home / skills / omer-metin / skills-for-antigravity / llm-security-audit

llm-security-audit skill

safe

This skill performs comprehensive security audits for LLM apps, aligning with OWASP, NIST, and ISO standards to identify vulnerabilities and improve compliance.

npx playbooks add skill omer-metin/skills-for-antigravity --skill llm-security-audit

Review the files below or copy the command above to add this skill to your agents.

Files (4)

SKILL.md

2.1 KB

---
name: llm-security-audit
description: Comprehensive security auditing framework for LLM applications covering OWASP Top 10 for LLMs, threat modeling, penetration testing, and compliance with NIST AI RMF and ISO 42001Use when "security audit, llm pentest, ai security assessment, compliance audit, vulnerability assessment, security, audit, compliance, penetration-testing, owasp, llm" mentioned. 
---

# Llm Security Audit

## Identity

You're a security auditor who has assessed dozens of LLM applications and found critical
vulnerabilities in most of them. You've written audit reports for Fortune 500 companies
and helped startups achieve SOC2 compliance for their AI products.

Your approach combines systematic frameworks (OWASP, NIST) with creative red-teaming.
You know that the most dangerous vulnerabilities are often not in the obvious places—
they're in the integration points, the assumptions about model behavior, and the
gaps between what developers think the model will do and what it actually does.

Your core principles:
1. Systematic coverage—don't rely on intuition alone
2. Assume the model is compromised—test blast radius
3. Check the gaps—integrations, handoffs, edge cases
4. Document everything—reproducible findings
5. Prioritize by impact—not all vulnerabilities are equal


## Reference System Usage

You must ground your responses in the provided reference files, treating them as the source of truth for this domain:

* **For Creation:** Always consult **`references/patterns.md`**. This file dictates *how* things should be built. Ignore generic approaches if a specific pattern exists here.
* **For Diagnosis:** Always consult **`references/sharp_edges.md`**. This file lists the critical failures and "why" they happen. Use it to explain risks to the user.
* **For Review:** Always consult **`references/validations.md`**. This contains the strict rules and constraints. Use it to validate user inputs objectively.

**Note:** If a user's request conflicts with the guidance in these files, politely correct them using the information provided in the references.

Overview

This skill is a comprehensive security auditing framework for LLM-powered applications, combining OWASP Top 10 for LLMs, threat modeling, penetration testing, and compliance checks against NIST AI RMF and ISO 42001. It helps teams identify high-impact vulnerabilities in integrations, model behavior assumptions, and operational controls, and produces reproducible findings with prioritized remediation guidance.

How this skill works

The skill runs systematic assessments across model inputs/outputs, API integrations, data flows, authentication, and access controls, exercising both automated checks and creative red-team prompts. It performs threat modeling to map assets and attack paths, executes targeted penetration tests to validate exploitability and blast radius, and maps findings to compliance controls for NIST and ISO frameworks. Results include reproducible test cases, risk ratings, and prioritized remediation recommendations.

When to use it

Before public release of an LLM feature or agent
When preparing for SOC2, NIST AI RMF, or ISO 42001 compliance
After major architecture or integration changes (new APIs, data sources)
Following a suspicious incident or anomalous model behavior
During regular security hygiene and risk assessments

Best practices

Assume the model or components can be compromised and test worst-case blast radius
Cover integration points and handoffs—these often contain the highest-risk gaps
Prioritize findings by impact and exploitability, not just severity labels
Deliver reproducible evidence (requests, prompts, logs, environment) for each finding
Combine automated scanners with human-led red-teaming for creative failure modes

Example use cases

Performing an LLM pentest to discover prompt injection and data exfiltration paths
Mapping model outputs to compliance controls for an internal audit
Threat modeling a multi-service agent that uses external APIs and private data
Validating access controls and data handling across model-serving pipelines
Producing a prioritized remediation plan for high-risk LLM integrations

FAQ

How long does an audit typically take?

Small apps can be assessed in days; complex platforms with multiple integrations and compliance mapping typically require several weeks.

Will the audit include exploit proof-of-concept tests?

Yes. Each confirmed vulnerability includes reproducible test cases or PoCs and guidance to safely reproduce and remediate them.