A curated cheat sheet for learning, practicing, and understanding AI prompt injection techniques within ethical security research and CTF environments.
ATTENTION
⚠️ This project is intended strictly for educational use, CTF competitions, and defensive AI security research.
- Understand how prompt injection works in LLMs
- Learn common CTF attack patterns and reasoning strategies
- Study model behavior under conflicting instructions
- Improve defensive awareness for AI application builders
- Provide a quick-reference guide during competitions
- Instruction hierarchy (system vs user vs developer)
- Context manipulation & indirect extraction
- Model compliance & refusal patterns
- Hidden system prompts & flag storage locations
- Instruction hierarchy (system vs user prompts)
- Context manipulation
- Model compliance behavior
- Refusal patterns & safety responses
- Role simulation
- Debug / audit prompting
- Instruction summarization
- Transformation-based extraction
- Multi-agent simulation
- Narrative framing
- Encoding & formatting tricks
- Realistic challenge scenarios
- Step-by-step prompt evolution
- Failure → refinement workflows
- Common flag exposure patterns
- How developers can prevent prompt injection
- Secure prompt design principles
- Input sanitization ideas
- Isolation strategies
AI-Prompt-Injection/ │ ├── concepts/ ├── techniques/ ├── examples/ ├── defensive-guides/ ├── templates/ └── README.md
- Debug / Audit Template
- Transformation / Encoding Template
- Multi-Agent Simulation Template
- Narrative / Fictional Framing Template
- CTF players
- AI security learners
- Prompt engineers
- Developers building LLM apps
- Cybersecurity students
- Red/Blue team researchers
This repository does not promote bypassing real-world AI safety systems.
All material is intended for:
- Capture The Flag competitions
- Educational experimentation
- Defensive AI security research
Always follow platform policies and responsible disclosure practices.
Contributions are welcome! You can help by:
- Adding new CTF techniques
- Improving explanations
- Sharing defensive mitigations
- Providing anonymized challenge writeups
This project is licensed under the MIT License.