The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the AI to use a mode: "bypassPermissions" setting when launching background agents. This is a direct attempt to subvert the agent's safety and permission boundaries.
[COMMAND_EXECUTION]: The instructions require the execution of shell commands such as ls to interact with the file system directly to list screenshot files.
[DATA_EXFILTRATION]: The skill reads and indexes course materials and system-generated manifest.json files. The presence of instructions to bypass permissions significantly elevates the risk associated with this data access.

visual-indexer