Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Design, implement, and evaluate risk-based software tests across unit, integration, contract, end-to-end, and regression layers. Use when adding tests, reproducing bugs, improving coverage, diagnosing flaky tests, or defining a test strategy; do not use to change production behavior unless the user also requests implementation.
.claude/skills/hashgraph-online-test-software/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -4% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -4% | 0% |
boundary under test.
files, snapshots, browsers, or external sandboxes.
services without explicit authorization and a verified isolation strategy.
Read references/test-strategy.md when choosing test layers or specialized methods, investigating flakes, or planning cross-service coverage.
Build deterministic fixtures, implement the smallest high-value set, and run focused tests repeatedly before broader required checks. Separate product, test, and environment failures.
For a strategy request, return:
For implementation, report focused and broader checks separately.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-20 | pass→pass | 13,497 | 12,378 | -8% | 1 | 1 | 0% | 1,293 | 1,760 | +36% | 0 | 0 | — |
case-01 | fail→pass | 45,969 | 30,496 | -34% | 1 | 1 | 0% | 6,744 | 4,462 | -34% | 0 | 0 | — |
case-02 | pass→pass | 20,142 | 18,898 | -6% | 1 | 1 | 0% | 2,427 | 2,218 | -9% | 0 | 0 | — |
case-03 | pass→pass | 21,685 | 13,092 | -40% | 1 | 1 | 0% | 3,152 | 2,748 | -13% | 0 | 0 | — |
case-04 | fail→pass | 42,995 | 35,973 | -16% | 1 | 1 | 0% | 4,980 | 4,357 | -13% | 0 | 0 | — |
case-05 | pass→pass | 13,022 | 10,239 | -21% | 1 | 1 | 0% | 1,783 | 2,144 | +20% | 0 | 0 | — |
case-06 | pass→pass | 22,179 | 22,945 | +3% | 1 | 1 | 0% | 2,516 | 3,105 | +23% | 0 | 0 | — |
case-07 | pass→pass | 15,882 | 13,545 | -15% | 1 | 1 | 0% | 1,724 | 1,670 | -3% | 0 | 0 | — |
case-08 | pass→pass | 26,409 | 19,946 | -24% | 1 | 1 | 0% | 2,880 | 2,723 | -5% | 0 | 0 | — |
case-09 | pass→pass | 17,002 | 5,329 | -69% | 1 | 1 | 0% | 1,826 | 1,193 | -35% | 0 | 0 | — |
case-10 | fail→pass | 17,553 | 16,538 | -6% | 1 | 1 | 0% | 2,378 | 2,271 | -4% | 0 | 0 | — |
case-19 | pass→pass | 17,065 | 12,945 | -24% | 1 | 1 | 0% | 1,776 | 1,503 | -15% | 0 | 0 | — |
case-11 | pass→pass | 13,557 | 11,666 | -14% | 1 | 1 | 0% | 2,342 | 1,980 | -15% | 0 | 0 | — |
case-12 | pass→pass | 16,476 | 13,154 | -20% | 1 | 1 | 0% | 2,389 | 2,560 | +7% | 0 | 0 | — |
case-13 | pass→pass | 14,319 | 18,479 | +29% | 1 | 1 | 0% | 2,295 | 2,448 | +7% | 0 | 0 | — |
case-14 | pass→pass | 22,730 | 14,953 | -34% | 1 | 1 | 0% | 2,349 | 1,752 | -25% | 0 | 0 | — |
case-15 | pass→pass | 19,445 | 9,360 | -52% | 1 | 1 | 0% | 1,904 | 958 | -50% | 0 | 0 | — |
case-16 | pass→pass | 12,679 | 6,896 | -46% | 1 | 1 | 0% | 2,082 | 1,237 | -41% | 0 | 0 | — |
case-17 | fail→pass | 40,294 | 28,557 | -29% | 1 | 1 | 0% | 5,621 | 3,807 | -32% | 0 | 0 | — |
case-18 | pass→pass | 15,598 | 21,617 | +39% | 1 | 1 | 0% | 2,192 | 3,081 | +41% | 0 | 0 | — |
case-21 | fail→pass | 19,370 | 17,306 | -11% | 1 | 1 | 0% | 2,252 | 2,164 | -4% | 0 | 0 | — |
case-22 | pass→pass | 18,745 | 31,965 | +71% | 1 | 1 | 0% | 2,826 | 3,505 | +24% | 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 +23 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.