2dg
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill determines its control flow by directly parsing user input provided through the
$ARGUMENTSplaceholder without sanitization. - Ingestion points: Raw user input from
$ARGUMENTSinSKILL.mdis used to select the execution mode and subject. - Boundary markers: The instructions lack delimiters or clear separation between system logic and user-provided data, which could allow a malicious user to manipulate the agent's logic flow.
- Capability inventory: The skill can load and follow external skill definitions (
../2d-generalist/SKILL.md), read local reference files, and perform system diagnostic checks. - Sanitization: There are no filters or validation steps applied to the user input before it is processed by the logic engine.
- [COMMAND_EXECUTION]: The
auditmode is explicitly designed to check for the presence of local software tools. This functionality requires the agent to execute shell commands (such aswhichor similar discovery utilities) on the host system to determine the environment's state. It also provides installation instructions which, if followed by the user, could lead to the execution of privileged commands or external software installation.
Audit Metadata