Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Async REPL Protocol
.claude/skills/async-repl-protocol/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -23% | 0% |
When working with Agentica's async REPL harness for testing.
await for Future-returning toolspythoncontent = await view_file(path) # NOT view_file(path) answer = await ask_memory("...")
Compute AND return in ONE block. Multiple blocks means only first executes.
python# GOOD: Single block content = await view_file(path) return any(c.isdigit() for c in content) # BAD: Split blocks (second block never runs) content = await view_file(path)
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 5,969 | 2,711 | -55% | 1 | 1 | 0% | 1,156 | 629 | -46% | 0 | 0 | — |
case-02 | fail→pass | 2,356 | 2,050 | -13% | 1 | 1 | 0% | 397 | 490 | +23% | 0 | 0 | — |
case-03 | fail→pass | 7,277 | 5,950 | -18% | 1 | 1 | 0% | 1,209 | 1,208 | -0% | 0 | 0 | — |
case-04 | fail→pass | 4,420 | 3,130 | -29% | 1 | 1 | 0% | 883 | 686 | -22% | 0 | 0 | — |
case-05 | fail→pass | 5,300 | 3,208 | -39% | 1 | 1 | 0% | 975 | 752 | -23% | 0 | 0 | — |
case-06 | fail→pass | 7,797 | 2,050 | -74% | 1 | 1 | 0% | 1,425 | 507 | -64% | 0 | 0 | — |
case-07 | fail→pass | 3,871 | 4,803 | +24% | 1 | 1 | 0% | 732 | 1,054 | +44% | 0 | 0 | — |
case-08 | fail→fail | 5,446 | 5,924 | +9% | 1 | 1 | 0% | 377 | 1,272 | +237% | 0 | 0 | — |
case-09 | fail→pass | 3,588 | 3,535 | -1% | 1 | 1 | 0% | 660 | 786 | +19% | 0 | 0 | — |
case-10 | fail→pass | 7,032 | 1,953 | -72% | 1 | 1 | 0% | 1,276 | 475 | -63% | 0 | 0 | — |
case-11 | fail→pass | 12,844 | 3,042 | -76% | 1 | 1 | 0% | 495 | 777 | +57% | 0 | 0 | — |
case-12 | fail→pass | 4,719 | 4,911 | +4% | 1 | 1 | 0% | 868 | 993 | +14% | 0 | 0 | — |
case-13 | fail→fail | 3,761 | 3,598 | -4% | 1 | 1 | 0% | 621 | 818 | +32% | 0 | 0 | — |
case-14 | fail→pass | 6,518 | 5,510 | -15% | 1 | 1 | 0% | 1,433 | 1,289 | -10% | 0 | 0 | — |
case-15 | fail→pass | 5,227 | 4,687 | -10% | 1 | 1 | 0% | 273 | 951 | +248% | 0 | 0 | — |
case-21 | pass→pass | 8,331 | 3,971 | -52% | 1 | 1 | 0% | 1,684 | 914 | -46% | 0 | 0 | — |
case-16 | fail→pass | 3,920 | 4,644 | +18% | 1 | 1 | 0% | 778 | 991 | +27% | 0 | 0 | — |
case-17 | fail→fail | 4,839 | 4,411 | -9% | 1 | 1 | 0% | 932 | 915 | -2% | 0 | 0 | — |
case-18 | fail→pass | 3,141 | 3,944 | +26% | 1 | 1 | 0% | 550 | 875 | +59% | 0 | 0 | — |
case-19 | fail→pass | 8,767 | 2,728 | -69% | 1 | 1 | 0% | 1,408 | 609 | -57% | 0 | 0 | — |
case-20 | pass→pass | 4,417 | 2,910 | -34% | 1 | 1 | 0% | 873 | 599 | -31% | 0 | 0 | — |
case-22 | pass→pass | 4,882 | 4,586 | -6% | 1 | 1 | 0% | 949 | 884 | -7% | 0 | 0 | — |
case-23 | pass→pass | 5,089 | 3,999 | -21% | 1 | 1 | 0% | 993 | 929 | -6% | 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, and 20 counted toward the lift figure. The other 3 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 +70 percentage points is the difference between those two pass rates over the 20 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/31/2026 | +64% |
Other measured skills in the registry, with their headline benchmark lift.