The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents a powerful CLI interface (obsidian) and provides shell script examples for vault management that include potentially destructive commands such as rm -f for backup cleanup.
[REMOTE_CODE_EXECUTION]: The skill provides detailed instructions and examples for executing arbitrary JavaScript within the Obsidian process context using obsidian eval and the Chrome DevTools Protocol Runtime.evaluate method. This allows an agent to perform high-privilege operations that are not restricted to standard vault management.
[DATA_EXFILTRATION]: The skill documentation describes how to read all vault files, access metadata, and capture screenshots of the application. When combined with documented Node.js networking capabilities (using the ws package), these features create a significant surface for unauthorized data access and exfiltration.
[PROMPT_INJECTION]: The skill establishes a high-risk capability tier by instructing the agent to read and process untrusted Markdown content from vaults while maintaining execution capabilities. There is a total absence of boundary markers or sanitization logic in the provided examples, making the agent vulnerable to indirect prompt injection where malicious instructions embedded in notes could trigger the described eval or CLI tools.
[SAFE]: The skill references standard developer resources and well-known tools like Chrome DevTools Protocol and Homebrew for installation, which are documented neutrally.

obsidian