mint
Fail
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPERSISTENCEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of system-critical commands, including
sudo update-grubfor bootloader changes,sudo cryptsetupfor managing disk encryption, andsudo aptfor package management. - [PRIVILEGE_ESCALATION]: Instructions frequently direct the agent to use
sudofor modifying sensitive system files such as/etc/default/gruband for managing system-level services and drivers. - [REMOTE_CODE_EXECUTION]: The skill is designed to source and execute shell scripts located in
~/.claude/execution/, effectively executing external code defined in the skill's infrastructure. - [DATA_EXFILTRATION]: The skill utilizes tools like
xclipandxselto read from the system clipboard, which may contain sensitive user information. It also accesses system logs and detailed hardware metadata. - [PERSISTENCE]: The skill provides workflows to modify the GRUB bootloader configuration and manage system services, which can be utilized to establish persistence on the host machine.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from the system clipboard and external web searches. Due to its high-privilege shell access, this poses a risk if processed data contains instructions designed to manipulate the agent.
- [DYNAMIC_EXECUTION]: The skill uses the
sourcecommand to dynamically execute bash scripts from the~/.claude/execution/directory at runtime during status checks and automated workflows.
Recommendations
- AI detected serious security threats
Audit Metadata