Bug Find Prompt Template: How to Get AI to Actually Prioritize What It Finds

"Find the bugs in this code" almost always works — the harder problem is what comes back: a list where a genuine null-pointer crash sits right next to a naming nitpick, with no signal for which one to fix first. The fix isn't a smarter model; it's a template that asks for severity up front.

The Bugs & Logic Errors Category

A focused bug-finding prompt should name what it's actually looking for, rather than leaving "bug" open to interpretation:

• Incorrect logic
• Off-by-one errors
• Improper null or undefined handling
• Race conditions

Naming these explicitly matters because a vague "find bugs" prompt tends to default to surface-level issues — a model will happily point out a typo in a comment before it traces through a subtle off-by-one in a loop boundary, unless specifically asked to look for the latter.

Why Severity Tagging Changes the Output

Asking for a severity level per issue — Critical, High, Medium, Low — forces a form of triage that a plain bug list doesn't. A race condition that could corrupt data under load is a different priority than an unused variable, and a template that requires severity per finding makes the model actually reason about that difference instead of listing everything as equally worth fixing.

An Example Template Output

Act as an experienced JavaScript code reviewer.

Review the code below for:
1. Bugs and logic errors — incorrect logic, off-by-one errors, improper null/undefined handling, and race conditions.

For each issue found, report its location, a description of the problem, and a severity level (Critical / High / Medium / Low).

Code to review:
```javascript
[PASTE YOUR CODE HERE]
```

Narrowing the focus to just this one category, with severity required, keeps the model from padding the response with quality or style comments that dilute the actual bug list.