ai-act-triage-pl

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill instructions are clear, pedagogical, and focused on compliance triage. There are no attempts to override agent behavior, bypass safety guardrails, or extract system prompts. It explicitly mandates human approval and verification of legal sources.
  • [DATA_EXFILTRATION]: No evidence of unauthorized data exposure or exfiltration. The skill uses the Read tool and the mcp-eu-compliance connector to fetch public legal documents from EUR-Lex. Data residency is restricted to local and PII egress is explicitly disabled.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted data from external sources and user inputs.
  • Ingestion points: External legal text fetched via the mcp-eu-compliance connector or EUR-Lex, and user-provided descriptions of AI systems.
  • Boundary markers: No explicit delimiters are used to wrap the external legal text to prevent instruction following within that content.
  • Capability inventory: The skill is restricted to Read operations and markdown reporting.
  • Sanitization: No specific sanitization or validation logic is defined for the external text before it is processed by the agent.
  • [REMOTE_CODE_EXECUTION]: No remote code execution patterns or unverifiable dependencies were found. The skill references legitimate connectors for compliance data.
  • [OBFUSCATION]: The content is provided in plain Polish with no hidden characters, encoded strings, or homoglyph substitutions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 08:27 AM
Security Audit — agent-trust-hub — ai-act-triage-pl