linux-shell-scripting
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Multiple scripts in SKILL.md include commands for sensitive operations such as file removal (rm), package installation (apt-get), and service management (systemctl).
- [PRIVILEGE_ESCALATION]: The toolkit provides scripts utilizing sudo, useradd, and passwd to modify system permissions and create administrative accounts.
- [PERSISTENCE]: The task scheduling script in SKILL.md automates the modification of the system crontab, which can be used to establish persistent execution of scripts.
- [REMOTE_CODE_EXECUTION]: The remote execution utility in SKILL.md uses SSH to pipe local script content into a remote bash shell.
- [INDIRECT_PROMPT_INJECTION]: The templates in SKILL.md ingest untrusted data through positional parameters (e.g., $1, $2) without boundary markers or sanitization, and combine this with capabilities like rm and ssh, creating an injection surface. The evidence chain includes: 1. Ingestion points: Positional parameters and variables assigned from them in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: File deletion (rm), remote execution (ssh), package management (apt-get), and system modifications (crontab, useradd, systemctl) in SKILL.md. 4. Sanitization: Absent; input is interpolated directly into shell strings.
Audit Metadata