AI-Assisted Security Script Development

Objective

Use AI to write a PowerShell script that generates a stale account report from Entra ID, then critically review the AI output for security issues, errors, and production readiness before deployment.

Required: Access to Claude. PowerShell environment with Microsoft Graph PowerShell SDK installed (or test in a lab).

Step 1: Generate the script

Role: You are a security automation engineer writing production 
PowerShell scripts.

Context: I need a script that connects to Microsoft Entra ID via 
the Microsoft Graph PowerShell SDK and generates a report of 
user accounts with no sign-in activity in the past 90 days.

Task: Write a PowerShell script that:
1. Connects to Microsoft Graph with appropriate scopes
2. Queries all user accounts with their last sign-in date
3. Filters to accounts with no sign-in in 90+ days
4. Excludes service accounts (UPN matching svc-* or app-*)
5. Excludes break-glass accounts
6. Outputs a CSV report with: UPN, DisplayName, Department, 
   LastSignIn, DaysSinceLastSignIn, AccountEnabled
7. Includes error handling and logging

Constraints:
- Use Microsoft.Graph PowerShell module (not AzureAD module)
- Minimal permissions (User.Read.All, AuditLog.Read.All)
- No hardcoded credentials
- Production-quality error handling

Step 2: Review the AI-generated script

Before running anything, review the script critically:

Security review checklist:

• Does it request minimum necessary Graph permissions?
• Does it handle authentication securely (no hardcoded secrets)?
• Does it handle errors gracefully (try/catch, meaningful messages)?
• Does it validate inputs before using them?
• Could it accidentally modify anything? (Should be read-only)

Functionality review:

• Is the Graph API endpoint correct for last sign-in data?
• Does the 90-day filter calculate correctly?
• Are service accounts and break-glass accounts properly excluded?
• Is the CSV output formatted correctly?

Ask the AI to review its own work:

Role: You are a security code reviewer.
Context: [Paste the script from Step 1]
Task: Review this script for:
1. Security vulnerabilities
2. Logic errors
3. Edge cases not handled
4. Production readiness gaps
5. Microsoft Graph API accuracy

Step 3: Test and refine

If you have a lab environment, run the script. If not, dry-run the logic:

1. Does Connect-MgGraph authenticate successfully?
2. Does the API call return user data with sign-in dates?
3. Does the filter correctly identify stale accounts?
4. Is the CSV output well-formed?

For any issues found, prompt the AI with the specific error message and ask for a fix. Track how many iterations are needed to reach a working script.

Step 4: Document the development process

Record:

This data supports the ROI case for AI-assisted development (Module 8).

Verification checklist

• Script generated with AI assistance
• Security review completed — no hardcoded credentials, minimal permissions
• Functionality review completed — correct API, proper filtering
• AI self-review requested and evaluated
• Script tested (or dry-run reviewed)
• Development process metrics documented