Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →USB device and host stack implementation expertise
.claude/skills/a5c-ai-usb-stack/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 8% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 28% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 92% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 180% | 0% |
This skill provides USB device and host stack implementation expertise for embedded systems, covering descriptor generation, device class implementation, and protocol debugging.
device-driver-development.js - USB driver implementationbootloader-implementation.js - USB DFU bootloaderhw-sw-interface-specification.js - USB interface specificationThis skill is invoked when tasks require:
ctusb_desc_interface_t cdc_interface = { .bLength = sizeof(tusb_desc_interface_t), .bDescriptorType = TUSB_DESC_INTERFACE, .bInterfaceNumber = 0, .bAlternateSetting = 0, .bNumEndpoints = 2, .bInterfaceClass = TUSB_CLASS_CDC, .bInterfaceSubClass = CDC_COMM_SUBCLASS_ACM, .bInterfaceProtocol = CDC_COMM_PROTOCOL_NONE, .iInterface = 0 };
cuint8_t const hid_report_desc[] = { HID_USAGE_PAGE(HID_USAGE_PAGE_DESKTOP), HID_USAGE(HID_USAGE_DESKTOP_MOUSE), HID_COLLECTION(HID_COLLECTION_APPLICATION), HID_USAGE(HID_USAGE_DESKTOP_POINTER), HID_COLLECTION(HID_COLLECTION_PHYSICAL), // ... buttons and axes HID_COLLECTION_END, HID_COLLECTION_END };
yamlusb: vid: 0x1234 pid: 0x5678 device_class: cdc | hid | msc | dfu | composite speed: full | high endpoints: - ep1_in: bulk, 64 - ep1_out: bulk, 64
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 13,040 | 12,597 | -3% | 1 | 1 | 0% | 2,562 | 2,673 | +4% | 0 | 0 | — |
case-02 | fail→pass | 16,203 | 16,201 | -0% | 1 | 1 | 0% | 3,651 | 4,372 | +20% | 0 | 0 | — |
case-03 | pass→pass | 43,075 | 12,738 | -70% | 1 | 1 | 0% | 2,740 | 2,952 | +8% | 0 | 0 | — |
case-04 | pass→pass | 20,649 | 16,750 | -19% | 1 | 1 | 0% | 3,335 | 4,267 | +28% | 0 | 0 | — |
case-05 | pass→pass | 5,634 | 23,901 | +324% | 1 | 1 | 0% | 925 | 1,773 | +92% | 0 | 0 | — |
case-06 | pass→pass | 3,779 | 6,307 | +67% | 1 | 1 | 0% | 713 | 1,998 | +180% | 0 | 0 | — |
case-07 | pass→pass | 5,462 | 9,301 | +70% | 1 | 1 | 0% | 891 | 2,375 | +167% | 0 | 0 | — |
case-08 | pass→pass | 5,648 | 7,115 | +26% | 1 | 1 | 0% | 985 | 2,108 | +114% | 0 | 0 | — |
case-09 | pass→pass | 5,375 | 5,573 | +4% | 1 | 1 | 0% | 1,029 | 1,770 | +72% | 0 | 0 | — |
case-10 | pass→pass | 16,628 | 6,309 | -62% | 1 | 1 | 0% | 1,618 | 1,908 | +18% | 0 | 0 | — |
case-11 | pass→pass | 6,251 | 8,864 | +42% | 1 | 1 | 0% | 1,141 | 2,020 | +77% | 0 | 0 | — |
case-12 | pass→pass | 6,445 | 6,275 | -3% | 1 | 1 | 0% | 1,015 | 1,953 | +92% | 0 | 0 | — |
case-13 | pass→pass | 10,863 | 11,808 | +9% | 1 | 1 | 0% | 1,857 | 2,745 | +48% | 0 | 0 | — |
case-14 | pass→pass | 16,650 | 18,690 | +12% | 1 | 1 | 0% | 2,616 | 3,814 | +46% | 0 | 0 | — |
case-15 | fail→fail | 5,422 | 4,276 | -21% | 1 | 1 | 0% | 969 | 1,436 | +48% | 0 | 0 | — |
case-16 | pass→pass | 12,043 | 11,419 | -5% | 1 | 1 | 0% | 1,815 | 2,685 | +48% | 0 | 0 | — |
case-17 | fail→fail | 14,934 | 15,593 | +4% | 1 | 1 | 0% | 2,954 | 3,794 | +28% | 0 | 0 | — |
case-18 | pass→pass | 9,860 | 8,481 | -14% | 1 | 1 | 0% | 1,995 | 2,591 | +30% | 0 | 0 | — |
case-19 | pass→pass | 2,824 | 3,248 | +15% | 1 | 1 | 0% | 472 | 1,209 | +156% | 0 | 0 | — |
case-20 | pass→pass | 19,417 | 19,347 | -0% | 1 | 1 | 0% | 3,160 | 4,048 | +28% | 0 | 0 | — |
case-21 | pass→pass | 7,448 | 6,756 | -9% | 1 | 1 | 0% | 1,575 | 1,799 | +14% | 0 | 0 | — |
case-22 | pass→pass | 14,551 | 14,509 | -0% | 1 | 1 | 0% | 2,386 | 3,278 | +37% | 0 | 0 | — |
case-23 | pass→pass | 9,317 | 5,593 | -40% | 1 | 1 | 0% | 1,898 | 1,815 | -4% | 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. 23 cases were attempted. The headline lift of +4 percentage points is the difference between those two pass rates over the 23 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.