Security audit
Security checks for vulnerabilities and agentic risk
The skill fits its advertised purpose, but logging defects can retain sensitive text and sometimes ignore the no-logging option.
Review the logging defects before using sensitive text. Disable audit logging with NIMBLE_NO_LOG=1 or an empty logFile; --no-log alone does not cover evaluations. Keep the endpoint local or choose a trusted remote server. The publisher should remove raw error bodies from logs and honor the opt-out during evaluations.
Without declared permissions the skill's intent is opaque and cannot be validated.
runEvalSuite calls logCall with disabled=false, so evaluations persist metadata even when the invocation specifies --no-log. This violates an explicit privacy control, although the recorded information is limited to evaluation metadata and logging can still be disabled through NIMBLE_NO_LOG=1 or an empty logFile.
For unsuccessful API responses, the code persists the first 200 characters of the raw response body, which can contain state or answers echoed by the server. This bypasses the stated content-free logging guarantee and could leave sensitive classification input in persistent logs; the local default endpoint reduces exposure but does not prevent accidental server echoes or risks from a configured remote endpoint.
Detected: suspicious.dangerous_exec
