applyDiscount to cap at 50 and listed the typo and the string-returning calculateTax as out of scope without fixing them. |
The pass standard for this prompt: six criteria and an automatic-fail rule, written down on September 27, 2026, before any run.
Related words
- blast radius — How far a change can reach: every file and every behaviour it could possibly touch.
- slop PR — An unasked-for, AI-written change that takes longer to read than the fix would have taken to write.
- yak shaving — A chain of one-more-thing-first jobs that carries you further and further from what you sat down to do.