The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled metadata and frontmatter from Obsidian notes, establishing an ingestion surface for untrusted data. * Ingestion points: SKILL.md documents functions to query note properties (file.properties) and metadata (file.path, file.tags). * Capability inventory: references/FUNCTIONS_REFERENCE.md lists powerful functions for logic, list processing (map, reduce), and UI rendering (html, image). * Boundary markers: No specific delimiters are provided to separate schema instructions from the ingested note data. * Sanitization: The presence of the html() function allows for rendering strings as HTML without mandatory sanitization, although escapeHTML() is available as a utility.
[DYNAMIC_EXECUTION]: The skill defines a domain-specific language for formulas that are evaluated at runtime to process note data and configure views. * The html(string) function provides a mechanism to render arbitrary strings as HTML content, which can be exploited if the input strings include note properties containing malicious scripts. * The formula engine supports complex transformations and functional programming patterns like map and reduce, expanding the potential impact of processed untrusted data.

obsidian-bases