Skip to content

Security: Ledge-Labs/GRND0

Security

SECURITY.md

Security policy

Supported release

Security fixes apply to the latest published research preview.

Report channel

Private vulnerability reports use the repository security-advisory channel. Public issues are not a vulnerability-report channel.

Deployment invariants

  • Host port mappings bind to loopback by default.
  • Public binding requires endpoint authentication.
  • Endpoint and gateway credentials are generated independently.
  • Debug routes are disabled by default and remain authenticated when enabled.
  • Core and capability profiles contain no writable host mounts.
  • Local model weights are mounted read-only; generated inference configuration resides in a named volume and contains no credential.
  • Local document access is limited to read-only mounts and GRND0_AUTHORIZED_READ_ROOTS.
  • Web and local-git connectors remain disabled until separate runtime grants are present.
  • Web fetches reject non-public destinations and validate redirect hops.
  • The browser harness is isolated from the core service network on a dedicated egress bridge.
  • Autonomous remediation is absent from the public runtime.
  • Provider credentials remain runtime environment values and never enter packs, receipts, or repository files.
  • Cloud backends remain disabled until an explicit enable flag, endpoint, model, and lane-role assignment are present.

Receipt handling

Only redacted receipts are valid public issue attachments. The tools/export-receipt.py transform removes credentials, absolute paths, network identifiers, and operator-defined sensitive keys before output.

There aren't any published security advisories