The Agent Skills Directory

Command Execution: The skill directs the agent to call dotnet new --help to identify reserved command names. This is a benign operation used to ensure template names do not conflict with platform subcommands.
Indirect Prompt Injection: The skill ingests user-provided template.json files, creating a potential ingestion surface. The evidence chain includes: (1) Ingestion Point: template.json via the workflow in SKILL.md. (2) Boundary Markers: A strict 'parse gate' and defined reporting table in SKILL.md. (3) Capability Inventory: File system read and dotnet new --help execution. (4) Sanitization: Mandatory JSON syntax validation before semantic analysis. These factors provide a structured environment for handling untrusted data.
Trusted External Sources: The skill links to official .NET documentation on GitHub, which is a recognized trusted repository for developer resources.

template-validation