Why Prompt Mirror Does Not Call an AI Model

1. Hypothesis

The hypothesis was that most prompts fail not because they lack creativity, but because they lack basic structural hygiene: a defined role, concrete specifics, and clear formatting constraints. If the tool could automatically enforce this structure locally, prompt outputs would improve without any prompt content leaving the browser.

2. What I built

Prompt Mirror — a client-side tool that rewrites prompts using five deterministic rules: "Add role", "Flag vagueness", "Set output format", "Remove hedging", and "Add context".

Crucially, it does not send the prompt to an LLM for "improvement". It uses regex functions to strip conversational filler ("please", "maybe") and insert placeholders for missing context.

3. Example Input/Output & UI Decisions

Input: "write me something about productivity please, maybe some tips or whatever, something good would be nice"

The UI Decision: Instead of a black-box "Magic Rewrite" button, I exposed the rules as distinct toggles. This shows the user why their prompt is changing.

Output (with all rules on): Context: [add relevant background here]

You are an expert. write me [specific something] about productivity, some tips or whatever, [specific something] [define "good" — e.g. concise, accurate] would be [define "nice" — e.g. concise, accurate]

Respond in clear, concise prose. Use bullet points if listing more than three items.

4. What I observed

• Observation: The output often looks clunky with literal placeholders like [specific something].
• Observation: The tool executes instantly because there is no network latency.
• Interpretation: The clunkiness is a feature, not a bug. By inserting [define "good"], the tool forces the user to actively think about their constraints rather than passively accepting an AI-generated rewrite. It acts as a "prompt structure linter."

5. What changed

No new rules have been added since the initial build. The five rules — Add role, Flag vagueness, Set output format, Remove hedging, Add context — remain as implemented.

6. Status

kept

FAQ

Why does the output look clunky?

The placeholders like [specific something] and [define "good"] are intentional. They force you to name what you actually want rather than accepting a vague AI-generated rewrite. The clunkiness is the feedback — a well-formed prompt should not need many placeholders.

Can I use the tool with only some rules enabled?

Yes. Each of the five rules is an independent toggle. You can disable "Remove hedging" if you want to keep polite phrasing, or activate only "Set output format" if that is the only structural gap in your prompt.

Does my prompt leave the browser?

No. Prompt Mirror uses local regex functions. The prompt text never reaches a server or an AI model.