The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from the application under test.
Ingestion points: Data enters the agent's context through maui devflow ui tree (UI hierarchy), maui devflow logs (application and console logs), and maui devflow webview snapshot (DOM content from Blazor views).
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when presenting this data to the LLM.
Capability inventory: The skill possesses significant capabilities including executing shell commands via dotnet, adb, and xcrun, as well as modifying files.
Sanitization: There is no mention of sanitizing or escaping the content retrieved from the app before interpolation.
[EXTERNAL_DOWNLOADS]: The skill encourages the installation of external development tools from trusted sources.
Evidence: Instructions include installing Microsoft.Maui.Cli, androidsdk.tool, and appledev.tools via the dotnet tool command.
Context: These are standard, well-known development tools provided by Microsoft and reputable community members.
[REMOTE_CODE_EXECUTION]: The skill includes a command to update its own instruction set from a remote source.
Evidence: The command maui devflow update-skill is documented as a way to download latest skill files from GitHub.
Context: This is a documented administrative feature of the CLI tool used to manage the agent's environment.

maui-ai-debugging