The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/create_winforms.ps1 includes an Export-WinFormsScript function that generates new PowerShell scripts by directly interpolating user-provided control names and text into a string template. This allows for script injection vulnerabilities where malicious PowerShell code can be embedded into the generated files.
[DYNAMIC_EXECUTION]: The scripts/build_wpf.ps1 script uses [System.Windows.Markup.XamlReader]::Load() to parse UI definitions. This .NET method is unsafe for untrusted input, as XAML can be crafted to execute arbitrary system commands via features like ObjectDataProvider.
[DYNAMIC_EXECUTION]: The TUI design script scripts/design_tui.ps1 executes scriptblocks passed via the MenuItems parameter using the call operator (&), creating a vector for executing arbitrary code defined in the menu configuration.
[COMMAND_EXECUTION]: The TypeScript wrapper scripts/ui_architect_wrapper.ts spawns powershell.exe and passes user-defined objects as command-line arguments. Although it attempts minimal escaping, the practice of passing complex, unsanitized data from a higher-level language into a shell environment is a common source of command injection bugs.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a bridge for user-supplied UI specifications to reach sensitive code-execution sinks without proper validation or boundary enforcement.
Ingestion points: UI parameter objects (controls, menu items, XAML paths) passed to the architect wrapper methods.
Boundary markers: None present in the instruction sets or scripts.
Capability inventory: Local script execution via spawn, file writing/script generation in create_winforms.ps1, and unsafe XAML parsing in build_wpf.ps1.
Sanitization: No evidence of input sanitization or filtering before data is used in script generation or dynamic loading.

powershell-ui-architect