prompt-injection

Detect prompt injection vulnerabilities in LLM-integrated applications and classify findings by OWASP LLM01.

44|128|Updated Mar 6, 2026
One-click install
npx skills add https://github.com/UnitOneAI/SecuritySkills --skill prompt-injection-unitoneai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: prompt-injection
Source: https://github.com/UnitOneAI/SecuritySkills/tree/main/skills/ai-security/prompt-injection
Command: npx skills add https://github.com/UnitOneAI/SecuritySkills --skill prompt-injection-unitoneai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you detect prompt injection vulnerabilities that can cause LLM applications to follow attacker-controlled instructions instead of your intended behavior.

Core Features & Use Cases

  • Direct injection testing: Reviews how user-supplied text may manipulate prompts, including boundary and concatenation weaknesses.
  • Indirect injection testing: Assesses RAG and external-content ingestion paths where poisoned documents or web content can plant malicious instructions.
  • Structured assessment reporting: Produces a categorized, severity-based report with defensive recommendations aligned to OWASP LLM01:2025 and MITRE ATLAS AML.T0051.
  • Defense posture evaluation: Checks key mitigations like instruction hierarchy, output filtering, privilege separation, and continuous red-teaming.

Quick Start

Run the assessment against your app code or a specific module by invoking prompt-injection on the target path you want reviewed.

Frequently Asked Questions about prompt-injection

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

FAQPage Schema
How do I detect prompt injection vulnerabilities in my LLM application?

Test prompt injection vulnerabilities by mapping interaction surfaces in your LLM app code and invoking an assessment to see how attacker-controlled content influences model behavior across direct input and indirect RAG pipelines.

What is indirect prompt injection in RAG pipelines?

Indirect prompt injection in RAG pipelines happens when poisoned documents or external web content plant malicious instructions into retrieved context, manipulating the LLM's behavior during ingestion and generation processes.

How do I secure my LLM app against OWASP LLM01 prompt injection attacks?

Secure your LLM app against OWASP LLM01 attacks by evaluating defense postures including instruction hierarchy, output filtering, privilege separation, and continuous red-teaming to classify and mitigate identified vulnerabilities.

Can I test multi-turn and tool-using LLM contexts for prompt injection?

Yes, you can test multi-turn and tool-using LLM contexts for prompt injection by assessing how attacker-controlled content influences model behavior across complex interaction surfaces and external-content ingestion paths.

Does prompt injection testing classify findings using MITRE ATLAS?

Yes, prompt injection testing classifies findings using MITRE ATLAS AML.T0051 categories and OWASP LLM01:2025 standards to produce a structured, severity-based report with defensive recommendations.

What are the limitations of automated prompt injection assessment?

Automated prompt injection assessment requires mapping interaction surfaces and may not capture all multi-turn manipulation vectors, making continuous red-teaming and defense posture evaluation necessary for comprehensive security.