acad-arx-wizard

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill automates project creation by instructing the AI agent to run local PowerShell scripts (e.g., New-ArxApp.ps1). These scripts are executed with the -ExecutionPolicy Bypass flag, which is a standard requirement for running unsigned scaffolding scripts in a Windows environment.
  • [DYNAMIC_EXECUTION]: A custom template engine (Invoke-TemplateExpansion.ps1) is used to dynamically generate C++ source files. It parses template directives like [!if] and [!output] and replaces them with values derived from user parameters. This process happens entirely within the local skill environment using non-executable string replacements.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: User-provided strings for project and class names enter the script context via command-line arguments (SKILL.md).
  • Boundary markers: None; inputs are directly interpolated into source code templates.
  • Capability inventory: The generator scripts perform filesystem operations including directory creation (New-Item) and file writing (Set-Content, Copy-Item) in New-ArxApp.ps1.
  • Sanitization: The skill implements strict regex validation (e.g., ^[A-Za-z_][A-Za-z0-9_]*$) for all user-provided identifiers. This ensures that inputs are valid C++ symbols and prevents directory traversal or shell command injection through the parameters.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 11:53 PM
Security Audit — agent-trust-hub — acad-arx-wizard