skills.rest
wgpsec
wgpsecOFFICIAL
Explore all skills by this author →
Website𝕏@wgpsecGitHub

ai-identity-security

Tests AI agent identity, role boundaries, permissions, sessions, and credential controls for bypass.

1.7k|238|Updated Dec 7, 2019
Search Tags:#penetration testing#ai security#credential leakage#session hijacking#mcp security#agent impersonation#role escape
One-click install
npx skills add https://github.com/wgpsec/AboutSecurity --skill ai-identity-security
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ai-identity-security
Source: https://github.com/wgpsec/AboutSecurity/tree/main/skills/ai-security/ai-identity-security
Command: npx skills add https://github.com/wgpsec/AboutSecurity --skill ai-identity-security

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

AI agents with tool-calling and external system access can exceed their intended business roles and permissions, turning a chatbot flaw into unauthorized actions. This Skill provides a structured methodology to verify whether role constraints, permission assignments, session isolation, and credential management in AI systems can actually be bypassed.

Core Features & Use Cases

  • Role Escape Testing: Covers assumed-scenario, assumed-role, forgetting-method, and goal-hijacking techniques (GAARM.0052) to break application-layer role settings.
  • Permission & Identity Attacks: Tests Action parameter tampering, MCP unauthorized resource access, multi-agent impersonation, and trust-chain deception (GAARM.0053/0057/0058/0059).
  • Session & Credential Auditing: Checks session ID predictability, CDN cache token leakage, cross-tenant isolation, and API key exposure via prompt injection.
  • Use Case: When assessing a customer-service agent with tool access, follow the phased workflow to attempt role escape, tamper with Action parameters to trigger admin operations, and verify whether session tokens can be stolen or replayed.

Quick Start

Use the ai-identity-security skill to run a full identity and permission assessment against my AI agent, covering role escape, Action privilege escalation, and credential leakage checks.

Frequently Asked Questions about ai-identity-security

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

FAQPage Schema
How do I test an AI agent for role escape vulnerabilities?▼

Role escape testing uses four techniques: assumed-scenario escape (fake audit modes), assumed-role escape (DAN-style personas), forgetting-method escape (reset prior instructions), and goal hijacking. Each attempts to make the model abandon its application-layer role constraints.

What is the difference between prompt jailbreaking and role escape?▼

Jailbreaking attacks the model's safety alignment to generate harmful content, while role escape attacks the application's business role settings, such as making a customer-service bot perform admin operations. The two can be combined but have different test objectives.

How to test MCP servers for unauthorized resource access?▼

Enumerate connected MCP servers and their tool permission levels, then attempt calling higher-privilege tools without authorization. Also verify server registration authentication, whether prompt injection can redirect the agent to attacker-controlled servers, and if MCP traffic is encrypted.

Can multi-agent systems be attacked through identity impersonation?▼

Yes. If inter-agent messages lack source verification, attackers can forge messages from high-privilege agents or exploit trust chains by compromising a low-privilege agent whose output is trusted by more privileged agents downstream.

What are the limitations of this AI identity testing methodology?▼

It focuses on application-layer identity and permission boundaries, not model-level safety alignment or infrastructure security. It assumes authorized testing access and does not cover areas like model extraction, training data poisoning, or prompt leaking, which belong to separate skill modules.

Related Skills

Explore More Software Engineering Skills→
lushly-devlushly-dev
4

audit-security

Audit web applications, APIs, MCP servers, and AI agents for OWASP/CWE vulnerabilities.

Official
Advanced
dapperdiversdapperdivers
1

report-generator

Generate structured markdown penetration test reports from JSON security scan findings.

Community
Intermediate
zaiyemeiyou404zaiyemeiyou404

ctf-web-encoding-bypass

Bypass web application defenses and extract data from CTF challenges.

Community
Advanced
ikenibornikeniborn
6

Authentication & Security

Implement JWT-based authentication, Telegram OAuth, and security middleware for REST APIs.

Community
Advanced