The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data from Azure Boards (work item descriptions and comments). It includes a mandatory evidence chain for managing this indirect prompt injection surface:
Ingestion points: Reads System.Description and System.WorkItemType from Azure DevOps via the wit_get_work_item and wit_list_work_item_comments tools.
Boundary markers: The instructions explicitly warn that work item content is provided in HTML and must be rendered to text to prevent tags from leaking into the logic.
Capability inventory: The skill has access to file writing (Write, Edit), shell execution (Bash for dotnet, make, and git), and agent management tools.
Sanitization: Instructs the agent to render HTML to text and reason over the content safely. It also explicitly forbids using git add -A to prevent accidental inclusion of sensitive files.
[CREDENTIALS_UNSAFE]: The skill mentions PERSONAL_ACCESS_TOKEN and how to configure it for the Azure DevOps MCP server. This is documented as a setup requirement for the user's environment and does not involve hardcoded credentials within the skill itself.
[COMMAND_EXECUTION]: The skill uses Bash to run standard development tools such as dotnet, make, and git. These are restricted to the intended purpose of building, testing, and committing code within a .NET repository.

plan-and-build-dotnet