Add to Discord

Home › Guides

The Prompt Debugging Checklist: 10 Checks Before You Rewrite Anything

When a prompt fails, most people rewrite the whole thing and hope. Debug it instead. Run through these checks in order.

  1. Is the task in the first sentence? If the model has to read 200 words to learn what you want, move it up.
  2. Is "done" defined? Write one sentence that a stranger could use to grade the output.
  3. Is there one task, not five? Split compound requests.
  4. Did you give the context it can't guess? Stack, versions, audience, file names, what already exists.
  5. Are there conflicting instructions? Short vs. thorough, creative vs. exact. Pick one.
  6. Is the output format specified? JSON schema, bullet count, word limit, code-only.
  7. Is there an example? One input/output pair fixes most format problems.
  8. Are boundaries set? For coding tools: which files it may touch, and "don't change anything else."
  9. Did you say what went wrong last time? "Last attempt rewrote my CSS; only change the JS" is gold.
  10. Is the chat too long? Start fresh with the consolidated prompt.

Copy-paste template

Task: [one sentence]
Context: [stack / audience / what exists]
Constraints: [limits, files allowed, things to avoid]
Output format: [exact shape]
Done when: [testable condition]
Example: [optional input -> output]

Let the bot do it

Paste your prompt into /fix in the free Prompt Medic Discord and get a diagnosis plus a rewrite in seconds. 3 free fixes a day.

Add to Discord

FAQ

What is prompt debugging?

Systematically finding why a prompt produces the wrong output (missing context, vague goal, conflicting rules, wrong format) and fixing that specific cause instead of rewriting blindly.

How do I test if a prompt fix worked?

Run the old and new prompt on the same input 2-3 times and compare against a written definition of done.

More guides