The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The eval command allows for the execution of arbitrary JavaScript within the Obsidian application context. This provides full access to the internal Obsidian API (including app.vault and app.workspace), enabling arbitrary code to be run with the privileges of the desktop application. Evidence: SKILL.md and command-reference.md Developer sections.
[EXTERNAL_DOWNLOADS]: The plugin:install and theme:install subcommands facilitate the downloading and activation of third-party code from community repositories. These extensions run within the Obsidian environment with full access to the vault. Evidence: SKILL.md Plugins and Themes sections.
[COMMAND_EXECUTION]: The skill exposes a wide range of file system operations, including creating, moving, and deleting files and folders within the vault. It also allows executing any registered Obsidian command via the command id= parameter. Evidence: SKILL.md Files and Commands sections.
[DATA_EXFILTRATION]: The skill enables reading any markdown file in the vault, searching through all notes, and capturing screenshots of the application window via dev:screenshot. This provides a significant surface for the exposure of sensitive information stored in the vault. Evidence: SKILL.md Read, Search, and Developer sections.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its ability to ingest untrusted data from the vault. Ingestion points: read, daily:read, search, search:context, property:read. Boundary markers: None present in instructions. Capability inventory: eval, plugin:install, delete, create, command. Sanitization: None present.

obsidian-cli