The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the dotnet CLI to manage a temporary preview project and execute rendering tasks. The scripts/render_variants.py script uses subprocess.run to invoke dotnet restore, dotnet add package, and dotnet run. These calls use list-based arguments, which is a secure practice to mitigate shell injection risks.
Evidence: subprocess.run calls in scripts/render_variants.py that handle project restoration, package management, and the execution of the rendering host.
[DYNAMIC_EXECUTION]: The skill creates and executes code at runtime by scaffolding a dedicated .NET scratch project in the .preview/ directory. It populates this project with XAML components generated by the agent, which are then compiled and executed via Avalonia.Headless to produce visual previews.
Evidence: scripts/scaffold_host.py defines the project infrastructure and boilerplate; scripts/render_variants.py dynamically updates the Variant/ content before each render run.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where user requirements influence the generation of XAML files and NuGet package lists (packages.txt). While these inputs are processed by the .NET toolchain, the risk is consistent with the skill's primary purpose as a development tool and is mitigated by the local, headless execution context.
Ingestion points: packages.txt and .axaml files generated in the .preview/variants/ directory, which are subsequently read and processed by scripts/render_variants.py.
Capability inventory: CLI access for .NET package management and project execution via subprocess.
Boundary markers: Not explicitly defined in the instruction set for generated XAML content.
Sanitization: Basic string splitting and stripping for NuGet package names in scripts/render_variants.py.

avalonia-pro-max/preview-server