| name | pci-dss-network-segmentation |
| description | Validates PCI-DSS v4.0 network segmentation and scope reduction—Requirement 1.x (firewalls, network security controls) and 2.x (secure configurations)—for Cardholder Data Environment (CDE) isolation. Trigger when scoping PCI environments, reviewing firewall rules, VLAN segmentation, agent/MCP access to CDE, or reducing assessment scope. Do not use for payment-page script audits (use pci-dss-script-audit) or general IAM reviews without CDE focus (use access-control-identity-audit). |
PCI-DSS Network Segmentation
Overview
This skill implements PCI-DSS v4.0 Requirements 1 and 2 for network segmentation and CDE scope reduction. Proper segmentation limits PCI assessment scope to systems that store, process, or transmit CHD/SAD (cardholder data/sensitive authentication data) and systems connected to them.
Key requirements:
- Req 1.2.x: Network security control (NSC) configuration standards, inbound/outbound rules, CDE segmentation
- Req 1.3.x: Prohibit direct public access to CDE; restrict connections between untrusted networks and CDE
- Req 1.4.x: Personal firewall on mobile/remote devices accessing CDE
- Req 2.x: Secure configurations—change defaults, harden NSCs and system components
AI agents and MCP servers must not bridge untrusted networks into the CDE without explicit NSC rules and business justification.
When to Use
Use this skill when:
- Scoping PCI assessment—identifying in-scope vs out-of-scope systems
- Reviewing firewall rules, security groups, ACLs, WAF between CDE and other zones
- Validating segmentation test (penetration test or connectivity probe) results
- Assessing whether agent/MCP infrastructure introduces CDE connectivity
- Scope reduction initiatives (moving payment to hosted iframe, P2PE)
- Hardening NSC default configurations (Req 2.1)
Do not use this skill when:
- Auditing checkout JavaScript (use
pci-dss-script-audit)
- General SOC 2 logical access without CDE (use
access-control-identity-audit)
- Encrypting data at rest (Req 3.x—note gap but use appropriate skill)
Core Process
Execute steps in order.
Step 1: CDE and connected-to identification
- Inventory systems that store, process, or transmit CHD/SAD.
- Identify systems connected to CDE (even if they don't touch CHD)—they enter scope unless adequately segmented.
- Document data flows: payment forms, tokenization, API gateways, MCP Postgres access.
- Artifact:
pci-scope-diagram-{id}.json with trust zones (CDE, DMZ, Corporate, Cloud).
Step 2: Network zone model
Define zones:
| Zone | Description | Default trust |
|---|
| CDE | CHD/SAD processing | Highest restriction |
| DMZ | Public-facing, no CHD storage | Restricted |
| Corporate | Internal business systems | Untrusted relative to CDE |
| Agent/MCP | Compliance agent runtime | Assess connectivity to CDE |
Map all inter-zone connections with protocol, port, source, destination, business purpose.
Step 3: Requirement 1.2 — NSC rule review
For each NSC (firewall, cloud SG, NACL, WAF):
- Export rule set (JSON/text) via cloud API or config backup.
- Verify deny-all default with explicit allow rules only.
- Check bidirectional justification for each CDE rule:
- Source IP/CIDR, destination, port, service owner, change ticket
- Flag violations:
ANY/ANY rules touching CDE
- Corporate zone → CDE without jump host or PAM
- Agent/MCP subnet → CDE database port (CRITICAL unless scoped tokenization proxy)
- Verify rule review at least every 6 months (Req 1.2.7)—collect review evidence.
Step 4: Requirement 1.3 — CDE boundary protection
- Confirm no direct Internet → CDE paths (except documented encrypted payment entry in DMZ).
- Verify DMZ → CDE limited to minimum ports (e.g., app tier only).
- Test from untrusted network (authorized pentest or automated probe):
- Attempt connectivity to CDE IPs on common DB/app ports
- Expected: connection refused/timeout unless explicitly allowed
- Document segmentation test methodology and results.
Step 5: Requirement 1.4 — Remote access
- Inventory remote access paths to CDE (VPN, ZTNA, admin consoles).
- Verify personal firewall on endpoints where applicable.
- Ensure multi-factor authentication on all remote CDE access (cross-ref CC6/Req 8).
Step 6: Requirement 2 — Secure NSC configuration
- Verify default passwords changed on all NSCs (Req 2.1).
- Remove unnecessary services, protocols, accounts (Req 2.2).
- Document configuration standards with version ID.
- Compare running config vs standard—drift = FAIL until remediated or authorized.
Step 7: Agent/MCP segmentation assessment
- Map compliance agent and MCP server network placement:
- Playwright MCP: runs from which subnet? Can it reach CDE checkout URLs only (expected) or CDE backend?
- Postgres MCP: which databases reachable? CDE DB = in scope for agent infrastructure
- If agent bridges zones, require:
- Dedicated jump architecture, read-only replicas outside CDE, or tokenized data only
- NSC rules documented with PCI QSA review
- Never store CHD in agent evidence DB without encryption and scope approval.
Step 8: Scope reduction validation
- Evaluate scope reduction strategies:
- SAQ A/A-EP: outsourced payment, iframe isolation
- Tokenization: agent sees tokens only, not PAN
- Network segmentation: documented isolation per Appendix D guidance
- If claiming reduced scope, provide segmentation test evidence proving out-of-scope systems cannot reach CDE.
Step 9: Findings and evidence
- Classify: CRITICAL (CDE exposed to Internet), HIGH (overly broad rule), MEDIUM (missing review documentation).
- Package: rule exports, segmentation test results, scope diagram, remediation plan.
Common Rationalizations
| Excuse the agent might generate | Required rebuttal |
|---|
| "We use tokenization, so network segmentation doesn't matter." | Tokenization reduces data scope but connected systems may still be in scope—segmentation still required. |
| "Cloud SG allows 0.0.0.0/0 on app port but WAF protects us." | NSC rules must be least privilege—reliance on WAF alone does not replace deny-by-default SG design. |
| "Compliance agent needs CDE DB access for audits." | Agent access expands PCI scope—use read replicas, sanitized data, or out-of-band QSA processes. |
| "Segmentation test passed last year." | Req 1.2.7 requires ongoing reviews and tests when topology changes—stale tests are insufficient. |
| "MCP runs in same VPC—that's internal, so safe." | Same VPC without zone segmentation often means flat network—explicit CDE isolation required. |
| "Temporary firewall rule for debugging is fine." | Undocumented temporary rules are audit failures—require ticket and expiry. |
Red Flags
- CDE database security group allows corporate LAN or 0.0.0.0/0
- Agent/MCP Postgres tool can query tables containing PAN or SAD
- Segmentation test not performed after major cloud migration
- NSC rule review overdue (>6 months) with no compensating evidence
- Default vendor credentials on firewall/cloud NSC console
- Flat VPC with CDE and general workloads sharing unrestricted east-west traffic
- CHD found in agent evidence store or MCP logs
Verification