The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to parse and render markdown content that may come from untrusted sources, including real-time AI generation.
Ingestion points: Untrusted content enters via the parseMarkdown function and framework-specific rendering components as documented in SKILL.md and references/parsing-ast.md.
Boundary markers: The library provides an autoClose feature to manage malformed or incomplete syntax during streaming, though this is primarily for syntax validity rather than security isolation.
Capability inventory: The skill supports rendering custom components with properties and slots, and integrates with plugins like Mermaid and KaTeX. It also supports template binding (references/rendering-angular.md), which interpolates data into the rendered output.
Sanitization: Parsing is handled by the markdown-it engine, but the safety of rendered custom components depends on user-defined implementation and framework-level sanitization.
[DYNAMIC_EXECUTION]: The skill uses dynamic component resolution to map markdown tags to executable framework components at runtime.
Evidence: The Angular renderer (references/rendering-angular.md) utilizes createComponent and reflectComponentType for dynamic instantiation. The Svelte renderer (references/rendering-svelte.md) demonstrates using import.meta.glob to dynamically import and resolve components based on computed paths derived from the markdown source.

comark