The Agent Skills Directory

[PROMPT_INJECTION]: The skill describes an indirect prompt injection surface by documenting how to ingest and process data from potentially untrusted sources.
Ingestion points: Functions like tp.file.content, tp.system.clipboard(), and tp.web.request() allow the agent to read external data into the template processing context.
Boundary markers: The documentation lacks instructions for the agent to treat external data as untrusted or to ignore embedded instructions.
Capability inventory: Templater can create new files (tp.file.create_new), rename or move files (tp.file.move), and execute arbitrary JavaScript code via execution tags (<%* %>).
Sanitization: There is no mention of content sanitization or validation before the data is used in template generation or execution.
[COMMAND_EXECUTION]: The skill mentions the obsidian-cli tool as a way to perform vault operations, which involves executing CLI commands such as obsidian read or obsidian properties.
[DATA_EXFILTRATION]: Documents the tp.web.request() function, which enables the skill to make HTTP GET requests to external URLs, creating a potential path for data transmission to non-whitelisted domains.

obsidian-templater