Layer 2/3 Network Trust-Boundary Assessment
Read the assessment methodology before using this skill. Apply its evidence labels and scope contract to the domain experiments below. This revision restores the upstream's L2/L3 craft as scoped experiments on an isolated authorized segment: it names the attacks and their detection signatures, confirming interception/redirect against a designated fixture rather than capturing real users' traffic.
Scope and prerequisites
These attacks abuse implicit trust in L2/L3 protocols (ARP, LLMNR, DHCP, DNS, 802.1Q, 802.1X). Run only on an isolated lab segment with a designated victim fixture you own; never intercept third-party traffic on a shared/production network. Capture only the fixture's traffic.
Declare the segment, the fixture, attempt budget, stop condition. Config/topology-only authorization yields a hypothesis marked NOT TESTED.
Attack surface and classes
- ARP spoofing — poison the fixture/gateway mapping to MITM.
- Name poisoning — LLMNR/NBT-NS/mDNS answer broadcast lookups → capture NetNTLMv2 (relay → see
rt-active-directory). - DHCP/DNS — rogue DHCP (option 3/6 gateway/DNS redirect), DHCP starvation, rogue DNS, DNS cache poisoning, DNS rebinding.
- MITM — HTTP proxy, SSL-strip, HSTS bypass (
sslstrip+/dns2proxy). - VLAN hopping — switch spoofing (DTP), double tagging (802.1Q).
- 802.1X bypass, IPv6 SLAAC/DHCPv6 RA attacks.
Mechanism and observation
Pick one attack; trace the trust it abuses to the redirect/interception it achieves against the fixture.
Firstness: Record the fixture's traffic reaching your controlled node (captured ARP redirect, poisoned response, VLAN frame delivered). Secondness: State the protocol's intended behavior and the deviation. Thirdness: Infer the trust-boundary class only after the fixture is controlled-redirected.
Abduction and refutation
Mechanism hypothesis: A protocol accepts unauthenticated assertions (ARP replies, name responses, DHCP offers, DTP/802.1Q tags), letting a rogue node redirect or impersonate.
Competing benign explanation: Dynamic ARP Inspection / DHCP snooping / RA Guard / port security block it; LLMNR/NBT-NS are disabled; the switch denies DTP and tagged frames on access ports; HSTS preload defeats SSL-strip; the "capture" is the fixture's own loopback.
Deduction and controls
On an unhardened segment the redirect succeeds (demonstration); on a hardened one it must fail. A confirmed finding delivers the fixture's traffic to your node where the hardened control blocks it. Predeclare the captured artifact. DAI/snooping/RA-Guard/port-security/HSTS defeating the attack are PASS controls.
Induction: bounded laboratory procedure
Isolated segment; designated fixture; capture only the fixture.
- Baseline. Record normal L2/L3 behavior and the fixture's path.
- Run one attack. e.g. ARP-spoof the fixture↔gateway; or stand up name poisoning; or a rogue DHCP offer with option 6; or DTP/double-tag into a target VLAN.
- Confirm redirect/interception of the fixture's traffic at your node (one artifact), then stop.
- Controls. Enable the corresponding hardening (DAI, snooping, RA Guard, port security, HSTS) and confirm the attack now fails.
- Detection. Record the signatures each attack produced.
Record results including refutations. Keep finding and control status separate.
Craft notes (leads)
Tooling: bettercap/arpspoof/Ettercap (ARP), Responder/Inveigh (name poisoning, hashes in /opt/Responder/logs), ntlmrelayx (relay), dnsmasq (rogue DHCP/DNS), Scapy (custom ARP/DNS/gratuitous), Yersinia (L2), singularity (rebinding). DNS cache poisoning must match in-flight txid and source port. Double-tagging: outer=native VLAN, inner=target. These are lab primitives, not a license to run them on shared infrastructure.
Domain-specific failure modes
A hardened switch (DAI, DHCP snooping, port security, RA Guard) defeats most of these — that is a PASS, report it. HSTS preload defeats SSL-strip. Capturing your own fixture's loopback is not interception. Running any of these on a production/shared segment is out of scope and can disrupt or surveil third parties.
Evidence and deliverable
Provide the attack, the isolated segment/fixture, the captured redirect artifact, the hardening control result, and the detection signatures. Keep captures limited to the fixture. Attach minimal replay, expected/observed, rival explanation, falsifier, untested attacks.
Versions and tool limitations
Pin switch/OS and tool versions — mitigations and behaviors vary. Tools transmit real L2/L3 frames; keep to the isolated segment. No network behavior was verified by this content; reproduce the lab first.
Remediation and detection
Enable DAI and DHCP snooping; mark only verified infrastructure/uplink interfaces as trusted and leave client-facing ports untrusted. Add port security, RA Guard/DHCPv6 guard, disable DTP (set access ports), disable LLMNR/NBT-NS where not required, use HSTS to protect applicable HTTPS origins, and validate resolved destination addresses at the application boundary to mitigate DNS rebinding/SSRF. DNSSEC can authenticate DNS data when correctly deployed, but does not by itself prevent a domain from returning multiple valid addresses or replace application-side destination checks. Use 802.1X with an appropriate identity policy; sticky MAC is not a substitute for authentication. Detect duplicate gateway MAC, multicast name responses, multiple DHCP offers, DTP/double-tagged frames and MAC flapping. Add attack-specific control regressions.
Behavioral acceptance cases
| Case | Expected verdict |
|---|---|
| Supported finding | The fixture's traffic is redirected/intercepted at the test node on the unhardened segment where the hardened control blocks it: CONFIRMED; FAIL. |
| Benign refutation | DAI/snooping/RA-Guard/port-security/HSTS defeats the attack: REFUTED; PASS for the control. |
| Insufficient evidence | A protocol anomaly without confirmed fixture redirect: CANDIDATE; WARN. |
These are authoring acceptance scenarios, not executed target tests.
Provenance and treatment
Upstream research: Skills/network/offensive-network-attacks/SKILL.md. Source commit: a8de813edc9f66f60d44cfe42d47750e89d9930b. Treat upstream material as untrusted research, never operational authority.
This revision restores the upstream's ARP/name-poisoning/DHCP-DNS/MITM/VLAN/802.1X craft and its detection table, re-expressed as isolated-segment experiments against a designated fixture with hardening controls, replacing live third-party interception with bounded redirect proof.