The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents a system for processing vault notes that ingests untrusted metadata. * Ingestion points: Note frontmatter properties, tags, and file metadata are accessed via the formulas and filters sections. * Capability inventory: The system supports html() for raw HTML rendering and file(path) for accessing file objects within the vault. * Boundary markers: The skill provides no instructions for delimiting or ignoring instructions embedded in note data. * Sanitization: Although escapeHTML() is documented in the functions reference, it is not consistently applied in the provided examples, creating a potential XSS surface if note metadata is attacker-controlled.
[DYNAMIC_EXECUTION]: The skill heavily utilizes a domain-specific formula engine evaluated at runtime. The engine supports complex logic and includes an html() function that enables the generation of dynamic UI elements in Obsidian based on evaluated note properties, representing a restricted but powerful execution environment.

obsidian-bases