security-review-owasp-llm-prompt-injection-prevention

Review LLM prompt injection risks and trust boundaries across execution paths.

Updated Mar 26, 2026
One-click install
npx skills add https://github.com/sjinks/ai-owasp-skillset --skill security-review-owasp-llm-prompt-injection-prevention
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: security-review-owasp-llm-prompt-injection-prevention
Source: https://github.com/sjinks/ai-owasp-skillset/tree/main/.github/skills/security-review-owasp-llm-prompt-injection-prevention
Command: npx skills add https://github.com/sjinks/ai-owasp-skillset --skill security-review-owasp-llm-prompt-injection-prevention

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill helps reviewers find prompt injection, unsafe tool use, context leakage, and weak guardrails in LLM systems before attackers can steer model behavior.

Core Features & Use Cases

  • Reviews prompts, retrieval paths, tools, memory, and output consumers for trust-boundary issues.
  • Checks least-privilege execution, human approval gates, logging, monitoring, and abuse containment.
  • Use it for agents, copilots, plugins, RAG pipelines, and any workflow where model output can trigger actions.

Quick Start

Use the security-review-owasp-llm-prompt-injection-prevention skill to review the attached LLM workflow for prompt injection and tool abuse risks.

Frequently Asked Questions about security-review-owasp-llm-prompt-injection-prevention

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review LLM workflows for prompt injection and tool abuse risks?

Review LLM workflows for prompt injection and tool abuse risks by validating untrusted content handling, tool constraints, context isolation, output validation, and monitoring across agent, retrieval, and plugin execution paths.

What is context isolation and how does it prevent LLM agent security issues?

Context isolation prevents LLM agent security issues by separating untrusted content from trusted system prompts, ensuring injected instructions cannot override guardrails or trigger unauthorized tool execution across trust boundaries.

How do I secure RAG pipelines and copilots against context leakage?

Secure RAG pipelines and copilots against context leakage by applying evidence-based validation of retrieval paths, output consumers, least-privilege execution, human approval gates, and abuse containment monitoring.

Can I use this prompt injection prevention review for plugins and copilots?

Yes, you can use this review for plugins, copilots, agents, and RAG pipelines, checking trust-boundary issues, unsafe tool use, memory handling, and weak guardrails in any workflow where model output triggers actions.

What are the limitations of automated LLM security reviews for agent abuse?

Automated LLM security reviews for agent abuse require evidence-based validation of untrusted content handling and tool constraints, but cannot fully replace human oversight of operational controls, logging, and human approval gates.

When do I need to check trust boundaries in LLM systems?

Check trust boundaries in LLM systems whenever model output can trigger actions, requiring validation of untrusted content handling, tool constraints, context isolation, output validation, and monitoring for abuse in code and architecture.