Skip to content

Security: loopgridio/loopgrid-openai-agents

Security

SECURITY.md

Security

  • Raw model/tool content is disabled by default (capture_content=False).
  • Use RunConfig(trace_include_sensitive_data=False) unless raw SDK trace content is intentionally required.
  • Reviewer identity must come from authenticated application state, not model output or browser-submitted fields.
  • Durable OpenAI Agents RunState snapshots must remain under trusted server-side control.
  • LoopGrid evidence verification proves integrity/provenance of recorded evidence; it does not prove that a business decision was correct.
  • Report security issues privately through the LoopGrid security contact/process rather than a public issue.

There aren't any published security advisories