The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documentation describes a conversational UI that processes untrusted data from AI models and user-uploaded files, which represents an indirect prompt injection surface.
Ingestion points: Untrusted data is ingested through the DxAIChat component via chat responses and the file attachment feature (DxAIChatFileUploadSettings) documented in examples/file-attachments.razor.
Boundary markers: The skill correctly uses ChatRole (System, User, Assistant) to delimit message types, helping to separate instructions from untrusted data.
Capability inventory: The skill facilitates chat rendering and file handling; it does not demonstrate or encourage unsafe operations like direct shell execution or arbitrary file system writes.
Sanitization: The documentation in references/features.md specifically advocates for the use of HtmlSanitizer and Markdig to sanitize Markdown responses before rendering them as HTML, mitigating XSS risks from AI output.
[SAFE]: The skill demonstrates excellent security posture by providing explicit warnings against hardcoding credentials and recommending standard secret management tools like dotnet user-secrets and Azure Key Vault.

devexpress-blazor-ai-chat