Summary
Adopt SPIFFE/SPIRE (CNCF graduated) as the workload identity layer to enable verifiable, federated agent identity across Kubernetes clusters. Opt-in per AgentWorkload; ServiceAccount + JWT remains the intra-cluster default.
Why
Enterprise NineVigil deployments span multiple clusters (multi-region, multi-tenant, multi-org). Today an agent in Cluster A has no zero-secret way to authenticate to a service or another agent in Cluster B. Existing options (shared secrets, mTLS without workload identity, central OIDC) all fail for air-gapped and regulated environments — which is exactly our ICP.
Triggered by @JacobSobolev on X (19 May 2026).
Design
Full design proposal: docs/rfcs/0001-cross-cluster-agent-identity.md
Validation gate (HARD)
This epic is in signal-collection mode. No implementation work begins until either:
- 6+ distinct external use cases in the linked GitHub Discussion (each comment naming a concrete deployment scenario), OR
- 1 paying customer explicitly requests this feature in writing.
Until the gate clears, only the RFC and tracking issues are active work.
Roadmap
Stories (sub-issues)
Linked below as sub-issues. All stories beyond Story 1 (RFC authoring) and Story 2 (validation gate) are blocked until the gate clears.
Related roadmap issues
After roadmap reconciliation on 2026-05-19, this epic absorbs and cross-links the following pre-existing issues. Each is preserved with the appropriate scope; this epic is the canonical implementation tracker:
Absorbed (closed as superseded):
Cross-linked (kept open, implementation tracked here):
Related but distinct scope (no implementation overlap):
Related
Summary
Adopt SPIFFE/SPIRE (CNCF graduated) as the workload identity layer to enable verifiable, federated agent identity across Kubernetes clusters. Opt-in per
AgentWorkload; ServiceAccount + JWT remains the intra-cluster default.Why
Enterprise NineVigil deployments span multiple clusters (multi-region, multi-tenant, multi-org). Today an agent in Cluster A has no zero-secret way to authenticate to a service or another agent in Cluster B. Existing options (shared secrets, mTLS without workload identity, central OIDC) all fail for air-gapped and regulated environments — which is exactly our ICP.
Triggered by @JacobSobolev on X (19 May 2026).
Design
Full design proposal:
docs/rfcs/0001-cross-cluster-agent-identity.mdValidation gate (HARD)
This epic is in signal-collection mode. No implementation work begins until either:
Until the gate clears, only the RFC and tracking issues are active work.
Roadmap
Stories (sub-issues)
Linked below as sub-issues. All stories beyond Story 1 (RFC authoring) and Story 2 (validation gate) are blocked until the gate clears.
Related roadmap issues
After roadmap reconciliation on 2026-05-19, this epic absorbs and cross-links the following pre-existing issues. Each is preserved with the appropriate scope; this epic is the canonical implementation tracker:
Absorbed (closed as superseded):
Cross-linked (kept open, implementation tracked here):
Related but distinct scope (no implementation overlap):
Related
docs/a2a-architecture.mddocs/security/threat-model.md