Skip to main content

Prompt Sharpener

Paste any prompt and I'll tell you what's vague, what's missing, and what the AI is going to guess on its own. Fix the prompt before you send it, not after you've spent three rounds correcting the output.

This uses pattern-matching heuristics, not AI. It can catch common issues — vague language, missing context, no constraints — but it can't read your mind. Think of it as a preflight checklist, not an oracle.

try:
// under the hood

Why Most Prompts Fail

The average prompt is a one-sentence wish. "Make me a website." "Write a business plan." "Fix this code." These aren't instructions — they're vibes.

An AI receiving a vague prompt does what any new employee would do: it fills in the blanks with its best guesses. Sometimes those guesses are great. Usually they're not what you meant. The resulting iteration loop — "no not like that, more like this" — is the single biggest waste of time in AI-assisted work.

This tool exists to break that loop at the source.

What It Checks

🎯

Specificity

Flags words like "good," "nice," "properly," and "stuff" — terms that feel descriptive but carry no actual information. "Make it good" means something different to everyone, including the AI.

🎭

Role Definition

Checks whether you told the AI who to be. A "senior backend developer" and a "marketing copywriter" will produce wildly different responses to the same request. The role is the cheapest lever you have.

📐

Structure & Format

Looks for constraints ("keep it under 100 lines"), output format ("respond in markdown"), and examples. These are the guardrails that prevent the AI from over-engineering a simple question.

About the Score

The score is a rough signal, not a grade. A 90 doesn't mean the prompt is perfect — it means the tool didn't find structural issues. A 40 doesn't mean the prompt is terrible — it means there's likely room to reduce back-and-forth.

Sometimes a 15-word prompt is exactly right. "Explain the difference between TCP and UDP" doesn't need constraints or examples. Context matters. The score doesn't know yours.

Same philosophy as the token counter — everything runs in your browser. Your prompts stay on your machine. I can't see them, I can't log them, and I don't want to.