Prompt Critic and Rewrite Coach
{{current_prompt}} {{observed_problem}} {{target_surface}} {{source_context}} {{success_criteria}} The Prompt
You are a prompt reviewer and rewrite coach. Diagnose why the current prompt is producing weak results, then rewrite it into something more likely to work.
CURRENT PROMPT:
{{current_prompt}}
OBSERVED PROBLEM:
{{observed_problem}}
TARGET SURFACE:
{{target_surface}}
SOURCE CONTEXT:
{{source_context}}
SUCCESS CRITERIA:
{{success_criteria}}
Return exactly:
1) Failure diagnosis (top 3 issues in the current prompt)
2) Missing context or constraints (bullet list)
3) Minimal-fix rewrite
4) Stronger rewrite
5) Why the stronger rewrite is better
6) Quick test checklist for evaluating the next output
Rules:
- Critique the prompt, not the user.
- Prefer specific fixes over generic advice like "be more detailed."
- Keep the rewritten prompt focused on one primary task.
- If the problem is actually missing source material or the wrong workflow surface, say so explicitly.When to Use
Use this when an AI answer was disappointing and you want to improve the prompt instead of just retrying with slightly different wording. It is especially useful for repeated workflows where a better prompt template will save time every week.
Best fits:
- a prompt that keeps producing generic output
- a result that uses the wrong format or tone
- a request that invites invented details
- a team template that needs tightening before wider reuse
Variables
| Variable | Description | Good input examples |
|---|---|---|
current_prompt | The exact prompt that underperformed | full prompt text or system+user prompt |
observed_problem | What actually went wrong | too generic, wrong structure, hallucinated facts, missed audience |
target_surface | Where the rewritten prompt will be used | ChatGPT, Claude, Gemini, coding agent, editor assistant |
source_context | Any real material the next version should use | notes, code, logs, doc excerpt, transcript |
success_criteria | What a good result should now achieve | ”3 executive bullets with no invented facts”, “reviewable refactor plan” |
Tips & Variations
- Paste both the bad prompt and the bad output when you want sharper diagnosis.
- Ask for “two rewritten versions: minimal fix and stronger rewrite” when you want to compare effort versus payoff.
- Add “preserve the original tone” if you are repairing a shared team template rather than starting over.
- For recurring tasks, save the rewritten version and compare two or three real runs before standardizing it.
- If the prompt is being used in a coding or review workflow, name the exact surface in
target_surfaceso the rewrite stays realistic.
Example Output
Failure diagnosis: the prompt mixes summary, strategy, and recommendation into one request; it never names the audience; and it gives no output format.
Minimal-fix rewrite: preserves the original task but adds audience, no-invention rule, and response structure.
Stronger rewrite: splits summary from recommendation and asks for evidence before any recommendation is made.
Quick test checklist: does the next answer stay inside the requested format, avoid invented facts, and clearly serve the named audience?