PromptShield

PromptShield is a resilient browser prompt enhancer for AI apps like ChatGPT, Claude, and Gemini.

It improves weak prompts and demonstrates resilience when the primary LLM or tool layer fails. The MVP uses a Chrome extension frontend, FastAPI backend, and a TrueFoundry AI Gateway-compatible OpenAI client.

What it does

• Floats over ChatGPT / Claude / Gemini
• Reads the user's draft prompt
• Enhances it based on mode: general, resume, coding, research, academic
• Calls a backend connected to TrueFoundry AI Gateway
• Falls back to a local rule-based prompt enhancer if the gateway/model fails
• Includes chaos mode for hackathon demo: simulate LLM failure
• Logs gateway success/failure and fallback behavior

Architecture

Browser AI App
   ↓
PromptShield Chrome Extension
   ↓ localhost:8000
FastAPI Backend
   ↓
TrueFoundry AI Gateway Virtual Model
   ↓
Primary LLM → fallback model(s)

If gateway/model fails:
FastAPI → local fallback enhancer → user still gets a useful prompt

Repo structure

promptshield/
  backend/
    main.py
    gateway_client.py
    resilience.py
    prompt_templates.py
    requirements.txt
    .env.example
  extension/
    manifest.json
    content.js
    styles.css
  README.md

Prerequisites

• Python 3.10+
• Google Chrome or Chromium-based browser
• TrueFoundry AI Gateway endpoint and API key

You can still run the app without TrueFoundry credentials to test local fallback behavior.

1. Run the backend

cd backend
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env

Edit .env:

TRUEFOUNDRY_API_KEY=your_truefoundry_gateway_key
TRUEFOUNDRY_GATEWAY_BASE_URL=https://your-gateway-url/v1
TRUEFOUNDRY_VIRTUAL_MODEL=promptshield-router

Start the server:

uvicorn main:app --reload --host 0.0.0.0 --port 8000

Open:

http://localhost:8000/docs

2. Test the backend directly

curl -X POST http://localhost:8000/enhance \
  -H "Content-Type: application/json" \
  -d '{"prompt":"make my resume better","mode":"resume"}'

Simulate LLM failure:

curl -X POST "http://localhost:8000/chaos/toggle?kind=llm_failure&enabled=true"

Turn it off:

curl -X POST "http://localhost:8000/chaos/toggle?kind=llm_failure&enabled=false"

View logs:

curl http://localhost:8000/logs

3. Load the Chrome extension

1. Open Chrome.
2. Go to chrome://extensions/.
3. Turn on Developer mode.
4. Click Load unpacked.
5. Select the extension/ folder.
6. Open https://chatgpt.com/.
7. You should see the PromptShield floating widget.

4. Demo script

1. Open ChatGPT.
2. Type: make my resume better.
3. Click Enhance Prompt.
4. Show the improved prompt.
5. Click Simulate LLM Failure.
6. Type another weak prompt.
7. Click Enhance Prompt again.
8. Show that PromptShield still returns a useful prompt using local fallback.
9. Open http://localhost:8000/logs to show resilience logs.

Buold for DEV AI/ML Hackathon

PromptShield: A resilient prompt copilot that improves user input and survives LLM/tool failures without breaking the user experience.

Failure scenarios shown:

• Primary LLM failure
• Gateway/client exception
• Slow response simulation
• Local fallback response
• User-facing recovery status
• Observability logs

Future improvements

• Add MCP server failure simulation
• Add provider-level fallback visualization
• Add prompt quality scoring
• Add user memory/preferences
• Add team prompt library
• Add SDK for agent builders