✨ MetaPrompt — AI Prompt Engineer & Optimizer

Turn any vague idea into a master-class prompt instantly. Free. Open source. No sign-up.

MetaPrompt is a browser extension that acts as your personal Prompt Engineer. It rewrites and structures your prompts before you send them to ChatGPT, Claude, or Gemini — so you always get dramatically better AI responses with zero extra effort.

🧠 How It Works

MetaPrompt works in two distinct modes depending on where you are:

1. 🚀 Optimizer Mode (For ChatGPT, Claude, Gemini)

Goal: Turn vague ideas into master-class prompts.

You type: "make a snake game in python"
MetaPrompt: Picks an Expert Persona (Senior Dev) + Chain of Thought structure.
Result: You get a professional, structured prompt that gets perfect code on the first try.

2. ✨ Refiner Mode (For LinkedIn, Instagram, Gmail, etc.)

Goal: Write better messages, DMs, and responses instantly.

You type intent: "tell her about her beauty and eyes"
MetaPrompt: Writes the actual message in a charming, casual tone.
You type a draft: "hey i want to meeting tomorrow at 5"
MetaPrompt: Polishes it into a professional invitation.

🎭 Multipurpose: Professional & Personal

MetaPrompt is more than just a prompt tool — it's your personal AI ghostwriter.

💼 Professional (LinkedIn/Email): Transform rough notes into articulate, high-level business communication. Perfect for cold DMs, follow-ups, and networking.
❤️ Personal & Casual (Instagram/DMs): Express yourself better. Whether you want to be heartfelt, playful, flirty, or deeply romantic, MetaPrompt understands your intent and writes the actual message for you.
✏️ Grammar & Correctness: Fix typos and punctuation while keeping your original voice perfectly intact.

Feature	Optimizer Mode	Refiner Mode
Core Job	Rewrites prompts to AI	Writes/Polishes messages for people
Best For	Coding, Writing, Analysis	DMs, Emails, Social Media
Platforms	ChatGPT, Gemini, Claude	LinkedIn, Instagram, WhatsApp, Gmail
Output	"Act as a..." structure	Direct, ready-to-send text

✨ Features

🎯 Dual-Engine Logic — Automatically switches between Prompt Engineering and Message Ghostwriting.
🧠 6 Prompt Techniques — Auto, Expert Persona, Chain of Thought, Few-Shot, Zero-Shot, ReAct.
🎨 3 Refining Tones — Professional, Casual/Personal, and Grammar-Only.
🔍 Intent Detection — Understands if you are giving instructions ("tell her...") or a draft ("hey ur pretty") and responds accordingly.
⚙️ Dynamic Model Selection — Choose exactly which model to use from a live dropdown.
🔍 Live Model Discovery — Detects which Gemini models are available on your API key.
🛡️ Smart Fallback — If a model is down or removed, silently tries the next working one.
📴 Works Offline — Built-in smart templates need zero internet.
🔒 Privacy First — Your API key never leaves your browser.
⌨️ Keyboard Shortcut — Alt + Shift + P to enhance any prompt instantly.
➕ External Site Support — Use MetaPrompt on ANY website (Instagram, LinkedIn, etc.) by adding them in the settings.
♾️ Always Alive — Dual keep-alive strategy prevents service worker sleep.

🚀 Quick Start

Step 1 — Get the Code

Option A — Direct ZIP Download (Easiest):

Download the extension: Download MetaPrompt (ZIP)
Extract the ZIP anywhere on your computer (e.g., C:\MetaPrompt\ or ~/MetaPrompt/)
The extracted folder exts-main contains the extension files.

Option B — Clone with Git:

git clone https://github.com/jigs1188/exts.git

Step 2 — Load into Your Browser

See the full browser-specific guide below.

Step 3 — Use It

Go to chatgpt.com, claude.ai, or gemini.google.com
Type any idea into the chat box
Click the ✨ floating button (bottom-right corner) or press Alt+Shift+P
Your prompt is instantly upgraded! 🚀

🌐 Browser-by-Browser Installation

🟢 Google Chrome

Versions supported: Chrome 88+

Open a new tab and go to: chrome://extensions
In the top-right corner, enable the Developer mode toggle
Click "Load unpacked" (appears after enabling Developer mode)
In the file picker, navigate to where you extracted the ZIP (or cloned the repo)
Select the exts-main (or exts) folder → click Select Folder
MetaPrompt appears in your extensions list ✅
Click the puzzle icon 🧩 in the toolbar → click the 📌 pin icon next to MetaPrompt

Verify it works: Visit chatgpt.com — you should see the ✨ button floating at the bottom-right.

🔵 Microsoft Edge

Versions supported: Edge 88+ (Chromium-based)

Open a new tab and go to: edge://extensions
In the bottom-left sidebar, enable the Developer mode toggle
Click "Load unpacked" in the top-left
Navigate to your extracted folder and select exts-main (or exts)
Click Select Folder → MetaPrompt appears ✅
Click the puzzle icon 🧩 in the toolbar → pin MetaPrompt

Verify it works: Visit chatgpt.com — ✨ button should appear.

🟠 Brave Browser

Versions supported: Brave 1.20+ (Chromium-based)

Open a new tab and go to: brave://extensions
Enable Developer mode toggle (top-right)
Click "Load unpacked"
Select the exts-main (or exts) folder
Click Select Folder → MetaPrompt is loaded ✅

Brave Shield note: If the ✨ button doesn't appear on some sites, click the Brave Shield 🛡️ icon in the address bar and set "Shields" to Down for that site. MetaPrompt doesn't need this normally, but Brave's aggressive blocking occasionally affects injected UI elements.

🔴 Opera / Opera GX

Versions supported: Opera 74+ (Chromium-based)

Open a new tab and go to: opera://extensions
(Opera GX: operagx://extensions)
Enable Developer mode (top-right toggle)
Click "Load unpacked"
Select the exts-main (or exts) folder → click OK ✅

🦊 Mozilla Firefox

Because Firefox handles background scripts slightly differently than Chrome, we have a build script that automatically prepares the Firefox version for you.

Step 1: Build for Firefox

Ensure you have Node.js installed.
Open your terminal in the extension folder and run:
```
npm install
npm run build
```
This creates a dist/firefox folder with the correct Firefox manifest.json.

Step 2: Load into Firefox (Temporary) Firefox removes unpacked extensions when you restart the browser. For immediate testing:

Open Firefox, go to: about:debugging#/runtime/this-firefox
Click "Load Temporary Add-on..."
Navigate to the dist/firefox folder you just built.
Select the manifest.json file inside it and click Open.
✅ MetaPrompt is ready to use!

Step 3: Make it Permanent To keep the extension permanently in Firefox without it disappearing on restart:

You must sign the extension using Mozilla's web-ext tool.
Install the tool globally: npm install -g web-ext
Generate an API key from Mozilla Add-ons (AMO)

Run the sign command inside the dist/firefox folder:

cd dist/firefox
web-ext sign --api-key=YOUR_API_KEY --api-secret=YOUR_API_SECRET

This generates a .xpi file in the web-ext-artifacts folder. You can drag and drop this .xpi file into Firefox for a permanent installation!

🔄 Applying Changes (Reloading)

If you modify the source code, change settings that seem stuck, or the extension hits a rate limit glitch, you must do two things:

Reload the Extension: Go to your browser's extensions page (chrome://extensions or about:debugging) and click the Reload 🔄 icon next to MetaPrompt.
Refresh the AI Chat Page (CRITICAL): Go to your open ChatGPT / Claude / Gemini tab and press Ctrl + R (or Cmd + R on Mac).
- Why? Browsers do not automatically update the code injected into already-open web pages. If you don't refresh the page, you will still be using the old, pre-reload version of the extension on that specific tab!

🔔 Staying Updated

Since this is an open-source project, new AI models, prompt techniques, and bug fixes are pushed to the GitHub repository directly.

How to get notified of updates:

Go to the MetaPrompt GitHub Repository page.
In the top right corner, click the Watch 👁️ button.
Select "Custom" -> check "Releases" (to only get notified of major updates) OR select "All Activity" to get an email whenever code changes.
When an update drops, simply run git pull (or download the ZIP again) and reload the extension in your browser!

🧭 Safari (macOS)

Requires: macOS 12+, Xcode 14+, Safari 15+

Safari requires converting the extension to a native Mac app. This is a one-time process.

Prerequisites: Install Xcode from the App Store.

# Convert the extension to a Safari extension project
xcrun safari-web-extension-converter /path/to/exts-main --project-location ~/Desktop

# Example with actual path:
xcrun safari-web-extension-converter ~/Downloads/exts-main --project-location ~/Desktop

Then:

Xcode opens automatically with the generated project
Click the ▶ Run button in Xcode (builds the app)
Open Safari → Safari menu → Settings → Extensions
Enable MetaPrompt in the list ✅
If you don't see it: Safari menu → Develop → Allow Unsigned Extensions (requires Developer menu enabled in Settings → Advanced)

🔑 Setup: Gemini API Key (Optional)

MetaPrompt works without an API key using built-in local templates. For AI-powered optimization:

Get a Free API Key

Go to Google AI Studio
Sign in with your Google account
Click "Create API Key" → select any project (or create new)
Copy the key (starts with AIza...)

Add Key to MetaPrompt

Click the ✨ MetaPrompt icon in your browser toolbar
Paste your key into the "Gemini API Key" field
Click 💾 Save Key
Optionally click 🔍 Check My Available Models to see which models work on your key

🔒 Privacy: Your key is stored in chrome.storage.sync (local browser storage). It is sent only to generativelanguage.googleapis.com (Google's official Gemini API). Never to any third-party server.

💡 Free tier limits: The free Gemini API tier has generous limits — typically 15 requests/minute and 1 million tokens/day with gemini-1.5-flash. MetaPrompt uses ~200-500 tokens per optimization.

💡 How to Use

Method 1 — Floating Button (Recommended)

Open any supported AI chat site (ChatGPT, Gemini, Claude) or a Custom External Site.
Type your raw idea or message in the input box.
Click the ✨ purple floating button at the bottom-right corner of the page.
Your text is enhanced instantly!

Method 2 — Highlight & Enhance (For DMs & Social Media)

If you are on a complex site like LinkedIn DMs or Instagram, use this much more reliable method:

Type your message or instruction (e.g., "tell her I love her eyes").
Highlight/Select the text with your mouse.
Click the ✨ floating button.
MetaPrompt will optimize only the selected part and replace it exactly where it was.

Method 3 — Keyboard Shortcut

While the cursor is in any input box:

Press Alt + Shift + P
The text is optimized instantly.

➕ Adding Custom External Sites

MetaPrompt isn't just for AI platforms. You can use it as a professional ghostwriter on LinkedIn, Instagram, Reddit, Gmail, or ANY other website.

How to add a new site:

Click the ✨ icon in your browser toolbar to open the MetaPrompt Popup.
Find the "External Sites (Optional)" section.
Type the domain name (e.g., linkedin.com or instagram.com).
Click ➕ Add Site.
Refresh the website you just added — the ✨ floating button will now appear there!

Why use this?

LinkedIn: Turn a rough draft into a high-level professional message.
Instagram/Social: Craft charming, flirty, or heartfelt responses.
Support/Emails: Polish your tone to be perfectly professional or casual.
Forums: Structure long posts or technical questions automatically.

Method 4 — Extension Popup (Quick Settings)

Click the ✨ icon in the toolbar to open the popup
Change your Technique or Model Tier
Return to the chat and click the floating button

🧠 Prompt Techniques Guide

Open the popup and choose how you want your prompt structured:

Technique	Icon	Best For	Example Output Style
Auto	🤖	Everything — safest default	Gemini picks the best technique per prompt
Expert Persona	🎭	Domain-specific tasks	"Act as a Senior Python Game Developer..."
Chain of Thought	🧠	Math, logic, debugging	"Think through this step by step. Stage 1: ..."
Few-Shot	📋	Creative, format-specific output	"Example 1: [input] → [output]... Now: ..."
Zero-Shot	⚡	Simple, fast, direct tasks	"Role: ... Task: ... Constraints: ... Format: ..."
ReAct	🔄	Research, planning, investigation	"Thought: ... Action: ... Observation: ..."

Which to use?

🤖 Auto — when unsure, this is always a great pick
🧠 Chain of Thought — "why does this code fail", "solve this math problem"
📋 Few-Shot — "write an email in this style", "generate data in this format"
🎭 Expert Persona — "build a REST API", "write a blog post about AI"
⚡ Zero-Shot — "summarize this", "translate to French"
🔄 ReAct — "research and plan a marketing strategy", "debug this step by step"

⚙️ Model Tier Guide

Tier	Icon	Models Used	Cost	Quality	Best For
Auto	🔄	Best → stable, in order	Varies	⭐⭐⭐⭐⭐	Recommended default
Flash	⚡	gemini-2.0-flash → 1.5-flash	Lowest	⭐⭐⭐⭐	Everyday use, fast responses
Pro	🧠	gemini-1.5-pro → 1.5-pro-latest	Medium	⭐⭐⭐⭐⭐	Complex prompts, better reasoning
Latest	🚀	gemini-2.5-flash-preview → 2.5-pro	Highest	⭐⭐⭐⭐⭐+	Cutting-edge, may not exist on free tier

Fallback behavior: Each tier has an ordered list of models. If the first model is unavailable (deprecated, not on your plan, etc.), MetaPrompt silently tries the next one. You always get a result.

Check Your Available Models

Add your API key → click 🔍 Check My Available Models in the popup
Each model shows ✅ Available or ❌ Not found
Results are cached for 24 hours (click the button again to force a recheck)
Use this to decide which tier makes sense for your API key

🔄 Reloading After Code Changes (For Developers)

When you edit any extension file, you need to reload the extension AND the target page.

Chrome / Edge / Brave / Opera

After changing `background.js` or `manifest.json`:

chrome://extensions → find MetaPrompt → click the 🔄 Reload icon

Then refresh the AI chat page (Ctrl+R / Cmd+R).

After changing `content/.js` or `utils/.js`:

Just refresh the AI chat page (Ctrl+R). Content scripts reload with the page.

After changing `ui/panel.html`, `panel.css`, or `panel.js`:

Just close and reopen the popup — no reload needed.

After changing `manifest.json` (permissions, scripts):

Must reload extension AND refresh all open tabs.

Quick Reload Shortcut

You can also add a bookmark with this JavaScript to reload instantly:

javascript:chrome.runtime.reload()

Or use the Extensions Reloader Chrome extension for one-click reloads.

Firefox (web-ext)

If you started Firefox with web-ext run, it auto-reloads on every file save. No manual action needed.

If you installed manually:

Go to about:debugging#/runtime/this-firefox
Find MetaPrompt → click Reload
Refresh the AI chat tab

Safari (macOS)

Make your code changes
In Xcode → click ▶ Run again to rebuild
Safari will prompt to reload the extension

🧪 Testing Guide

✅ Quick Sanity Check (No API Key Required)

Install the extension in your browser
Open chatgpt.com (or claude.ai, gemini.google.com)
Look for the ✨ purple button in the bottom-right corner — if it's there, the extension loaded ✅
Type "explain recursion" in the chat box
Click the ✨ button
The text should transform into a structured, expert-level prompt ✅

🧠 Testing Techniques

For each technique below, type the same prompt and switch techniques between tests:

Test prompt: "write a function to find prime numbers"

Technique	What to check in output
Auto	Smart structure — likely Expert Persona or Zero-Shot
Expert Persona	Starts with "Act as a Senior Software Engineer..."
Chain of Thought	Contains numbered reasoning steps
Few-Shot	Contains 2-3 examples before the actual ask
Zero-Shot	Contains labeled sections: Role, Task, Constraints, Format
ReAct	Contains Thought → Action → Observation structure

How to test:

Open popup → select technique
Go to chat site → type the test prompt → click ✨
Check the output matches the expected structure
Repeat for each technique

⚙️ Testing Model Selection (API Key Required)

Add your Gemini API key in the popup
Click the 🔄 Refresh button next to the Model Dropdown
The dropdown will populate with the exact models your API key has access to (e.g., gemini-2.0-flash).

Verify fallback works:

Select Auto (Best available) in the dropdown.
Optimize any prompt
Open the browser console (F12 → Console tab)
Look for: [MetaPrompt] Trying: gemini-3.1-flash-lite-preview
If that model fails: [MetaPrompt] gemini-3.1-flash-lite-preview failed (404) — trying next
Then it tries the next model automatically ✅

🔍 Testing Model Caching

Open the popup → click the 🔄 Refresh button
The button will spin while it fetches models from the Gemini API.
Close the popup and reopen it → the models load instantly from local storage.
The cache automatically expires after 24 hours to ensure your model list is always up to date.

🛑 Testing Without API Key (Offline Fallback)

Clear any saved API key (popup → 🗑️ Clear)
Optimize any prompt → should still work using local templates
The result won't use the selected Technique or Model (those require the API)
A structured, useful prompt is still produced ✅

📋 Full Test Checklist

SETUP
[ ] Extension loads without errors in chrome://extensions
[ ] ✨ floating button appears on chatgpt.com
[ ] ✨ floating button appears on claude.ai
[ ] ✨ floating button appears on gemini.google.com
[ ] Popup opens when clicking the toolbar icon
[ ] Popup shows Technique Picker (6 chips)
[ ] Popup shows Model Tier Picker (4 cards)

WITHOUT API KEY
[ ] Prompt optimization works (uses local templates)
[ ] Alt+Shift+P keyboard shortcut works
[ ] No error messages appear

WITH API KEY
[ ] API key saves and persists after popup close
[ ] API key clears with Clear button
[ ] "Check My Models" button works and shows results
[ ] Model discovery results persist after popup close/reopen
[ ] Flash tier produces a result
[ ] Pro tier produces a result
[ ] Latest tier falls back gracefully if model unavailable
[ ] Auto technique works
[ ] Expert Persona technique produces "Act as..." style output
[ ] Chain of Thought produces step-by-step structure
[ ] Few-Shot produces examples before the task
[ ] Zero-Shot produces Role/Task/Constraints/Format sections
[ ] ReAct produces Thought/Action/Observation structure

PERSISTENCE
[ ] Selected technique is remembered after popup close
[ ] Selected model tier is remembered after popup close
[ ] Stats (enhancement count) increase after each use

📁 Project Structure

exts/                                ← Root repo (clone this)
│
├── README.md                        ← This file
├── .gitignore
│
├── manifest.json                    ← Extension config (MV3)
    ├── background.js                ← Service worker: Gemini API + keep-alive
    ├── LICENSE
    ├── README.md
    │
    ├── icons/
    │   ├── icon16.png
    │   ├── icon48.png
    │   └── icon128.png
    │
    ├── content/
    │   ├── content.js               ← Page init + keyboard shortcut
    │   ├── observer.js              ← Floating button injection + notifications
    │   └── optimizer.js             ← Local templates + API call dispatcher
    │
    ├── utils/
    │   └── detectPlatform.js        ← ChatGPT/Gemini/Claude detection & text I/O
    │
    └── ui/
        ├── panel.html               ← Extension popup UI
        ├── panel.css                ← Popup styles
        └── panel.js                 ← Popup logic (API key, technique, model, discovery)

Storage keys used (all local, never transmitted):

Key	Type	Purpose
`geminiApiKey`	`sync`	Your Gemini API key
`selectedTechnique`	`sync`	Chosen prompt technique
`selectedModel`	`sync`	Chosen model tier
`enhanceCount`	`local`	Total enhancements counter
`lastUsed`	`local`	Timestamp of last use
`modelAvailability`	`local`	Cached model discovery results
`modelCheckTimestamp`	`local`	When the last check was run

❓ Troubleshooting

The ✨ button doesn't appear

Make sure you're on a supported URL: chatgpt.com, claude.ai, or gemini.google.com
Reload the page (Ctrl+R)
Check chrome://extensions — ensure MetaPrompt is enabled (toggle is blue)
If the button still doesn't appear: disable/enable the extension, then reload the page

"Optimization failed" or API error

Verify your API key is correct (check AI Studio)
Your key may have hit the free tier rate limit — wait 1 minute and try again
The extension falls back to local templates automatically — you still get a result

The text doesn't appear in the chat box after optimization

Click inside the chat input box first, then click ✨
Some pages re-render the input — the text may appear briefly then disappear; type a character in the box to stabilize it

"Receiving end does not exist" error

The service worker went to sleep between your browser session and the next day
Just click ✨ a second time — it reconnects automatically
This should be rare since we use dual keep-alive (alarm heartbeat + port connection)

Firefox: Extension disappears after browser restart

This is normal if you used about:debugging (temporary install).
To keep it permanently, you must use Firefox Developer Edition, build an .xpi via web-ext, and disable extension signatures in about:config. For most users, reloading via about:debugging is recommended.

Safari: Extension not showing in Safari settings

Make sure you clicked Develop → Allow Unsigned Extensions in Safari's menu bar
If Develop menu is missing: Safari → Settings → Advanced → check "Show Develop menu"

Specific model doesn't work

If you manually selected a model like gemini-2.5-pro and it fails, it means your API key lacks access or quota for that specific model.
Switch the dropdown to Auto (Best available) — it will automatically hunt for a working model.

🤝 Contributing

Pull requests are welcome! Areas especially needing help:

🌍 Adding support for new AI platforms (Perplexity, Mistral, Grok, etc.)
🧠 Improving prompt templates for specific domains
🌐 Internationalization (non-English prompts)
🧪 Writing automated tests with web-ext

To contribute:

Fork the repo
Create a branch: git checkout -b feature/your-feature-name
Make your changes in the codebase
Test in at least Chrome and Firefox
Submit a pull request

📄 License

MIT License — see LICENSE for details.

Free to use, modify, fork, and distribute. No attribution required (but appreciated!).

Built with ❤️ for the AI community. Star ⭐ the repo if MetaPrompt helps you!

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
content		content
icons		icons
ui		ui
utils		utils
web-ext-artifacts		web-ext-artifacts
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
background.js		background.js
build.js		build.js
extension.zip		extension.zip
manifest.json		manifest.json
package-lock.json		package-lock.json
package.json		package.json
test.html		test.html

Folders and files

Latest commit

History

Repository files navigation