The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to install the markstream-react package. As this is not a pre-approved or widely recognized library, users should verify its origin and integrity before installation.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the rendering of data provided via the content and nodes props, which are intended to hold AI-generated output. This creates a surface for indirect prompt injection attacks.
Ingestion points: The skill processes external content through the content prop in components from the markstream-react library (e.g., in SKILL.md workflows).
Boundary markers: The instructions note that htmlPolicy defaults to safe and Mermaid strict mode is enabled, providing a layer of protection against malicious input.
Capability inventory: The renderer supports markdown, HTML, and Mermaid diagrams. If safety policies are lowered, this could enable malicious scripts or UI components to be injected into the application.
Sanitization: While defaulting to a safe policy, the skill explicitly provides instructions on how to use htmlPolicy="trusted" and mermaidProps={{ isStrict: false }}, which disables built-in sanitization and increases exposure to Cross-Site Scripting (XSS) and other content-based attacks.

markstream-react