omniCell
Warn
Audited by Gen Agent Trust Hub on Mar 7, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
nuShelltool inscripts/nu_shell.pyallows the execution of arbitrary system commands using Nushell syntax. This gives the agent broad authority to run processes, scripts, and development tools on the host environment. - [DATA_EXFILTRATION]: Through shell commands such as
ls,open, andsys, the skill can access sensitive system information, configuration files, and private data. The referencedsys_querytool inreferences/navigation.mdfurther enables precise extraction of source code using AST patterns, which could be used to expose intellectual property. - [PROMPT_INJECTION]: The tool relies on a
commandstring argument which is highly susceptible to prompt injection. An attacker could influence the command content to bypass safety protocols, override intent classification (observe vs mutate), or execute destructive operations.
Audit Metadata