design-visual-design
Fail
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [REMOTE_CODE_EXECUTION]: The file
references/layout-principles.mdcontains embedded PowerShell script commands (e.g.,Set-Content -Path "D:\j4flmao-org\..." -Encoding UTF8). This is a clear attempt to perform unauthorized file system modifications using absolute Windows paths. - [COMMAND_EXECUTION]: The skill uses PowerShell 'Here-String' syntax (
@"..."@) within its documentation files. This is indicative of an attempt to trick the AI agent into executing shell commands via its available toolset. - [INDIRECT_PROMPT_INJECTION]: The skill includes eight reference files (e.g.,
architecture-patterns.md,deployment-pipelines.md,state-management.md) that each contain 150 repetitive sections of unrelated technical text. This functions as a 'token bomb' intended to saturate the agent's context window, which can be used to displace safety instructions or hide malicious behavior. - [METADATA_POISONING]: Although the skill is named and described as a 'Visual Design' utility, it ships with thousands of lines of irrelevant documentation regarding distributed systems and kernel-level networking (eBPF). This deceptive metadata structure is used to mask the injection of large amounts of unrelated data into the agent's context.
Recommendations
- AI detected serious security threats
Audit Metadata