design
Warn
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes shell commands (Node.js and PowerShell) using variables such as $PluginName and $selectedDesign that are derived from user input or library manifests. There is no evidence of validation or escaping for these variables, which could allow for command or argument injection if the input is maliciously crafted.
- [PRIVILEGE_ESCALATION]: The instructions for previewing the design explicitly recommend using the -ExecutionPolicy Bypass flag with PowerShell. This allows the execution of scripts that might otherwise be blocked by the system's security policy.
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting user-provided design requirements and metadata from a local library and interpolating them into design specifications and JSON map files.
- Ingestion points: User responses during Phase 1 requirements gathering (Style, Layout, etc.) and design metadata from design_library/manifest.json.
- Boundary markers: There are no visible delimiters or warnings used to prevent the agent from interpreting embedded instructions within the user-supplied data.
- Capability inventory: The skill possesses capabilities for file system modification (Out-File, Copy-Item) and shell command execution (node, powershell).
- Sanitization: No sanitization or validation logic is implemented for the interpolated content.
Audit Metadata