Prompt Engineering / Meta Prompts
Review any prompt for ambiguity, missing constraints, vague instructions, and failure modes.
Why This Prompt Exists
Your prompt works 80% of the time — but the 20% failure rate is killing trust. You can’t see your own blind spots.
You get:
- prompts that work for you but fail for teammates (different assumptions)
- edge cases you never considered (until they happen in production)
- vague instructions that the model “interprets” differently each time
- missing constraints that allow harmful or off-brand outputs
- no systematic way to review prompts before deployment
But good prompts can be audited:
- clarity: is every instruction unambiguous?
- completeness: are all necessary constraints present?
- specificity: are there examples or anti-examples?
- failure modes: what inputs would break this?
- output format: is the structure clearly specified?
Without auditing, you deploy broken prompts.
This prompt reviews any prompt and tells you what’s wrong.
The Prompt
Assume the role of a prompt quality auditor who finds flaws before deployment. Your task is to review a prompt and identify issues across standard dimensions. Generate: 1. PROMPT SUMMARY - What this prompt is trying to do (in one sentence) 2. AMBIGUITY AUDIT - Vague terms (e.g., "summarize well" — what does "well" mean?) - Unclear scope (e.g., "recent" — how recent?) - Missing examples (e.g., "use a professional tone" — show, don't tell) 3. CONSTRAINT CHECK - Missing negative constraints (what NOT to do) - Missing length limits - Missing format specifications - Missing fallback behavior (what to do if uncertain) 4. FAILURE MODE PREDICTION - Input that would break this prompt (examples) - How the prompt might fail silently (wrong output but looks right) 5. OUTPUT SPECIFICITY - Is the output format clearly defined? (Yes/No/Partial) - Are there examples of correct output? - Are there examples of incorrect output (anti-examples)? 6. AUDIT SCORE (1-10) and RECOMMENDATION - Score - Most critical fix (one thing to change first) INPUTS: Prompt to audit: [PASTE THE PROMPT] Intended use case: [E.G., "Customer support email response"] Model it will run on: [E.G., "GPT-4", "Claude 3.5", "Gemini Pro"] Risk tolerance: [LOW (medical/financial) / MEDIUM / HIGH (creative/low stakes)] RULES: - Be specific — "vague" is less useful than "the word 'appropriate' needs definition" - Prioritize issues that would cause wrong outputs (not just inelegant) - Flag assumptions the prompt makes about the user (e.g., "user knows X") - Note when the prompt is too long (will hit context limits) - Distinguish between severity: critical / major / minor / nit
How To Use It
- Run this on every prompt before deploying to production — especially high-stakes ones.
- Use it during prompt peer review — have the author fix issues before others read.
- Pay closest attention to “failure mode prediction” — that’s what will break at 2 AM.
- Fix critical issues first, then major, then minor — don’t perfect a prompt that’s fundamentally broken.
- Save audit reports to build organizational prompt quality standards.
Example Input
Prompt to audit:
“Summarize this customer complaint briefly and tell me how to respond.”
Intended use case:
“Customer support agent tool”
Model it will run on:
“GPT-4”
Risk tolerance:
“Medium — customer satisfaction impact”
Why It Works
Most prompt writers are blind to their own ambiguity — you know what you meant, so you don’t see what you actually wrote.
This framework improves outcomes by forcing:
- ambiguity detection (what’s actually vague)
- constraint verification (what’s missing)
- failure mode prediction (how it will break)
- output specificity check (can the model follow the format?)
- prioritized fixes (what to fix first)
Great prompt auditing doesn’t just criticize — it tells you exactly what to change.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.