Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Applies microkernel architecture with minimal core and plugin extensibility. Use when building platforms where third parties extend core functionality.
.claude/skills/athola-architecture-paradigm-microkernel/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -10% | 0% |
These vocabulary items name the concrete tools and abstractions that show up when the paradigm is implemented. They are not required dependencies and they are not part of the skill's tools: frontmatter (which is reserved for Claude Code tool restrictions). Use this list to disambiguate during architecture discussions.
plugin-loader: discovers, validates, and activates plugins at runtimesandbox-executor: runs each plugin in an isolated context with a constrained capability setsdk-generator: produces language-specific SDKs from the kernel's stable interface(registration, capability descriptors, lifecycle hooks), and the permission model.
exists before any third-party plugin development begins.
and gates plugin certification.
published and updated whenever the core plugin contract changes.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 17,604 | 10,988 | -38% | 1 | 1 | 0% | 2,596 | 2,423 | -7% | 0 | 0 | — |
case-06 | pass→pass | 14,428 | 7,403 | -49% | 1 | 1 | 0% | 2,097 | 1,812 | -14% | 0 | 0 | — |
case-01 | fail→pass | 21,965 | 21,472 | -2% | 1 | 1 | 0% | 3,192 | 4,151 | +30% | 0 | 0 | — |
case-02 | fail→pass | 27,073 | 20,826 | -23% | 1 | 1 | 0% | 4,076 | 4,212 | +3% | 0 | 0 | — |
case-03 | fail→fail | 26,437 | 22,354 | -15% | 1 | 1 | 0% | 4,035 | 4,353 | +8% | 0 | 0 | — |
case-04 | pass→pass | 14,993 | 10,945 | -27% | 1 | 1 | 0% | 2,167 | 2,486 | +15% | 0 | 0 | — |
case-07 | pass→pass | 17,800 | 15,232 | -14% | 1 | 1 | 0% | 2,522 | 3,002 | +19% | 0 | 0 | — |
case-08 | pass→pass | 18,152 | 14,097 | -22% | 1 | 1 | 0% | 2,717 | 2,989 | +10% | 0 | 0 | — |
case-09 | fail→pass | 12,251 | 3,414 | -72% | 1 | 1 | 0% | 2,040 | 1,280 | -37% | 0 | 0 | — |
case-10 | fail→pass | 14,991 | 3,273 | -78% | 1 | 1 | 0% | 2,307 | 1,269 | -45% | 0 | 0 | — |
case-11 | pass→pass | 12,196 | 3,608 | -70% | 1 | 1 | 0% | 1,844 | 1,391 | -25% | 0 | 0 | — |
case-12 | pass→pass | 25,874 | 14,497 | -44% | 1 | 1 | 0% | 2,006 | 2,966 | +48% | 0 | 0 | — |
case-13 | pass→pass | 15,587 | 11,477 | -26% | 1 | 1 | 0% | 2,272 | 2,477 | +9% | 0 | 0 | — |
case-14 | pass→pass | 17,343 | 14,708 | -15% | 1 | 1 | 0% | 2,335 | 3,009 | +29% | 0 | 0 | — |
case-15 | pass→pass | 14,403 | 6,772 | -53% | 1 | 1 | 0% | 2,056 | 1,933 | -6% | 0 | 0 | — |
case-16 | pass→pass | 15,629 | 16,304 | +4% | 1 | 1 | 0% | 2,582 | 3,401 | +32% | 0 | 0 | — |
case-17 | pass→pass | 18,528 | 15,389 | -17% | 1 | 1 | 0% | 2,806 | 3,312 | +18% | 0 | 0 | — |
case-18 | pass→fail | 13,123 | 11,324 | -14% | 1 | 1 | 0% | 2,005 | 2,645 | +32% | 0 | 0 | — |
case-19 | pass→pass | 13,912 | 13,352 | -4% | 1 | 1 | 0% | 2,245 | 2,946 | +31% | 0 | 0 | — |
case-20 | fail→pass | 10,714 | 4,158 | -61% | 1 | 1 | 0% | 1,596 | 1,438 | -10% | 0 | 0 | — |
case-21 | pass→pass | 18,367 | 10,208 | -44% | 1 | 1 | 0% | 2,799 | 2,390 | -15% | 0 | 0 | — |
case-22 | pass→pass | 9,121 | 6,249 | -31% | 1 | 1 | 0% | 1,399 | 1,712 | +22% | 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 +18 percentage points is the difference between those two pass rates over the 22 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.