human-gate-designer

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill documents an architectural pattern for 'Feedback Injection' that is susceptible to indirect prompt injection.
  • Ingestion points: External data enters the agent context via the 'human feedback' text input during the modification phase (SKILL.md).
  • Boundary markers: The template uses basic double quotes as delimiters, which are easily bypassed by adversarial input.
  • Capability inventory: According to the frontmatter, the skill environment allows Read, Write, and Edit tools, which could be abused if an injected instruction is executed (SKILL.md).
  • Sanitization: There is no evidence of input escaping or instructions to ignore embedded commands within the feedback string.
  • [SAFE]: No evidence of hardcoded credentials, unauthorized network communication, or persistent mechanisms was found. The skill primarily provides instructional content and design patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 08:31 PM
Security Audit — agent-trust-hub — human-gate-designer