anfra-aml
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The AML language schema defined in references/type_definitions includes 'ai_context' fields intended to provide additional context to AI agents. This creates a surface where untrusted content in modeling files could attempt to influence the agent's behavior. * Ingestion point: 'ai_context' property in .aml files. * Boundary markers: Not specified in instructions. * Capability inventory: The agent has access to shell execution via the 'anfra' tool and file writing capabilities. * Sanitization: No sanitization or validation of the ai_context content is mentioned.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute arbitrary shell commands using the vendor's 'anfra' CLI tool to study database schemas. Evidence: 'anfra query --lang sql --data-source ' in SKILL.md.
Audit Metadata