Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Adversaries may attempt to cause a denial of service (DoS) by reflecting a high-volume of network traffic to a target.
.claude/skills/cyberstrikeus-t1498-002-reflection-amplification/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 111% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-14 | ✓→✗ | ▼ Worse | 67% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 100% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | -18% | 0% |
> Sub-technique of: T1498
Adversaries may attempt to cause a denial of service (DoS) by reflecting a high-volume of network traffic to a target. This type of Network DoS takes advantage of a third-party server intermediary that hosts and will respond to a given spoofed source IP address. This third-party server is commonly termed a reflector. An adversary accomplishes a reflection attack by sending packets to reflectors with the spoofed address of the victim. Similar to Direct Network Floods, more than one system may be used to conduct the attack, or a botnet may be used. Likewise, one or more reflectors may be used to focus traffic on the target. This Network DoS attack may also reduce the availability and functionality of the targeted system(s) and network.
Reflection attacks often take advantage of protocols with larger responses than requests in order to amplify their traffic, commonly known as a Reflection Amplification attack. Adversaries may be able to generate an increase in volume of attack traffic that is several orders of magnitude greater than the requests sent to the amplifiers. The extent of this increase will depending upon many variables, such as the protocol in question, the technique used, and the amplifying servers that actually produce the amplification in attack volume. Two prominent protocols that have enabled Reflection Amplification Floods are DNS and NTP, though the use of several others in the wild have been documented. In particular, the memcache protocol showed itself to be a powerful protocol, with amplification sizes up to 51,200 times the requesting packet.
Platforms: Windows, IaaS, Linux, macOS
> Note: No Atomic Red Team tests available for this technique. See Atomic Red Team GitHub for updates.
When flood volumes exceed the capacity of the network connection being targeted, it is typically necessary to intercept the incoming traffic upstream to filter out the attack traffic from the legitimate traffic. Such defenses can be provided by the hosting Internet Service Provider (ISP) or by a 3rd party such as a Content Delivery Network (CDN) or providers specializing in DoS mitigations.
Depending on flood volume, on-premises filtering may be possible by blocking source addresses sourcing the attack, blocking ports that are being targeted, or blocking protocols being used for transport.
As immediate response may require rapid engagement of 3rd parties, analyze the risk associated to critical resources being affected by Network DoS attacks and create a disaster recovery plan/business continuity plan to respond to incidents.
| Finding | Severity | Impact | | --------------------------------------------- | -------- | ------ | | Reflection Amplification technique applicable | Low | Impact |
| CWE ID | Title | | ------- | --------------------------------- | | CWE-400 | Uncontrolled Resource Consumption |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 31,196 | 36,183 | +16% | 1 | 1 | 0% | 3,829 | 6,321 | +65% | 0 | 0 | — |
case-02 | fail→fail | 21,538 | 28,808 | +34% | 1 | 1 | 0% | 3,452 | 5,079 | +47% | 0 | 0 | — |
case-03 | fail→pass | 25,566 | 9,828 | -62% | 1 | 1 | 0% | 887 | 1,868 | +111% | 0 | 0 | — |
case-04 | pass→pass | 9,526 | 2,225 | -77% | 1 | 1 | 0% | 793 | 1,586 | +100% | 0 | 0 | — |
case-05 | pass→pass | 19,943 | 4,842 | -76% | 1 | 1 | 0% | 2,434 | 1,997 | -18% | 0 | 0 | — |
case-06 | fail→pass | 15,290 | 13,473 | -12% | 1 | 1 | 0% | 1,698 | 2,679 | +58% | 0 | 0 | — |
case-07 | pass→pass | 15,191 | 10,403 | -32% | 1 | 1 | 0% | 1,635 | 2,111 | +29% | 0 | 0 | — |
case-08 | pass→pass | 18,989 | 12,409 | -35% | 1 | 1 | 0% | 2,070 | 2,415 | +17% | 0 | 0 | — |
case-09 | fail→fail | 23,360 | 14,898 | -36% | 1 | 1 | 0% | 2,506 | 3,325 | +33% | 0 | 0 | — |
case-10 | pass→pass | 13,078 | 9,111 | -30% | 1 | 1 | 0% | 1,342 | 1,903 | +42% | 0 | 0 | — |
case-11 | pass→pass | 19,317 | 7,926 | -59% | 1 | 1 | 0% | 2,358 | 1,720 | -27% | 0 | 0 | — |
case-12 | pass→pass | 9,192 | 7,281 | -21% | 1 | 1 | 0% | 1,357 | 1,561 | +15% | 0 | 0 | — |
case-13 | pass→pass | 14,174 | 8,212 | -42% | 1 | 1 | 0% | 1,262 | 1,627 | +29% | 0 | 0 | — |
case-14 | pass→fail | 22,645 | 26,152 | +15% | 1 | 1 | 0% | 2,489 | 4,169 | +67% | 0 | 0 | — |
case-15 | pass→pass | 17,307 | 21,626 | +25% | 1 | 1 | 0% | 2,561 | 3,667 | +43% | 0 | 0 | — |
case-16 | pass→pass | 12,132 | 2,308 | -81% | 1 | 1 | 0% | 1,071 | 1,575 | +47% | 0 | 0 | — |
case-21 | pass→pass | 20,972 | 20,275 | -3% | 1 | 1 | 0% | 2,197 | 3,529 | +61% | 0 | 0 | — |
case-17 | pass→pass | 13,184 | 15,125 | +15% | 1 | 1 | 0% | 2,087 | 2,806 | +34% | 0 | 0 | — |
case-18 | pass→pass | 14,859 | 3,732 | -75% | 1 | 1 | 0% | 1,567 | 1,776 | +13% | 0 | 0 | — |
case-19 | pass→pass | 6,738 | 3,547 | -47% | 1 | 1 | 0% | 912 | 1,800 | +97% | 0 | 0 | — |
case-20 | pass→pass | 15,657 | 12,287 | -22% | 1 | 1 | 0% | 1,750 | 3,036 | +73% | 0 | 0 | — |
case-22 | fail→fail | 11,213 | 10,746 | -4% | 1 | 1 | 0% | 1,067 | 2,245 | +110% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +5 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.