Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the user explicitly asks to build, refresh, onboard, or maintain a MaxCompute semantic package.
.claude/skills/aliyun-build/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -26% | 0% |
Use this only for explicit build / refresh / onboarding / maintenance tasks. Do not load this workflow while answering a data question.
mcs -f json doctor only if setup is unclear.mcs build with the user-requested profile/schema/table overrides.mcs skill get enrich for the full review workflow.bash mcs package propose --from-suggestions
bash mcs -f json status --tables Look at has_ai_context and columns_with_description per table.
for ambiguous columns (names shared across tables). If you find a build suggestion with the wrong role/subtype, include the corrected role / dim_type / agg / id_type in the same YAML. Then propose: bash mcs package propose --from-stdin <<'EOF' tables:
ai_context: "Each row is one driver's race result in a single Grand Prix." columns: points: {role: measure, agg: SUM, description: "Points scored in this single race (0-50 scale)."} position: {role: dimension, dim_type: ordinal, description: "Finishing position in this race (1=winner)."}
ai_context: "Each row is one driver's cumulative championship standing after a race." columns: points: {role: measure, agg: SUM, description: "Cumulative season championship points after this race."} position: {role: dimension, dim_type: ordinal, description: "Championship standing rank after this race (1=leader)."} EOF
bash mcs package list-proposals mcs package show-proposal <id>
bash mcs package apply <id> mcs package reject <id> --reason "..."
bash mcs -f json status --tables Every table should have has_ai_context: true. Tables with shared-name columns should have columns_with_description > 0.
results = per-race outcome vs driverstandings = cumulative championship rank). Without it the query agent falls back to column-name heuristics which fail when tables share column names.
(e.g. results.points = "race points" vs driverstandings.points = "season points"). Table-level ai_context says what a row is; column-level description says what this column means in this table.
Write both from your domain knowledge — what entity each row represents, what each column means in context. This is general semantic metadata.
mcs build is a persistent semantic-package maintenance operation. It can take minutes and should never be used as a fallback for a query flow.
See references/build.md for flags and maintenance details.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 27,863 | 5,095 | -82% | 1 | 1 | 0% | 395 | 988 | +150% | 0 | 0 | — |
case-02 | fail→fail | 7,156 | 4,968 | -31% | 1 | 1 | 0% | 388 | 993 | +156% | 0 | 0 | — |
case-03 | fail→fail | 6,450 | 5,295 | -18% | 1 | 1 | 0% | 411 | 1,040 | +153% | 0 | 0 | — |
case-04 | fail→pass | 10,411 | 2,688 | -74% | 1 | 1 | 0% | 1,472 | 1,275 | -13% | 0 | 0 | — |
case-05 | fail→pass | 8,629 | 4,961 | -43% | 1 | 1 | 0% | 1,215 | 1,631 | +34% | 0 | 0 | — |
case-06 | fail→pass | 9,881 | 2,486 | -75% | 1 | 1 | 0% | 1,607 | 1,204 | -25% | 0 | 0 | — |
case-07 | fail→pass | 5,493 | 1,745 | -68% | 1 | 1 | 0% | 910 | 1,063 | +17% | 0 | 0 | — |
case-08 | fail→pass | 9,212 | 1,760 | -81% | 1 | 1 | 0% | 1,419 | 1,056 | -26% | 0 | 0 | — |
case-09 | fail→pass | 9,973 | 1,251 | -87% | 1 | 1 | 0% | 1,667 | 995 | -40% | 0 | 0 | — |
case-10 | fail→pass | 13,341 | 2,058 | -85% | 1 | 1 | 0% | 2,233 | 1,178 | -47% | 0 | 0 | — |
case-11 | fail→pass | 9,354 | 1,817 | -81% | 1 | 1 | 0% | 1,545 | 1,116 | -28% | 0 | 0 | — |
case-12 | pass→pass | 7,576 | 2,803 | -63% | 1 | 1 | 0% | 1,196 | 1,406 | +18% | 0 | 0 | — |
case-13 | fail→pass | 11,721 | 3,038 | -74% | 1 | 1 | 0% | 2,005 | 1,366 | -32% | 0 | 0 | — |
case-14 | fail→pass | 7,444 | 2,283 | -69% | 1 | 1 | 0% | 1,260 | 1,186 | -6% | 0 | 0 | — |
case-15 | pass→pass | 7,715 | 2,661 | -66% | 1 | 1 | 0% | 1,210 | 1,157 | -4% | 0 | 0 | — |
case-16 | fail→pass | 6,995 | 1,645 | -76% | 1 | 1 | 0% | 1,230 | 1,020 | -17% | 0 | 0 | — |
case-17 | fail→pass | 5,995 | 1,526 | -75% | 1 | 1 | 0% | 1,020 | 1,065 | +4% | 0 | 0 | — |
case-18 | fail→pass | 16,100 | 2,554 | -84% | 1 | 1 | 0% | 2,838 | 1,195 | -58% | 0 | 0 | — |
case-19 | fail→pass | 18,310 | 1,905 | -90% | 1 | 1 | 0% | 3,314 | 1,218 | -63% | 0 | 0 | — |
case-20 | fail→pass | 10,314 | 3,152 | -69% | 1 | 1 | 0% | 1,765 | 1,441 | -18% | 0 | 0 | — |
case-21 | pass→pass | 10,536 | 4,187 | -60% | 1 | 1 | 0% | 1,878 | 1,602 | -15% | 0 | 0 | — |
case-22 | fail→pass | 14,514 | 2,393 | -84% | 1 | 1 | 0% | 2,477 | 1,191 | -52% | 0 | 0 | — |
case-23 | pass→pass | 8,727 | 2,244 | -74% | 1 | 1 | 0% | 1,360 | 1,175 | -14% | 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.
Other measured skills in the registry, with their headline benchmark lift.