Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when recording, recalling, listing, showing, removing, clearing, or running reindex for verified/failed SQL and domain notes.
.claude/skills/aliyun-memory/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 239% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -64% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -61% | 0% |
Use memory after a query succeeds or fails, or when past verified SQL can help answer a new question.
bashmcs memory recall "keyword" mcs memory verify --question Q --sql '<SQL>' --tables T1,T2 mcs memory fail --question Q --sql '<SQL>' --error-msg '<msg>' --remediation '<fix>' mcs memory note 'domain note' mcs memory list mcs memory show ID mcs memory remove ID mcs memory reindex
For multi-source profiles, pass FQN table names in --tables when a bare table name is ambiguous.
See references/memory.md.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 4,257 | 2,783 | -35% | 1 | 1 | 0% | 696 | 513 | -26% | 0 | 0 | — |
case-10 | fail→fail | 10,334 | 4,905 | -53% | 1 | 1 | 0% | 1,912 | 259 | -86% | 0 | 0 | — |
case-01 | fail→pass | 3,264 | 7,993 | +145% | 1 | 1 | 0% | 467 | 1,581 | +239% | 0 | 0 | — |
case-02 | fail→pass | 9,423 | 6,391 | -32% | 1 | 1 | 0% | 1,657 | 1,275 | -23% | 0 | 0 | — |
case-03 | fail→fail | 8,781 | 7,099 | -19% | 1 | 1 | 0% | 1,415 | 265 | -81% | 0 | 0 | — |
case-04 | fail→pass | 12,174 | 3,369 | -72% | 1 | 1 | 0% | 1,903 | 687 | -64% | 0 | 0 | — |
case-06 | fail→pass | 7,068 | 1,757 | -75% | 1 | 1 | 0% | 934 | 365 | -61% | 0 | 0 | — |
case-07 | fail→pass | 7,796 | 1,873 | -76% | 1 | 1 | 0% | 976 | 381 | -61% | 0 | 0 | — |
case-08 | fail→fail | 8,510 | 4,226 | -50% | 1 | 1 | 0% | 571 | 292 | -49% | 0 | 0 | — |
case-09 | fail→pass | 14,095 | 1,622 | -88% | 1 | 1 | 0% | 2,201 | 386 | -82% | 0 | 0 | — |
case-11 | fail→fail | 3,671 | 7,894 | +115% | 1 | 1 | 0% | 586 | 592 | +1% | 0 | 0 | — |
case-12 | fail→fail | 3,734 | 7,660 | +105% | 1 | 1 | 0% | 279 | 515 | +85% | 0 | 0 | — |
case-13 | fail→pass | 14,869 | 3,395 | -77% | 1 | 1 | 0% | 2,228 | 767 | -66% | 0 | 0 | — |
case-14 | fail→fail | 1,590 | 7,514 | +373% | 1 | 1 | 0% | 208 | 404 | +94% | 0 | 0 | — |
case-15 | fail→fail | 20,659 | 11,987 | -42% | 1 | 1 | 0% | 1,792 | 658 | -63% | 0 | 0 | — |
case-16 | fail→pass | 6,119 | 9,090 | +49% | 1 | 1 | 0% | 1,211 | 1,571 | +30% | 0 | 0 | — |
case-17 | fail→pass | 4,227 | 1,572 | -63% | 1 | 1 | 0% | 600 | 228 | -62% | 0 | 0 | — |
case-18 | fail→fail | 6,448 | 4,499 | -30% | 1 | 1 | 0% | 780 | 354 | -55% | 0 | 0 | — |
case-19 | fail→fail | 7,503 | 4,241 | -43% | 1 | 1 | 0% | 1,295 | 319 | -75% | 0 | 0 | — |
case-20 | fail→fail | 2,809 | 5,490 | +95% | 1 | 1 | 0% | 386 | 482 | +25% | 0 | 0 | — |
case-21 | pass→fail | 3,240 | 7,217 | +123% | 1 | 1 | 0% | 516 | 405 | -22% | 0 | 0 | — |
case-22 | pass→pass | 3,621 | 3,223 | -11% | 1 | 1 | 0% | 657 | 687 | +5% | 0 | 0 | — |
case-23 | pass→pass | 10,229 | 21,293 | +108% | 1 | 1 | 0% | 1,829 | 2,854 | +56% | 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 12 counted toward the lift figure. The other 11 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 +39 percentage points is the difference between those two pass rates over the 12 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.
Other measured skills in the registry, with their headline benchmark lift.