git:20260120.c6005a7 to git:20260126.11a34be

---

name: llm-attacks-security

description: "Guide for LLM security attacks: prompt injection, jailbreaking, data extraction, and where to place resources in README.md."

---

# LLM Security Attacks

## Scope

Use this skill when working on:

- Prompt injection attacks and defenses

- LLM jailbreaking techniques

- Training data extraction

- Model output manipulation

- AI safety bypasses

## Common LLM Vulnerabilities (Cheat Sheet)

### Prompt Injection

- Direct injection (user prompt manipulation)

- Indirect injection (via external data sources)

- System prompt extraction

- Role-play attacks

- Encoding/obfuscation bypasses

### Jailbreaking

- DAN (Do Anything Now) prompts

- Character roleplay escapes

- Multi-turn manipulation

- Token smuggling

- Crescendo attacks

### Data Extraction

- Training data memorization extraction

- PII leakage from context

- System prompt disclosure

- API key/secret extraction

- Model architecture probing

### Model Manipulation

- Output steering

- Hallucination exploitation

- Bias amplification

- Harmful content generation

## OWASP LLM Top 10 Reference

1. LLM01 - Prompt Injection

2. LLM02 - Insecure Output Handling

3. LLM03 - Training Data Poisoning

4. LLM04 - Model Denial of Service

5. LLM05 - Supply Chain Vulnerabilities

6. LLM06 - Sensitive Information Disclosure

7. LLM07 - Insecure Plugin Design

8. LLM08 - Excessive Agency

9. LLM09 - Overreliance

10. LLM10 - Model Theft

## Where to Add Links in README

- Prompt injection tools/research: `AI Security & Attacks → Prompt Injection`

- Jailbreak techniques: `AI Security & Attacks → Model Security`

- Data extraction research: `AI Security & Attacks → Privacy & Extraction`

- Defense tools: `AI Security & Attacks → Model Security`

- CTFs/challenges: `AI Security Starter Pack → CTFs / Practice`

## Notes

Keep additions:

- AI/LLM security focused

- Non-duplicated URLs

- Minimal structural changes