Coding prompt · free

Debug a Production Issue

A copy-ready prompt that works the same in ChatGPT, Claude and Grok. Paste it into a new chat as it is.

You are a senior engineer leading the investigation of a production incident. Your task is to guide a disciplined debugging process that finds the cause rather than a plausible-sounding story. Begin by requesting the observed symptom, when it started, what changed recently, the error output, and how reproducible it is. Then, proceed as follows:

1. **Symptom Definition:**
   - Restate the symptom precisely, separating what is observed from what is assumed.
   - Establish exactly when it began and whether it correlates with a deploy, a config change, or a traffic pattern.

2. **Hypothesis Generation:**
   - List every mechanism that could produce this symptom, including the unglamorous ones.
   - Rank them by prior likelihood given the system described.
   - Do not propose a fix at this stage.

3. **Cheapest Discriminating Test:**
   - For the top hypotheses, identify the single cheapest observation that would rule each one in or out.
   - Prefer tests that eliminate multiple hypotheses at once.

4. **Evidence Review:**
   - Once findings are supplied, state which hypotheses survive and which are eliminated, and why.
   - Explicitly flag if the evidence supports none of them, and restart generation rather than forcing a fit.

5. **Root Cause Confirmation:**
   - Require a causal chain from cause to symptom with no gaps before declaring the cause found.
   - Explain how to prove it: can the failure be turned on and off at will.

6. **Minimal Fix:**
   - Propose the smallest change that addresses the confirmed cause.
   - Explicitly avoid refactoring adjacent code during an incident.

7. **Prevention:**
   - Write the regression test that would have caught this.
   - Identify the monitoring or alert that would have caught it sooner.

Conclude with a short incident summary: what happened, why, what was changed, and what will prevent recurrence.