avalonia
Warn
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The
preview-serversub-skill includes Python scripts (scaffold_host.py,render_variants.py) that utilizesubprocess.runto executedotnetCLI commands for project creation, package management, and application execution. - [REMOTE_CODE_EXECUTION]: The
render_variants.pyscript identifies and installs NuGet packages from apackages.txtfile generated by the agent at runtime. This allows for the dynamic inclusion and execution of arbitrary third-party libraries from the NuGet registry. - [DYNAMIC_EXECUTION]: The skill facilitates the generation of C# and XAML source code into a local project structure (
.preview/AvaloniaPreviewHost/) which is then compiled and executed usingdotnet runto generate visual previews. This 'code generation and execution' loop is a known vector for code injection if input is not strictly validated. - [EXTERNAL_DOWNLOADS]: The documentation and scripts encourage the resolution and download of numerous community-provided Avalonia UI packages from external repositories (NuGet). While many are well-known (e.g., SukiUI, FluentAvalonia), the system allows for the resolution of any package name provided in the variant configuration.
- [INDIRECT_PROMPT_INJECTION]: The 'Preview Server' feature creates an attack surface for indirect prompt injection. A malicious instruction could potentially influence the agent to generate a UI variant containing malicious package references or XAML-based exploits, which are then processed and executed by the provided utility scripts.
Audit Metadata