The Agent Skills Directory

Indirect Prompt Injection Surface: The architecture described in the skill facilitates the ingestion of MDX content which is then rendered alongside interactive code execution components. This creates a surface where untrusted content could potentially influence the behavior of the learning environment.
Ingestion points: The LessonPage component in references/architecture.md reads MDX files from the local file system using readFile based on URL parameters.
Boundary markers: No specific delimiters or "ignore instructions" warnings are present in the provided reference implementations to separate content from instructions.
Capability inventory: The skill includes components like CodePlayground and TaskRunner (found in references/code-learning.md) that are designed to execute code snippets, providing a functional execution environment.
Sanitization: The provided code snippets do not include explicit sanitization or validation of the MDX source content.
Dynamic File Path Construction: In references/architecture.md, the LessonPage component uses path.join with URL parameters (courseSlug, moduleSlug, lessonSlug) to locate MDX files. This is a common pattern in content-driven sites but is a consideration for directory traversal risks if the parameters are not properly validated to exclude sequences like ...
Runtime Content Compilation: The skill utilizes next-mdx-remote for runtime compilation of MDX content. While this is essential for providing the interactive features described, it represents a dynamic execution pattern that relies on the integrity of the content files to prevent script injection or other cross-site concerns.

geist-learning-lab