The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the filesystem (Obsidian notes) and has the capability to execute commands via the terminal tool.
Ingestion points: File content read via read_file or search_files (SKILL.md).
Boundary markers: None specified to differentiate between note content and agent instructions.
Capability inventory: terminal (shell access), write_file, and patch (filesystem modification).
Sanitization: No sanitization or validation of note content before processing is mentioned.
[COMMAND_EXECUTION]: The instructions explicitly permit the use of the terminal tool for resolving environment variables (e.g., OBSIDIAN_VAULT_PATH) or performing file operations when native tools are insufficient.

obsidian