Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create a validated, authorized backend endpoint with error mapping and tests.
.claude/skills/velcrafting-endpoint-scaffold/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-11 | ✓→✗ | ▼ Worse | -8% | 0% |
| case-22 | ✓→✗ | ▼ Worse | 74% | 0% |
| case-01 | ✓→✓ | = Same ✓ | 67% | 0% |
Add or extend an API endpoint in a way that is:
This skill is the default entry point for backend feature exposure.
<repo>/REPO_PROFILE.jsonbackend/domain-logic-module)backend/persistence-layer-change)api/contract-update if needed)1) Discover repo endpoint conventions:
REPO_PROFILE.json if present2) Define contract explicitly:
3) Enforce authorization:
4) Call domain logic via a module boundary:
5) Map errors to a stable taxonomy:
6) Add tests at the highest stable layer available:
7) If this endpoint introduces branching, retries, async coordination, or multi-step behavior, recommend system/state-machine-mapper and pause until the behavior is modeled or explicitly waived.
shared/error-taxonomy before shipping.backend/authz-policy or $decision-capture.meta/ask-questions-if-underspecified or api/contract-update.backend/domain-logic-module.block completion and recommend system/state-machine-mapper.
Log:
backend/endpoint-scaffold<method> <path> (if known)public | authenticated | role | resourceunit | integration | nonesuccess | partial | blocked| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 11,102 | 12,926 | +16% | 1 | 1 | 0% | 2,026 | 3,389 | +67% | 0 | 0 | — |
case-02 | fail→pass | 21,071 | 8,519 | -60% | 1 | 1 | 0% | 3,992 | 2,313 | -42% | 0 | 0 | — |
case-03 | pass→pass | 12,509 | 14,109 | +13% | 1 | 1 | 0% | 2,616 | 3,964 | +52% | 0 | 0 | — |
case-04 | pass→pass | 15,234 | 14,939 | -2% | 1 | 1 | 0% | 3,070 | 4,042 | +32% | 0 | 0 | — |
case-05 | pass→pass | 9,785 | 5,669 | -42% | 1 | 1 | 0% | 1,680 | 1,744 | +4% | 0 | 0 | — |
case-06 | pass→pass | 9,420 | 3,111 | -67% | 1 | 1 | 0% | 1,592 | 1,311 | -18% | 0 | 0 | — |
case-07 | pass→pass | 8,182 | 3,875 | -53% | 1 | 1 | 0% | 1,390 | 1,415 | +2% | 0 | 0 | — |
case-08 | pass→pass | 15,362 | 9,518 | -38% | 1 | 1 | 0% | 2,830 | 2,408 | -15% | 0 | 0 | — |
case-09 | pass→pass | 11,006 | 12,393 | +13% | 1 | 1 | 0% | 2,167 | 3,147 | +45% | 0 | 0 | — |
case-10 | pass→pass | 13,255 | 9,669 | -27% | 1 | 1 | 0% | 2,479 | 2,253 | -9% | 0 | 0 | — |
case-11 | pass→fail | 10,380 | 5,541 | -47% | 1 | 1 | 0% | 1,833 | 1,683 | -8% | 0 | 0 | — |
case-12 | pass→pass | 9,128 | 6,203 | -32% | 1 | 1 | 0% | 1,585 | 1,911 | +21% | 0 | 0 | — |
case-13 | fail→pass | 15,266 | 2,287 | -85% | 1 | 1 | 0% | 2,602 | 1,207 | -54% | 0 | 0 | — |
case-14 | pass→pass | 12,660 | 14,524 | +15% | 1 | 1 | 0% | 2,509 | 3,998 | +59% | 0 | 0 | — |
case-15 | pass→pass | 18,023 | 15,538 | -14% | 1 | 1 | 0% | 3,633 | 4,102 | +13% | 0 | 0 | — |
case-16 | pass→pass | 11,663 | 10,213 | -12% | 1 | 1 | 0% | 2,053 | 2,338 | +14% | 0 | 0 | — |
case-17 | pass→pass | 9,788 | 4,683 | -52% | 1 | 1 | 0% | 1,764 | 1,646 | -7% | 0 | 0 | — |
case-18 | pass→pass | 8,532 | 1,841 | -78% | 1 | 1 | 0% | 1,552 | 1,088 | -30% | 0 | 0 | — |
case-19 | fail→fail | 11,261 | 6,428 | -43% | 1 | 1 | 0% | 1,897 | 1,898 | +0% | 0 | 0 | — |
case-20 | pass→pass | 6,668 | 12,676 | +90% | 1 | 1 | 0% | 1,373 | 3,526 | +157% | 0 | 0 | — |
case-21 | pass→pass | 12,086 | 12,919 | +7% | 1 | 1 | 0% | 2,193 | 3,287 | +50% | 0 | 0 | — |
case-22 | pass→fail | 15,406 | 22,937 | +49% | 1 | 1 | 0% | 3,527 | 6,133 | +74% | 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 0 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are 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.