The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from local Obsidian notes to populate views and evaluate computed properties.
Ingestion points: Note frontmatter (note.author, author), file metadata (file.properties, file.tags), and note content referenced via properties.
Boundary markers: Data is processed within a YAML-based configuration structure, but formulas interpolate property values directly into evaluation logic without enforced boundary markers or warnings to ignore embedded instructions.
Capability inventory: The skill provides an html() function for rendering strings as HTML and a file(path) function for accessing file objects within the vault. It also includes extensive string and list manipulation functions.
Sanitization: While an escapeHTML() function is documented in the reference, its use is not required or demonstrated in the provided workflows or examples, increasing the risk of XSS or injection if note properties contain malicious strings.
[DYNAMIC_EXECUTION]: The skill implements a domain-specific language for formulas and filters that are evaluated at runtime.
Evidence: The formulas and filters sections in SKILL.md define logic that is dynamically executed based on vault data. The html(string) function in references/FUNCTIONS_REFERENCE.md enables the dynamic rendering of content based on runtime evaluations.

obsidian-bases