---

🎯 Overview

NullSec PromptInject is a curated library of prompt injection payloads and an automated tester for LLM-powered applications. It targets system prompt extraction, instruction hijacking, context manipulation, and output steering across chatbots, RAG pipelines, AI agents, and function-calling systems.

⚡ Features

Feature	Description
Payload Library	500+ categorised prompt injection payloads
System Prompt Extraction	Techniques to leak hidden system instructions
Instruction Override	Payloads that hijack model behaviour
Context Manipulation	Indirect injection via RAG document poisoning
Function Call Abuse	Exploit tool-use / function-calling APIs
Multi-Language	Payloads in EN, ZH, JA, DE, FR, ES, AR
Auto-Tester	Batch-test payloads against target endpoints

📋 Payload Categories

Category	Count	Targets
System Prompt Extraction	80+	Chatbots, assistants
Instruction Override	90+	Any LLM app
Jailbreak Chains	60+	Safety-aligned models
Indirect Injection	50+	RAG, email agents
Function Call Abuse	40+	Tool-use agents
Output Steering	45+	Content generators
Encoding Bypass	35+	Input filters
Multi-turn Escalation	30+	Conversation systems

🚀 Quick Start

# Test all payloads against a target endpoint
nullsec-promptinject test --target http://chatbot.example.com/api --category all

# Extract system prompt
nullsec-promptinject extract --target http://chatbot.example.com/api --techniques top20

# Test RAG indirect injection
nullsec-promptinject indirect --target http://rag.example.com/query --inject-doc malicious.txt

# List available payload categories
nullsec-promptinject list --categories

🔗 Related Projects

Project	Description
nullsec-llmred	LLM red-teaming framework
nullsec-adversarial	Adversarial ML attack toolkit
nullsec-modelaudit	ML model security auditing
nullsec-datapoisoning	Training data poisoning detection
nullsec-linux	Security Linux distro (140+ tools)

⚠️ Legal

For authorized security testing only. Never use prompt injection against systems without explicit written permission.

📜 License

MIT License — @bad-antics

---