The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses the .NET SDK to restore standard NuGet packages from official registries during the build process. The dependencies, which include Microsoft.Agents.AI and Azure.AI.OpenAI, originate from trusted organizations. It also provides a link to a well-known observability platform for tracing verification.
[COMMAND_EXECUTION]: The skill executes dotnet CLI commands to build projects, run smoke tests, and execute a local helper script. It also uses curl to verify service health. These operations are essential to the skill's purpose and are appropriately scoped to the local environment.
[DYNAMIC_EXECUTION]: A local C# script (copy-template.cs) is executed to manage the deployment of template assets. This script is part of the skill package and contains specific security checks, such as symlink validation, to prevent path-traversal vulnerabilities during file operations.
[INDIRECT_PROMPT_INJECTION]: The agents provided by this skill ingest external data such as Markdown documents and JSON tickets, creating a surface for indirect prompt injection. However, each template includes comprehensive system instructions to treat processed data as untrusted and ignore any embedded commands.
Ingestion points: Markdown manuals in docs/, support tickets in data/tickets.json, and CSV metrics.
Boundary markers: System prompts explicitly label retrieved content as untrusted data and forbid the agent from following instructions within it.
Capability inventory: Tools are restricted to local file retrieval and data aggregation; no capabilities for arbitrary shell execution or network writes are present.
Sanitization: Input parameters for all tools are strictly validated and normalized in the application code.

build-template-agent