The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from Obsidian drawings and vault-relative JSON files, which represents a potential attack surface for instructions embedded in that data.
Ingestion points: The skill reads drawing elements using the :read command and accepts external JSON input via the input parameter in SKILL.md.
Boundary markers: There are no instructions or delimiters provided in SKILL.md to ensure the agent ignores potentially malicious instructions embedded within the drawing data or JSON files.
Capability inventory: The skill possesses significant capabilities described in SKILL.md, including file writing (:export), CLI interaction, and script execution (:script).
Sanitization: No sanitization, filtering, or validation steps are mentioned in SKILL.md for the external data before it is processed by the agent or the plugin.
[DYNAMIC_EXECUTION]: The skill enables the execution of local script files through the plugin's API.
The :script command in SKILL.md allows for the execution of existing local files with a trusted=true flag.
The instructions explicitly state that the execution environment is not a sandbox, meaning executed scripts may have access to other APIs and system resources available to the host application.
[COMMAND_EXECUTION]: The skill instructs the agent to interact with the local environment by executing commands through the obsidian CLI tool to communicate with the obsidian-excalidraw-plugin.

obsidian-excalidraw-cli