The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and render MDX content stored in a database, creating a surface for indirect prompt injection where malicious instructions embedded in blog data could influence the agent's behavior.
Ingestion points: SKILL.md describes the BlogPostContent component which takes a content string (sourced from a database) and passes it to the serialize function.
Boundary markers: The provided instructions and code snippets do not include explicit boundary markers or instructions to the agent to ignore potentially embedded directives within the content.
Capability inventory: The skill itself focuses on rendering, but the agent environment typically includes capabilities for file system access, command execution, and network operations that could be targeted by an injection.
Sanitization: The skill utilizes standard MDX serialization without specific sanitization aimed at preventing instruction-based attacks.
[DYNAMIC_EXECUTION]: The skill utilizes next-mdx-remote/serialize and the <MDXRemote /> component to dynamically compile and execute MDX strings at runtime. While this is the intended and primary purpose of the skill for blog rendering, it involves the execution of code generated from external data.

nextjs-mdx-blog