| name | design-segmentation-and-zero-trust |
| description | Design network segmentation and a zero-trust network access posture by traversing the segmentation decision tree (what to isolate -> blast radius -> enforcement point -> identity model), then return the segment plan (VLANs/VRFs/microsegmentation), the east-west + north-south policy model, the NAC/802.1X posture, and the SASE/SD-WAN or firewall enforcement point. Reach for this when the user asks "how do we segment <IoT/OT/guest/PCI> off the network?". Used by `network-architect` (primary); security verdicts escalate to security-engineering. |
Skill: design-segmentation-and-zero-trust
Invoked by: network-architect (primary); network-operations-engineer when an ops fix touches policy.
When to invoke: "segment IoT/OT/guest/PCI off the rest"; "how do we do zero-trust on the network?"; "limit east-west blast radius".
Output: segment plan + policy model + NAC posture + enforcement point. The security verdict (is this policy sufficient?) escalates to security-engineering.
Procedure
- Name what must be isolated and why — IoT, OT/ICS, guest, BYOD, PCI/cardholder data, dev/test, vendor/3rd-party, management plane. Each has a different trust level and a regulatory driver.
- Assess the blast radius — what can a compromised host on each segment reach today? Flat networks let one compromise pivot everywhere; that is the problem segmentation solves.
- Traverse the segmentation decision tree (
../../knowledge/network-topology-decision-trees.md):
- Coarse isolation, traditional → VLANs + inter-VLAN ACLs / firewall on a stick.
- Routing/tenant isolation → VRFs (separate routing tables).
- Fine-grained, workload-level, "default-deny east-west" → microsegmentation (host-based / hypervisor / NSX / Illumio / Cisco ACI).
- Remote/branch/cloud-edge identity-aware access → ZTNA / SASE (replace flat VPN with per-app, identity-gated access).
- Define the policy model — north-south (to/from internet/DC) and east-west (segment↔segment). Default-deny east-west where blast radius warrants; allow-list the flows the apps actually need (get the flow inventory — don't guess).
- Specify the identity/posture layer — 802.1X / NAC for wired+wireless admission (who+what is on the port), device posture (patched? managed?), and dynamic VLAN/segment assignment. Zero-trust = never trust the network location alone; authenticate identity + device every time.
- Name the enforcement point(s) — switch ACL, distributed firewall, NGFW, NAC, SASE PoP — and where logging/visibility lands.
- Escalate the security verdict to
security-engineering: this skill designs the segmentation; whether it's sufficient against the threat model is their call.
House guardrails
- Segment by trust, not by convenience. A VLAN that mixes IoT and finance is not segmentation.
- Default-deny east-west is the goal; allow-list from a real flow inventory — broad
any/any rules between segments defeat the purpose (the hook flags these).
- The management plane is a segment too — out-of-band management, not reachable from user VLANs.
- Zero-trust is a posture (identity + device + least-privilege), not a single product — don't let a SASE purchase substitute for the policy design.
Output contract
Isolate: <segments + driver (security/regulatory)>
Blast radius today: <what a compromise can reach>
Mechanism: <VLAN/VRF/microseg/ZTNA — per decision tree>
Policy: <north-south + east-west, default-deny scope, allow-list source>
Identity/posture: <802.1X/NAC, device posture, dynamic assignment>
Enforcement + visibility: <where policy is enforced and logged>
Security verdict: ESCALATE to security-engineering
Then emit the Structured Output Protocol JSON block.