The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-generated markdown files within a local Obsidian vault, which serves as an entry point for untrusted data. Maliciously crafted notes could attempt to influence the agent's actions during vault maintenance.
Ingestion points: The agent explores the file system using rg --files and find, and reads the contents of various markdown notes to perform categorization and updates.
Boundary markers: The instructions do not include explicit delimiters or commands for the agent to ignore instructions embedded within the notes it processes.
Capability inventory: The agent has permissions to create, update, move, and rename files, and is tasked with performing syntax validation on Python code blocks.
Sanitization: No sanitization or validation logic is defined to filter out potential instructions from the ingested note content.
[COMMAND_EXECUTION]: The skill utilizes standard system utilities including rg (ripgrep) and find to map the vault's structure. These commands are targeted at the specific local path /Users/huangrx6/Documents/obsidian and are legitimate for the skill's file management functions.

obsidian-personal-knowledge-base