design-taste-frontend

Fail

Audited by Gen Agent Trust Hub on Aug 7, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill is configured to extract and run a pre-compiled Windows executable (taste-ui.exe) via scripts/ensure-ui.mjs. Bundling and executing opaque binaries is a high-risk pattern that allows for the execution of unreviewable code.
  • [COMMAND_EXECUTION]: The scripts/ensure-ui.mjs script performs shell command execution using execSync (PowerShell) to extract archives and spawn (cmd.exe) to launch the binary in a hidden, detached process.
  • [PROMPT_INJECTION]: The skill provides an indirect prompt injection surface by ingesting user-provided design briefs to generate code.
  • Ingestion points: User-supplied design briefs in SKILL.md.
  • Boundary markers: None provided to isolate input from instructions.
  • Capability inventory: Execution of shell commands and external binaries in scripts/ensure-ui.mjs.
  • Sanitization: No validation or sanitization procedures for the input brief are described.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 7, 2026, 09:42 PM
Security Audit — agent-trust-hub — design-taste-frontend