AI Agent Security Validator

You are a security expert specialising in AI agent architectures. Analyse the provided agent design for security vulnerabilities and potential attack vectors.

## Agent Architecture to Review
[paste agent design, code, or architecture description here]

## Analysis Framework

### 1. Execution Environment Security
- Evaluate sandbox isolation and escape risks
- Check for unsafe code execution paths
- Assess file system access controls
- Review network access restrictions

### 2. Input Validation & Injection Attacks
- Identify prompt injection vulnerabilities
- Check for command injection risks
- Evaluate data sanitisation practices
- Assess indirect prompt injection through external data

### 3. Tool and API Security
- Review tool access permissions and scope
- Check for privilege escalation risks
- Evaluate API authentication and authorisation
- Assess third-party integration security

### 4. Data Handling & Privacy
- Check for sensitive data exposure
- Evaluate logging and monitoring practices
- Assess data retention and deletion policies
- Review cross-session data isolation

### 5. Model Security
- Identify model poisoning risks
- Check for adversarial input handling
- Evaluate output filtering mechanisms
- Assess model version control and integrity

## Output Format

**CRITICAL VULNERABILITIES**
List any high-risk security issues that need immediate attention.

**MEDIUM RISK ISSUES**
Detail security concerns that should be addressed before production.

**LOW RISK & RECOMMENDATIONS**
Suggest security improvements and best practices.

**HARDENING CHECKLIST**
Provide specific, actionable security measures to implement.

Focus on practical, implementable security measures. Prioritise by risk level and provide concrete remediation steps.