Skip to content

Security: brunovicco/stateops

Security

docs/SECURITY.md

Security and privacy model

StateOps is a demonstration runtime, not production incident automation.

  • Remediations are allowlisted simulations; there is no shell, Kubernetes, cloud, or database mutation tool.
  • A human interrupt gates every simulated effect.
  • Modified approvals can change only existing bounded string parameters.
  • Production LLM calls use the Governed LLM Gateway client. Provider credentials, model selection, retry, and fallback never enter StateOps.
  • Gateway output, HTTP input, checkpoint values, and resume payloads are untrusted at their boundaries.
  • Redis provides durable checkpoints and an atomic SET NX idempotency claim.
  • Serializer module allowlisting is explicit and pickle fallback is disabled.
  • Logs, traces, history responses, and governance evidence are metadata-only by default.

The local Compose password is development-only and its published ports bind to loopback. Real deployments must supply secret-managed Redis authentication, TLS, least-privilege ACLs, an eviction policy that protects workflow data, retention policy, network isolation, and backup/recovery appropriate to their environment.

There aren't any published security advisories