The Agent Skills Directory

[PROMPT_INJECTION]: The implementation patterns for the Angular chat interface include a vulnerability surface for Indirect Prompt Injection through unsafe content rendering.
Ingestion points: Untrusted data enters the agent context through the ChatInputComponent (user input) and the streaming response chunks processed in ChatComponent (external AI data).
Boundary markers: The UI uses role-based message bubbles to separate user and assistant content, but these are for display only and do not provide programmatic instruction isolation for the agent.
Capability inventory: The skill utilizes HttpClient for network requests and the marked library for markdown parsing.
Sanitization: A significant sanitization failure was identified in reference/streaming-patterns.md. The StreamingMessageComponent uses DomSanitizer.bypassSecurityTrustHtml to render the output of the markdown parser. This explicitly disables Angular's built-in Cross-Site Scripting (XSS) protections. If an attacker-controlled AI response contains malicious HTML or script tags that pass through the markdown parser, they will be executed in the user's browser context.

ai-chat