arkenSTONE

🧠 Why arkenSTONE?

As AI models evolve, so do the methods to break them. Keeping track of prompt injections, jailbreaks, and zero-day vulnerabilities in spreadsheets is chaotic. arkenSTONE brings order to the chaos. It provides a highly visual, secure, and structured environment to build your personal database of AI vulnerabilities.

Everything is stored locally on your machine. Your zero-days stay yours.

---

⚡ The Arsenal: Core Functionality

arkenSTONE isn't just a database; it's a complete workflow engine for AI security research.

Feature	What it does	Why it matters
🔐 Secure Local Vault	Encrypts your workspace with a PIN lock.	Keeps sensitive exploit payloads away from prying eyes.
📝 Technique Catalog	Logs specific payloads with metadata (Severity, Vector, Category).	Easily search and filter your arsenal during a red team engagement.
🤖 Model Matrix	Maps the status of your payloads against specific LLMs (e.g., GPT-4, Claude).	Instantly know which models are vulnerable to which specific attacks.
📊 Grafana-Style Dashboard	Generates real-time heatmaps, treemaps, and trend charts of your data.	Identify patterns—like seeing that "Role Confusion" is highly effective against a specific model.
💰 Bounty Tracker	Links successful exploits to bug bounty submissions.	Calculate your total earnings and track pending rewards directly in-app.
📄 1-Click Reports	Compiles your findings into clean, professional vulnerability reports.	Simplifies the submission process for AI vendors and bug bounty programs.

---

🎯 Real-World Use Cases

How are professionals using arkenSTONE?

🛑 AI Red Teamers
"I'm hired to stress-test a new corporate chatbot."
Use Case: You open arkenSTONE, filter your vault for Severity: Critical and Category: System Prompt Override, and instantly have a curated list of payloads to test against the client's infrastructure.

🐛 Bug Bounty Hunters
"I found a new token smuggling bypass for a major AI platform."
Use Case: You document the exact payload, mark the target model as Confirmed vulnerable, set the bounty status to Pending, and use the Report Generator to export a beautifully formatted proof-of-concept to submit to HackerOne.

🛡️ AI Defensive Engineers
"I need to track regression testing for our internal models."
Use Case: You track known vulnerabilities. Whenever your team deploys a new guardrail, you run the payloads from your vault and update the model status from Confirmed to Patched.

---

🚀 A Day in the Life (Your Workflow)

1. Discover: You find a new way to trick an LLM using a nested instruction.
2. Catalog: Hit Cmd/Ctrl + N. You name it "Nested Persona Bypass", set the severity to 🔥 High, and paste your prompt.
3. Test: You test it against 5 different models. You mark 3 as Confirmed and 2 as Patched in the Model Matrix.
4. Analyze: You check your Dashboard and notice your "Nested Instruction" success rate has spiked this week.
5. Profit: You link the finding to an active bug bounty program, generate a PDF report, submit it, and mark the bounty as ⏳ Pending.

---

💻 Getting Started

Get up and running in less than 2 minutes.

Prerequisites

• Node.js installed on your local machine.

Quick Start

1. Open your terminal and navigate to the folder:

   cd "e:\New folder (5)"
   

2. Install the required packages:

   npm install
   

3. Ignite the server:

   npm run dev
   

4. Start Hacking: Open your browser to http://localhost:3000 (or 3001 if port 3000 is taken). Set your master PIN, and welcome to your new vault.

---