devexpress-wpf-ai-chat-control
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data through a chat interface (user messages and file attachments). The instructions address this attack surface by mandating the use of HTML sanitization (e.g., via the
HtmlSanitizerlibrary) when converting AI-generated Markdown to HTML for rendering. - Ingestion points: User input area and file attachment settings in
AIChatControl(SKILL.md). - Boundary markers: Not explicitly required in snippets, but instructions focus on output sanitization.
- Capability inventory: The control connects to external LLM providers via
IChatClientimplementations. - Sanitization: Documentation explicitly requires sanitizing HTML output before assigning it to
e.HtmlTextin theMarkdownConvertevent. - [DATA_EXPOSURE]: The skill includes an
App.configfile containing placeholder credentials ('demo') for a DevExpress evaluation endpoint. These are clearly marked as rate-limited evaluation credentials and do not represent a leak of production secrets. - [SAFE]: The skill references standard, well-known libraries and NuGet packages from trusted vendors, including DevExpress, Microsoft, and established open-source projects like Markdig and OllamaSharp.
Audit Metadata