Languages: Python, JavaScript/TypeScript, Java, Go, Ruby, PHP, C/C++, C#, Scala, Kotlin, Rust, HCL (Terraform), YAML (Kubernetes)

llm-security

Security guidelines for LLM applications based on the OWASP Top 10 for Large Language Model Applications 2025.

Use when:

• Building LLM-powered applications
• Implementing RAG systems
• Securing AI/ML pipelines
• Reviewing code that interacts with language models

Categories covered:

| Impact | Category | Description | |--------|----------|-------------| | Critical | Prompt Injection | Input validation, content segregation, output filtering | | Critical | Sensitive Information Disclosure | PII detection, permission-aware RAG | | Critical | Supply Chain | Model verification, safetensors, ML-BOM | | Critical | Data and Model Poisoning | Training data validation, anomaly detection | | Critical | Improper Output Handling | Context-aware encoding, parameterized queries | | High | Excessive Agency | Least privilege, human-in-the-loop | | High | System Prompt Leakage | External guardrails, no secrets in prompts | | High | Vector and Embedding Weaknesses | Permission-aware retrieval, tenant isolation | | High | Misinformation | RAG, fact verification, confidence scoring | | High | Unbounded Consumption | Rate limiting, budget controls |

Frameworks: OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF

semgrep

Run Semgrep static analysis scans and create custom detection rules for security vulnerabilities and bug patterns.

Use when:

• Running security scans with Semgrep
• Creating custom Semgrep rules
• Detecting specific vulnerability patterns
• Setting up Semgrep in CI/CD pipelines

Capabilities:

| Feature | Description | |---------|-------------| | Quick Scans | Run semgrep --config auto or use curated rulesets | | Rulesets | security-audit, owasp-top-ten, cwe-top-25, trailofbits | | Custom Rules | Pattern matching and taint mode for data flow analysis | | Test-Driven | Write test cases first with ruleid: and ok: annotations | | CI/CD | GitHub Actions integration with diff-aware scanning |

Rule Creation Workflow:

1. Analyze the vulnerability pattern
2. Create test cases first (test-driven development)
3. Analyze AST structure with semgrep --dump-ast
4. Write the rule (taint mode for injection, pattern matching for syntax)
5. Iterate until 100% tests pass
6. Optimize patterns

When to use taint mode: SQL injection, command injection, XSS, path traversal, SSRF - any vulnerability where untrusted data flows to a dangerous sink.

Usage

Skills are automatically available once installed. The agent will use them when relevant tasks are detected.

Examples:

Review this React component for security issues

Help me implement input validation for my LLM chat endpoint

Create a Semgrep rule to detect hardcoded API keys in Python

Development

Building Skills

make install     # Install dependencies
make validate    # Validate all skills
make build       # Build AGENTS.md for all skills
make zip         # Create distribution packages
make             # All of the above

Single Skill Operations

make validate-skill SKILL=code-security
make build-skill SKILL=llm-security

Skill Structure

Each skill contains:

• SKILL.md - Instructions for the agent
• rules/ - Individual rule files (for skills with rules)
• scripts/ - Helper scripts for automation (optional)
• references/ - Supporting documentation (optional)

Acknowledgments

Originally created by @DrewDennison at Semgrep. This work was heavily inspired by Vercel's React Best Practices.