The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion and rendering of streaming output from AI models, which is an attack surface for indirect prompt injection.
Ingestion points: The content prop in the MarkdownRender component directly processes strings originating from external AI generations.
Boundary markers: The implementation example does not demonstrate the use of system-level delimiters to separate instructions from rendered content.
Capability inventory: The component supports Markdown, Mermaid diagrams, and HTML rendering, which could be exploited for UI redressing or client-side code execution if misconfigured.
Sanitization: The instructions explicitly warn against enabling 'trusted HTML' and mandate html-policy="safe", providing significant mitigation for the injection surface.
[EXTERNAL_DOWNLOADS]: The skill documentation guides the installation of several packages from public registries and references external code from the Simon-He95 repository.
Evidence: The workflow and example code require installing and importing markstream-vue, shiki, and monaco-editor to enable the renderer's core and peer functionality.

markstream-vue