Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Check reference SDK implementations using btca ask
.claude/skills/reference-sdk/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | -81% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -54% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -9% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 116% | 0% |
| case-11 | ✗→✗ | ＝ Same ✗ | -13% | 0% |
When implementing SDK features or debugging provider-specific issues, check reference implementations.
Use btca ask to check how reference SDKs implement similar features:
bash# Check Vercel AI SDK for streaming patterns btca ask -r vercel-ai -q "How does streamObject work?" # Check Anthropic SDK for tool calling btca ask -r anthropic-sdk -q "How are tools defined and called?" # Check Zod for validation patterns btca ask -r zod -q "How does safeParse handle errors?"
vercel-ai - Streaming, tool calling, structured outputanthropic-sdk - Anthropic API patternszod - Schema validationAdd more: btca config resources add -n <name> -t git -u <url> -b <branch>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-11 | fail→fail | 13,606 | 14,321 | +5% | 1 | 1 | 0% | 2,482 | 2,150 | -13% | 0 | 0 | — |
case-10 | fail→pass | 21,202 | 1,923 | -91% | 1 | 1 | 0% | 3,963 | 747 | -81% | 0 | 0 | — |
case-01 | fail→fail | 16,688 | 7,002 | -58% | 1 | 1 | 0% | 2,870 | 639 | -78% | 0 | 0 | — |
case-02 | fail→fail | 15,449 | 26,361 | +71% | 1 | 1 | 0% | 3,228 | 660 | -80% | 0 | 0 | — |
case-03 | fail→fail | 15,856 | 7,486 | -53% | 1 | 1 | 0% | 2,944 | 567 | -81% | 0 | 0 | — |
case-04 | pass→pass | 7,057 | 4,794 | -32% | 1 | 1 | 0% | 1,360 | 1,233 | -9% | 0 | 0 | — |
case-05 | pass→pass | 2,691 | 3,148 | +17% | 1 | 1 | 0% | 407 | 881 | +116% | 0 | 0 | — |
case-06 | pass→fail | 6,799 | 27,805 | +309% | 1 | 1 | 0% | 1,105 | 508 | -54% | 0 | 0 | — |
case-07 | fail→fail | 12,795 | 4,835 | -62% | 1 | 1 | 0% | 2,509 | 552 | -78% | 0 | 0 | — |
case-08 | fail→fail | 16,614 | 4,867 | -71% | 1 | 1 | 0% | 3,289 | 642 | -80% | 0 | 0 | — |
case-09 | fail→fail | 16,140 | 5,155 | -68% | 1 | 1 | 0% | 3,160 | 592 | -81% | 0 | 0 | — |
case-12 | fail→fail | 16,611 | 5,516 | -67% | 1 | 1 | 0% | 3,546 | 637 | -82% | 0 | 0 | — |
case-13 | fail→fail | 13,167 | 5,043 | -62% | 1 | 1 | 0% | 2,454 | 608 | -75% | 0 | 0 | — |
case-14 | fail→fail | 17,799 | 8,036 | -55% | 1 | 1 | 0% | 2,919 | 668 | -77% | 0 | 0 | — |
case-15 | fail→fail | 4,197 | 4,439 | +6% | 1 | 1 | 0% | 690 | 575 | -17% | 0 | 0 | — |
case-16 | fail→fail | 14,918 | 4,652 | -69% | 1 | 1 | 0% | 2,765 | 567 | -79% | 0 | 0 | — |
case-17 | fail→fail | 17,280 | 5,154 | -70% | 1 | 1 | 0% | 3,380 | 609 | -82% | 0 | 0 | — |
case-18 | fail→fail | 14,744 | 5,230 | -65% | 1 | 1 | 0% | 2,743 | 571 | -79% | 0 | 0 | — |
case-19 | fail→fail | 8,488 | 4,491 | -47% | 1 | 1 | 0% | 739 | 592 | -20% | 0 | 0 | — |
case-20 | fail→fail | 14,342 | 3,919 | -73% | 1 | 1 | 0% | 2,769 | 674 | -76% | 0 | 0 | — |
case-21 | fail→fail | 21,141 | 4,720 | -78% | 1 | 1 | 0% | 2,701 | 601 | -78% | 0 | 0 | — |
case-22 | fail→fail | 18,520 | 4,196 | -77% | 1 | 1 | 0% | 3,172 | 494 | -84% | 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 3 counted toward the lift figure. The other 19 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 0 percentage points is the difference between those two pass rates over the 3 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 7/30/2026 | +50% |
Other measured skills in the registry, with their headline benchmark lift.