Setting the file. One moment.
Subchapter 2.2
README.mdMarkdown3 KBView on GitHub
Security guidelines for LLM applications based on the OWASP Top 10 for Large Language Model Applications 2025.
llm-security/
├── SKILL.md # Skill definition (loaded by agents)
├── rules/ # Security rule files
│ ├── _sections.md # Index of all categories
│ ├── prompt-injection.md
│ ├── sensitive-disclosure.md
│ └── ... # 10 rule files total
└── README.md # This fileInstall the skill:
npx skills add semgrep/skillsThe agent will automatically reference these guidelines when building or reviewing LLM applications.
From the repo root:
make validate # Validate all skills
make build # Build all skills
make zip # Create distribution packages
make # All of the aboveOr for this skill only:
cd packages/skill-build
pnpm install
pnpm validate llm-security # Validate rule files
pnpm build-agents llm-security # Build AGENTS.mdrules/{category}.md---
title: Category Title
impact: HIGH
impactDescription: Brief description of the impact
tags: security, llm, category-name, owasp-llmXX
---
## Category Title
Brief explanation of the vulnerability.
**Vulnerable (description):**
```python
# Vulnerable code
```
**Secure (description):**
```python
# Secure code
```rules/_sections.mdmake validate to check formattingmake to rebuild everything| Level | Description |
|---|---|
| CRITICAL | Data exfiltration, model compromise, unauthorized actions |
| HIGH | Information disclosure, service degradation, significant risk |
Created by @DrewDennison (opens in a new tab) at Semgrep (opens in a new tab).
Rules derived from the OWASP Top 10 for LLM Applications 2025 (opens in a new tab).