responsive-design

Warn

Audited by Gen Agent Trust Hub on Aug 12, 2026

Risk Level: MEDIUMPROMPT_INJECTION
Full Analysis
  • [METADATA_POISONING]: The file skill-report.json contains a fabricated security audit report with fake model references (e.g., 'gpt-5.5') and pre-determined 'false positive' justifications. This metadata is deceptive and designed to mislead automated security assessment tools about the skill's safety status.
  • [INDIRECT_PROMPT_INJECTION]: The skill contains self-referential content in skill-report.json that attempts to pre-emptively validate its own security status ('risk_level': 'safe', 'is_blocked': false).
  • Ingestion points: The file skill-report.json is provided as part of the skill package and is ingested during analysis.
  • Boundary markers: Absent; the deceptive audit content is presented as structured metadata.
  • Capability inventory: The skill provides instructions and examples for CSS Grid, Flexbox, media queries, and responsive images in SKILL.md.
  • Sanitization: Absent; the agent or scanner processes the provided JSON files as authoritative metadata.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 12, 2026, 11:24 AM
Security Audit — agent-trust-hub — responsive-design