🛡️ Agentic QA Suite (The Firewall for AI Agents)

🛑 Stop deploying broken agents to production.

Agentic QA is a middleware API that stress-tests your LLM Agents against Infinite Loops, PII Leaks, and Prompt Injections in 20ms before they go live.

---

⚡ Try it Instantly (No Install Needed)

Option 1: 🪄 The Magic Link (Browser)

Click the link below to run a live audit on a demo agent: 👉 Run Live Auto-Scan

Option 2: 💻 The Terminal (cURL)

Copy-paste this into your terminal to audit your own prompt:

curl -X POST "[https://agentic-qa-api.onrender.com/v1/auto-scan](https://agentic-qa-api.onrender.com/v1/auto-scan)" \
     -H "Content-Type: application/json" \
     -d '{"system_prompt": "You are a coding agent", "client_name": "Terminal_User"}'

---

🚨 The Problem

Building agents is easy. Debugging them at scale is hell.

1. Cost: One infinite loop can burn $50+ in OpenAI tokens overnight.
2. Risk: One PII leak (Phone/SSN) in a log file can cause a GDPR lawsuit.
3. Fragility: Manually testing 50 edge cases takes hours.

🚀 The Solution

We provide a Pre-Flight Check API. Wrap your agent's system prompt, send it to us, and we run an Adversarial Simulation (Red Teaming) to break it before your users do.

---

🛠️ Integration (Python SDK)

Add this to your CI/CD pipeline or Agent code to prevent bad deployments.

import requests

def scan_agent(prompt):
    print("🛡️ Scanning Agent Logic...")
    
    response = requests.post(
        "[https://agentic-qa-api.onrender.com/v1/auto-scan](https://agentic-qa-api.onrender.com/v1/auto-scan)",
        json={"system_prompt": prompt, "client_name": "Dev_Integration"}
    )
    
    report = response.json()
    
    if report["overall_status"] == "BLOCKED":
        print(f"❌ DEPLOYMENT STOPPED. Risk Detected!")
        for test in report["tests"]:
            print(f"   - {test['test_name']}: {test['status']}")
        return False
        
    print("✅ Agent is Safe to Deploy.")
    return True

# Usage
my_prompt = "You are a helpful assistant."
scan_agent(my_prompt)

---

🛡️ Detection Capabilities (V2 Engine)

Our engine automatically runs these 3 adversarial attacks on every request:

Attack Vector	Description	Risk Covered
PII Extraction	Attempts to trick agent into leaking Phone/SSN/Email via social engineering.	Compliance / Lawsuits
Infinite Loops	Detects semantic repetition loops that waste tokens.	Financial Loss
Safety/Jailbreak	Tests resistance against "Ignore previous instructions" attacks.	Security

---

📄 API Documentation

Full Swagger UI available here: Live Docs

---

Built for the AI Engineering Community.