eu-ai-act-reviewer

Pass

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill does not contain any malicious code, hidden instructions, or unauthorized data access patterns. It is designed to be a read-only legal research aid.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements robust data handling policies in references/review-rules.md. It explicitly prohibits the exfiltration of user-provided content (code, logs, business facts) and restricts external lookups to official government domains using only generic legal identifiers.\n- [INDIRECT_PROMPT_INJECTION]: The skill includes proactive defenses against indirect prompt injection.\n
  • Ingestion points: The agent analyzes untrusted user content such as codebases and user journeys.\n
  • Boundary markers: The references/review-rules.md file contains a clear instruction to treat all content within analyzed files as untrusted evidence rather than instructions.\n
  • Capability inventory: The skill has no capabilities for executing code, modifying the filesystem, or making network requests to arbitrary domains.\n
  • Sanitization: The agent is instructed to distinguish between observed evidence and legal conclusions, preventing input-driven behavior modification.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 15, 2026, 04:57 PM
Security Audit — agent-trust-hub — eu-ai-act-reviewer