The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates a web clipping automation that processes untrusted content from the browser into the Obsidian vault.
Ingestion points: External data is ingested in SKILL.md via the web_clipper workflow, specifically capturing {{page.title}}, {{page.url}}, and {{selection}} from arbitrary web pages.
Boundary markers: Absent. The templates interpolate untrusted content directly into notes without delimiters or instructions to the agent to disregard embedded commands.
Capability inventory: The skill uses tools including obsidian_create_note, obsidian_search, and obsidian_link. While limited to the Obsidian environment, these could be leveraged to create malicious notes or manipulate the internal knowledge graph if the agent obeys instructions hidden in clipped web content.
Sanitization: Absent. There is no evidence of content filtering, escaping, or validation for the data extracted from the web clipper or research sources.

Obsidian Automation