The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs file system modifications such as deleting directories and renaming files to uninstall or disable community plugins. These actions are governed by safety rules requiring explicit user approval and logging of all operations.
[DATA_EXFILTRATION]: The skill accesses local paths within the agent's configuration directory to read audit logs and identify plugin files. While these paths are sensitive, access is restricted to the plugin ecosystem and no network exfiltration was found.
[PROMPT_INJECTION]: The skill interacts with external data by reading the names and paths of community skills from a log file. It includes defensive instructions to ignore any commands embedded in third-party content, reducing the risk of indirect prompt injection.

skill-manager