owasp-llm

owasp-llm

Field	Value
Kind	skill
Source	.github/skills/security/owasp-llm
Invocation	Loaded on demand by referencing agents
Interactive	No

What it does

OWASP Top 10 for LLM Applications (2025) knowledge base for identifying, assessing, and remediating large language model security risks.

When to use it

Have an agent load this reference when reviewing an LLM application that accepts untrusted inputs, retrieves documents, handles model output, or exposes sensitive context. Its OWASP 2025 reference set includes retrieval, output handling, excessive agency, and consumption risks. Use owasp-agentic alongside it when autonomous goals, tools, memory, or delegation introduce additional system-level boundaries.

Example usage

Ask a security reviewing agent: "Load owasp-llm and review this retrieval-based help assistant using synthetic documents, the retrieval access policy, and the HTML-rendering code. Do not call a model or retrieve real user records."

The review should identify relevant references for prompt injection, sensitive information disclosure, vector and embedding weaknesses, and output handling. Expect findings that name the supplied evidence, affected boundary, and proposed mitigation or test. Success means distinguishing demonstrated code or design gaps from hypotheses requiring later evaluation. A read-only review does not establish resistance to attacks or replace privacy assessment.