Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Deep integration with JTAG/SWD debug probes for hardware-level debugging and flash programming
.claude/skills/a5c-ai-jtag-swd-debug/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 81% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 16% | 0% |
This skill provides deep integration with JTAG/SWD debug probes for hardware-level debugging, flash programming, and real-time firmware analysis. It enables direct hardware interaction through industry-standard debug interfaces.
hardware-bring-up.js - Initial board validation and debug setupbootloader-implementation.js - Bootloader debugging and flash operationsdevice-driver-development.js - Driver debugging with hardware accessfield-diagnostics.js - Field debugging and analysisThis skill is invoked when tasks require:
yamldebug_probe: type: jlink | stlink | openocd | cmsis-dap interface: swd | jtag speed: 4000 # kHz target: cortex-m4 | cortex-m7 | etc.
bashopenocd -f interface/stlink.cfg -f target/stm32f4x.cfg \ -c "program firmware.elf verify reset exit"
bashJLinkExe -device STM32F407VG -if SWD -speed 4000 \ -CommanderScript read_memory.jlink
bashopenocd -f board/stm32f4discovery.cfg \ -c "init; halt; reg; resume; exit"
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 13,197 | 5,200 | -61% | 1 | 1 | 0% | 917 | 1,658 | +81% | 0 | 0 | — |
case-01 | pass→pass | 6,138 | 7,477 | +22% | 1 | 1 | 0% | 1,172 | 1,356 | +16% | 0 | 0 | — |
case-03 | fail→pass | 8,911 | 3,991 | -55% | 1 | 1 | 0% | 1,286 | 1,244 | -3% | 0 | 0 | — |
case-04 | pass→pass | 3,946 | 2,447 | -38% | 1 | 1 | 0% | 645 | 970 | +50% | 0 | 0 | — |
case-05 | fail→pass | 9,330 | 3,119 | -67% | 1 | 1 | 0% | 1,526 | 927 | -39% | 0 | 0 | — |
case-06 | fail→pass | 6,847 | 2,187 | -68% | 1 | 1 | 0% | 1,109 | 909 | -18% | 0 | 0 | — |
case-07 | pass→pass | 7,644 | 2,026 | -73% | 1 | 1 | 0% | 1,038 | 948 | -9% | 0 | 0 | — |
case-08 | pass→pass | 4,851 | 2,685 | -45% | 1 | 1 | 0% | 900 | 1,009 | +12% | 0 | 0 | — |
case-09 | pass→pass | 12,878 | 2,354 | -82% | 1 | 1 | 0% | 1,910 | 949 | -50% | 0 | 0 | — |
case-10 | pass→pass | 26,744 | 2,489 | -91% | 1 | 1 | 0% | 4,302 | 946 | -78% | 0 | 0 | — |
case-11 | fail→pass | 18,999 | 25,080 | +32% | 1 | 1 | 0% | 3,147 | 3,637 | +16% | 0 | 0 | — |
case-12 | pass→pass | 10,506 | 7,522 | -28% | 1 | 1 | 0% | 1,650 | 1,848 | +12% | 0 | 0 | — |
case-13 | pass→pass | 21,075 | 8,198 | -61% | 1 | 1 | 0% | 2,907 | 2,028 | -30% | 0 | 0 | — |
case-14 | pass→pass | 18,139 | 15,999 | -12% | 1 | 1 | 0% | 3,010 | 2,908 | -3% | 0 | 0 | — |
case-15 | pass→fail | 19,446 | 19,206 | -1% | 1 | 1 | 0% | 3,046 | 3,531 | +16% | 0 | 0 | — |
case-16 | pass→pass | 26,173 | 5,288 | -80% | 1 | 1 | 0% | 4,013 | 1,513 | -62% | 0 | 0 | — |
case-17 | pass→pass | 19,778 | 5,101 | -74% | 1 | 1 | 0% | 3,160 | 1,509 | -52% | 0 | 0 | — |
case-18 | fail→pass | 16,957 | 3,232 | -81% | 1 | 1 | 0% | 2,918 | 1,173 | -60% | 0 | 0 | — |
case-19 | pass→pass | 16,226 | 6,515 | -60% | 1 | 1 | 0% | 2,166 | 1,551 | -28% | 0 | 0 | — |
case-20 | pass→pass | 18,409 | 12,346 | -33% | 1 | 1 | 0% | 2,556 | 2,487 | -3% | 0 | 0 | — |
case-21 | pass→pass | 21,530 | 19,881 | -8% | 1 | 1 | 0% | 3,103 | 3,815 | +23% | 0 | 0 | — |
case-22 | pass→pass | 14,416 | 12,215 | -15% | 1 | 1 | 0% | 2,424 | 2,734 | +13% | 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 +23 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.