agentsclimarketplace

Cilium network policy review

Skill Raishin/vanguard-frontier-agentic/skills/cilium/cilium-network-policy-review

Use this skill for Cilium network policy review across the three policy formats (Kubernetes NetworkPolicy, CiliumNetworkPolicy, CiliumClusterwideNetworkPolicy), L7 policy via embedded Envoy, ClusterMesh cross-cluster semantics, Hubble flow observability, and CiliumEgressGatewayPolicy. Trigger when the user asks whether a network policy is too broad, whether default-deny is in place, whether L7 rules will actually be enforced, whether ClusterMesh policy semantics are correct, or whether an egress gateway IP collision is possible.From its SKILL.md

Install
npx -y skills add Raishin/vanguard-frontier-agentic --skill cilium-network-policy-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 20 stars20 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.8 KB, 705 tokens by cl100k_base, as published. Nobody here has run it

Cilium Network Policy Review

Purpose

Review Cilium policy resources against zero-trust correctness, blast radius, and the operational traps unique to eBPF-backed networking. Cilium's policy surface is broader than native Kubernetes NetworkPolicy — CiliumNetworkPolicy adds L7 rules, FQDN matching, ICMP control, and identity-based selectors; CiliumClusterwideNetworkPolicy applies cluster-wide; CiliumEgressGatewayPolicy controls SNAT egress IPs; and policy-default-local-cluster changes how policy evaluates across ClusterMesh.

Lean operating rules

  • Prefer live cluster evidence (kubectl get networkpolicies,ciliumnetworkpolicies,ciliumclusterwidenetworkpolicies,ciliumegressgatewaypolicies -A -o yaml, cilium policy get, cilium clustermesh inspect-policy-default-local-cluster, and Hubble flow observation) when the active client exposes it; otherwise fall back to official Cilium documentation (docs.cilium.io) and sanitized YAML.
  • Separate confirmed facts from inference. If Cilium agent state, ClusterMesh peer status, or Hubble flow data was not queried, say so.
  • Treat removal of a default-deny NetworkPolicy in a namespace as a critical finding — pods become reachable from any source/destination unless another policy provides isolation.
  • Treat CiliumNetworkPolicy egress with toCIDRSet: [{cidr: 0.0.0.0/0}] (no except for sensitive CIDRs) as a critical finding — unrestricted egress is a documented data exfiltration path.
  • Treat any change to policy-default-local-cluster in a ClusterMesh deployment as critical-blast-radius — every existing policy's cross-cluster semantics flip simultaneously.
  • Challenge CiliumEgressGatewayPolicy with the same egressIP used in two policies — silent connection breakage when both match.
  • Challenge L7 rules in CiliumNetworkPolicy for namespaces where Envoy proxy is not enabled — L7 fields require the proxy.
  • Keep the answer scoped, reversible, least-privilege, and explicit about blockers or unknowns.

References

Load these only when needed:

  • Evidence path and tooling — use when choosing live cluster evidence, confirming Cilium version and ClusterMesh state, or switching to documentation mode.
  • Workflow and output contract — use when executing the full review, applying stress checks across the three policy formats and ClusterMesh, or formatting the final answer.
  • Official sources — use when you need the detailed Cilium documentation list, CRD schema, and grounded insights.

Response minimum

Return, at minimum:

  • the scoped target (namespace NetworkPolicy, namespace CiliumNetworkPolicy, cluster-wide CiliumClusterwideNetworkPolicy, CiliumEgressGatewayPolicy) and evidence level,
  • the default-deny posture in the affected namespace(s),
  • the L7 enforcement assessment (Envoy proxy enabled / required) and whether L7 rules will actually run,
  • the ClusterMesh assessment when applicable (policy-default-local-cluster semantics),
  • the safest next actions and rollback plan,
  • the assumptions or blockers that prevent stronger conclusions.

What ships with it: 4 files

17.9 KB alongside SKILL.md

Keep looking

Skills are one crate of 326,782. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.