AI Prompt Enhancer
Score your prompt against the 8 components prompt engineers care about — role, task, context, constraints, format, examples, audience, tone — then rewrite it with structured sections. Markdown / XML / ALL-CAPS styles. 100% browser-side.
What is the AI Prompt Enhancer?
AI Prompt Enhancer is a heuristic prompt analyser and rewriter — entirely client-side, with no LLM call. It scores your input against the 8 components prompt-engineering literature consistently cites: a role for the model, a clear task verb, surrounding context, constraints, output format, examples, audience and tone. A 0–100 completeness score and per-component pass/fail card pinpoints exactly what's missing. The rewriter then composes a structured prompt with placeholders for every missing component, choosing one of three styles you pick: Markdown sections (## Role / ## Task / …) as a clean universal default; XML tags (`<role>…</role>`) optimised for Claude; or ALL-CAPS sections as a plain-text fallback for older models. Your original prompt is appended at the bottom for reference. Because everything runs locally, your prompt — and any sensitive context inside it — never leaves your browser.
How to use it
- Paste your prompt into the editor.
- Read the 8-check pass/fail card and the completeness score.
- Pick a section style — Markdown (default), XML, or ALL-CAPS.
- Copy the enhanced prompt into ChatGPT, Claude, Gemini or any other LLM.
Benefits
- Heuristic scoring against the 8 components prompt engineers care about — no API key required.
- Live completeness score (0–100) updates as you type with React deferred values.
- Pass/fail card for each component with actionable hints when something's missing.
- Three rewrite styles: Markdown sections, XML tags (for Claude), ALL-CAPS sections.
- Preset prompts cover common starting points: vague request, blog idea, code help, email draft.
- Original prompt preserved at the bottom of the rewrite for reference.
- Persists last input and style in localStorage for one-click resume.
- Runs 100% in your browser — your prompt is never sent to any AI service.
Frequently asked questions
Does this call ChatGPT or Claude to improve my prompt?
No — the whole tool is heuristic, client-side. We pattern-match for the 8 prompt components and compose a structured rewrite. That keeps your prompt private and the tool free.
What are the 8 components scored?
Role (you are a …), task (action verb), context (background), constraints (rules and limits), format (output shape), examples (few-shot pairs), audience (who it's for), tone (voice/style). They're weighted in the 100-point score.
Why is the score sometimes low even though the prompt is short and clear?
Short prompts often miss audience, tone, format and examples — even if the task is clear. The 8-check grid shows which ones are missing so you can decide whether to add them.
Which rewrite style should I use?
Markdown sections are the universal default and work everywhere. XML tags are documented to give the best results with Claude. ALL-CAPS sections are a plain-text fallback for older or stricter models.
What if the rewrite has placeholders I don't want?
Edit them out — the rewrite is a scaffold, not a final prompt. Removing a section you don't need is faster than typing it from scratch.
Does it work in other languages?
The pattern matching is English-centric, so the score will be lower for non-English prompts. The structural scaffold still works for any language — just translate the section headings.
How big can the prompt be?
There's no hard cap. Larger prompts work but the analysis runs on every keystroke (debounced) — multi-thousand-word prompts may feel slightly less responsive.
Is my prompt logged or stored?
Only locally — your last prompt and style choice are saved to localStorage on this device. Toollyz has no server that ever sees the text.
Can I share the enhanced prompt?
Yes — copy or download the rewrite as Markdown, XML or .txt and paste into any LLM. Markdown reads cleanly in Cursor, ChatGPT, Claude and most other UIs.
Should I always use the rewrite verbatim?
No — treat it as a scaffold. Fill in the placeholders with real context, remove sections you don't need and adjust the tone. The structure is the value, the words inside are yours.
Why isn't there an example library?
Examples are highly task-specific — we surface a placeholder block and trust you to add a real input/output pair. A generic example library would dilute the signal.