Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Expert skill for load balancer configuration, algorithms, and high availability design
.claude/skills/a5c-ai-load-balancer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 42% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 73% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 61% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 60% | 0% |
Expert skill for load balancer configuration, algorithm implementation, and high availability design.
HAProxy - High-performance TCP/HTTP load balancerNGINX - Web server and reverse proxykeepalived - HA and failoverLVS/IPVS - Linux Virtual Serverhaproxyctl - HAProxy managementhaproxybackend web_servers balance roundrobin option httpchk GET /health http-check expect status 200 server web1 10.0.0.1:80 check weight 100 server web2 10.0.0.2:80 check weight 100 backup
nginxupstream backend { least_conn; server 10.0.0.1:8080 weight=5; server 10.0.0.2:8080; keepalive 32; }
bashecho "show stat" | socat stdio /var/run/haproxy/admin.sock
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 5,754 | 6,685 | +16% | 1 | 1 | 0% | 889 | 1,261 | +42% | 0 | 0 | — |
case-02 | pass→pass | 3,416 | 4,814 | +41% | 1 | 1 | 0% | 743 | 1,289 | +73% | 0 | 0 | — |
case-03 | pass→pass | 3,716 | 4,100 | +10% | 1 | 1 | 0% | 840 | 1,352 | +61% | 0 | 0 | — |
case-04 | pass→pass | 3,361 | 2,861 | -15% | 1 | 1 | 0% | 641 | 1,023 | +60% | 0 | 0 | — |
case-05 | pass→pass | 2,924 | 2,947 | +1% | 1 | 1 | 0% | 601 | 1,202 | +100% | 0 | 0 | — |
case-06 | pass→pass | 7,277 | 7,353 | +1% | 1 | 1 | 0% | 1,345 | 1,834 | +36% | 0 | 0 | — |
case-07 | pass→pass | 4,548 | 5,089 | +12% | 1 | 1 | 0% | 917 | 1,557 | +70% | 0 | 0 | — |
case-08 | pass→pass | 2,120 | 3,946 | +86% | 1 | 1 | 0% | 353 | 1,233 | +249% | 0 | 0 | — |
case-09 | pass→pass | 3,819 | 4,637 | +21% | 1 | 1 | 0% | 803 | 1,404 | +75% | 0 | 0 | — |
case-10 | fail→pass | 7,120 | 5,319 | -25% | 1 | 1 | 0% | 1,220 | 1,376 | +13% | 0 | 0 | — |
case-11 | pass→pass | 2,835 | 4,659 | +64% | 1 | 1 | 0% | 522 | 1,406 | +169% | 0 | 0 | — |
case-12 | pass→pass | 5,662 | 5,497 | -3% | 1 | 1 | 0% | 1,075 | 1,282 | +19% | 0 | 0 | — |
case-13 | pass→pass | 4,131 | 5,454 | +32% | 1 | 1 | 0% | 858 | 1,602 | +87% | 0 | 0 | — |
case-14 | pass→pass | 8,408 | 4,846 | -42% | 1 | 1 | 0% | 1,600 | 1,419 | -11% | 0 | 0 | — |
case-15 | pass→pass | 6,417 | 5,046 | -21% | 1 | 1 | 0% | 1,143 | 1,529 | +34% | 0 | 0 | — |
case-16 | pass→pass | 3,843 | 3,904 | +2% | 1 | 1 | 0% | 630 | 1,177 | +87% | 0 | 0 | — |
case-17 | pass→pass | 5,258 | 6,216 | +18% | 1 | 1 | 0% | 1,126 | 1,598 | +42% | 0 | 0 | — |
case-18 | pass→pass | 6,705 | 5,506 | -18% | 1 | 1 | 0% | 1,305 | 1,393 | +7% | 0 | 0 | — |
case-19 | pass→pass | 4,163 | 7,756 | +86% | 1 | 1 | 0% | 749 | 1,709 | +128% | 0 | 0 | — |
case-20 | pass→pass | 4,005 | 3,757 | -6% | 1 | 1 | 0% | 795 | 1,213 | +53% | 0 | 0 | — |
case-21 | pass→pass | 3,003 | 4,207 | +40% | 1 | 1 | 0% | 642 | 1,278 | +99% | 0 | 0 | — |
case-22 | pass→pass | 4,691 | 4,343 | -7% | 1 | 1 | 0% | 768 | 1,126 | +47% | 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. The headline lift of +5 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.