model-audit

The skill focuses on enumerating and validating AI model usage by scanning configuration files, resolving aliases to real models, and reporting a routing table. It does not involve credential handling, direct data exfiltration, or remote execution. The footprint is coherent with a model-audit/observability objective and poses low security risk provided appropriate access controls are in place to protect internal model mappings. Primary concerns are potential information disclosure of internal mappings and ensuring deprecated/renamed models are tracked, but there is no evidence of malicious activity or high-risk data flows in the described scope.