frontend-design
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: The skill exhibits deceptive metadata across multiple files. The README.md falsely attributes authorship to Anthropic employees, whereas the plugin.json identifies a different author. Furthermore, the skill-report.json file contains a fabricated security audit that claims verification by a non-existent AI model (gpt-5.5), which appears intended to mislead reviewers regarding the skill's origin and safety.
- [INDIRECT_PROMPT_INJECTION]: The skill contains self-referential metadata designed to influence the behavior of security analysis tools. The skill-report.json file includes pre-computed verdicts such as risk_level: safe and is_blocked: false, attempting to inject safety conclusions into the evaluation process. * Ingestion points: skill-report.json metadata file. * Boundary markers: Absent. * Capability inventory: The skill code in index.js is an implementation stub, and SKILL.md contains design instructions. * Sanitization: None.
Audit Metadata