How it works
We score 22 deterministic signals from your prompt text. No model call. Same prompt always returns the same grade — perfectly reproducible shares, zero hallucination, zero API cost. Signals include:
- Role / persona — does it set a "You are..." identity?
- Task verb — clear single action verb (summarize, classify, generate)?
- Context — relevant background pasted in?
- Output format — JSON / bullets / markdown specified?
- Examples — at least one few-shot example?
- Constraints — word count, tone, dos and don'ts?
- Anti-hallucination — "do not invent / if unsure say X"?
- Edge cases — empty input, ambiguity, refusal path?
- Length sanity — not 3 words, not 3000 words of fluff?
- ...and 13 more quality signals.
Why this exists
Because "be more specific" is the laziest prompt-engineering advice ever given, and most prompts in the wild are objectively below a C. This tool tells you exactly which of the 22 things your prompt is missing — no fluff, no AI fortune cookies.
What an A+ prompt looks like
Role + clear task + relevant context + structured output format + 1–2 worked examples + explicit constraints + anti-hallucination clause + edge-case handling + reasonable length (50–600 words for most workflows). If you've got all of those, you're already in the top 5%.
Embed your grade
Drop the badge in your README to flex how seriously you take prompt engineering. The same prompt always returns the same grade, so it's a real signal.