agent-security-audit

Enumerate permissions, data flows, and guardrails to produce a structured security audit report.

141|14|Updated Mar 22, 2026
One-click install
npx skills add https://github.com/OWASP/secure-agent-playbook --skill agent-security-audit-owasp
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agent-security-audit
Source: https://github.com/OWASP/secure-agent-playbook/tree/main/skills/agent-security-audit
Command: npx skills add https://github.com/OWASP/secure-agent-playbook --skill agent-security-audit-owasp

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps security teams and AI developers identify and remediate security risks in AI agent configurations, including excessive permissions, prompt-injection surfaces, data exfiltration paths, and absent guardrails.

Core Features & Use Cases

  • Permission Inventory: enumerate every tool, MCP server, file system path, network access, and credential the agent possesses, flagging capabilities beyond its stated purpose.
  • Prompt Injection Surface Analysis: assess input paths (user messages, tool outputs, MCP resources, RAG documents) for potential instruction hijacking or data leakage.
  • Data Exfiltration Path Analysis: map how sensitive data could leave the agent boundary through tools, logs, or cross-server forwarding.
  • Guardrail Evaluation: verify system prompts, tool-call confirmations, output filtering, audit logging, and sandboxing exist and are effective.

Quick Start

Run a full security audit of the agent setup following the defined procedure and generate a structured findings report.

Frequently Asked Questions about agent-security-audit

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

FAQPage Schema
How do I audit AI agent security configurations for prompt injection risks?

To audit AI agent security, you enumerate permissions and assess input paths like user messages and tool outputs for instruction hijacking. This process maps the injection surface and identifies data leakage vulnerabilities across your agent deployment.

What is an AI agent data exfiltration path analysis?

Data exfiltration path analysis maps how sensitive data leaves the agent boundary through tools, logs, or cross-server forwarding. It identifies unauthorized data transmission routes within your MCP configs and orchestration code to prevent information leakage.

How do I inventory permissions and capabilities for an AI agent?

Inventorying agent permissions involves enumerating every tool, MCP server, file system path, network access, and credential the agent possesses. This flags any capabilities beyond the agent's stated purpose to ensure least privilege.

Can I use a security audit on CLAUDE.md references and MCP configs?

Yes, a security audit can be applied directly to CLAUDE.md references, MCP configs, and orchestration code. It evaluates typical agent deployments to produce a structured output including a Permission Summary and Risk Findings report.

How do I evaluate if AI agent guardrails and system prompts are effective?

Evaluating guardrails involves verifying that system prompts, tool-call confirmations, output filtering, audit logging, and sandboxing exist within your agent setup. It checks their effectiveness against potential security bypasses and data exfiltration attempts.

What are the limitations of agent security audits for MCP environments?

Agent security audits rely on analyzing static MCP configs, CLAUDE.md references, and orchestration code. They may not detect dynamic runtime vulnerabilities or zero-day exploits that emerge during active agent execution without continuous monitoring.