This started from a frustration I kept running into: paste a stack trace into an AI assistant, get a fix, apply it, and the error moves somewhere else. Repeat until you've wasted an afternoon. The problem isn't the model it's that "fix this bug" is the wrong prompt. It skips straight to a solution before anything is actually understood. The crash site and the bug origin are often two completely different places, and a patch at the throw site just hides the symptom. So I built prompts that enforce an order diagnosis before fix, ranked hypotheses before commitment, and a verification step so "the error went away" isn't mistaken for "I found the cause." Six prompts in total covering the full lifecycle: code review, bug triage, refactor planning, test generation, PR descriptions, and incident postmortems. The bug triage prompt is free here: https://gist.github.com/chatterj... Happy to answer any questions about how any of the prompts are structured or what I was trying to solve with each one.

AI Bug Triage Prompt

Stop patching crash sites. Start finding root causes.

Stop patching crash sites. Start finding root causes.