Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要在银行零售金融场景下，对客户投诉进行归因拆解、驱动项识别、问题定位与改进建议输出时使用本技能。适合输出归因结论、证据链与后续改进动作。
.claude/skills/aifinlab-bank-t168-retail-finance-complaint-attribution-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 35% | 0% |
本技能用于将投诉数据转化为“可解释的归因结论 + 可落地的改进动作”。强调口径一致、证据链清晰、相关性与因果区分，不用于替代审计或法律意见。
time_window、complaintsreferences/input-schema.mdbashpython scripts/run_skill.py --input assets/example-input.json --format markdown python scripts/run_skill.py --input assets/example-input.json --format json
脚本入口：scripts/run_skill.py（调用 shared/retail_service_ops_skill_engine.py 的 t168 场景）。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 31,348 | 27,124 | -13% | 1 | 1 | 0% | 3,448 | 3,904 | +13% | 0 | 0 | — |
case-01 | fail→fail | 27,842 | 40,657 | +46% | 1 | 1 | 0% | 3,375 | 4,898 | +45% | 0 | 0 | — |
case-02 | fail→fail | 34,506 | 23,084 | -33% | 1 | 1 | 0% | 4,369 | 3,820 | -13% | 0 | 0 | — |
case-03 | fail→pass | 20,384 | 35,378 | +74% | 1 | 1 | 0% | 2,332 | 2,962 | +27% | 0 | 0 | — |
case-05 | fail→pass | 22,220 | 16,254 | -27% | 1 | 1 | 0% | 2,714 | 2,525 | -7% | 0 | 0 | — |
case-06 | fail→pass | 20,703 | 19,490 | -6% | 1 | 1 | 0% | 2,629 | 2,900 | +10% | 0 | 0 | — |
case-07 | fail→pass | 22,660 | 24,052 | +6% | 1 | 1 | 0% | 2,821 | 3,810 | +35% | 0 | 0 | — |
case-08 | fail→fail | 15,631 | 16,192 | +4% | 1 | 1 | 0% | 2,721 | 3,096 | +14% | 0 | 0 | — |
case-25 | fail→pass | 14,518 | 5,083 | -65% | 1 | 1 | 0% | 2,174 | 1,145 | -47% | 0 | 0 | — |
case-09 | fail→pass | 11,587 | 16,282 | +41% | 1 | 1 | 0% | 1,748 | 2,737 | +57% | 0 | 0 | — |
case-10 | fail→fail | 13,233 | 18,549 | +40% | 1 | 1 | 0% | 2,008 | 2,985 | +49% | 0 | 0 | — |
case-11 | fail→pass | 37,083 | 19,301 | -48% | 1 | 1 | 0% | 3,583 | 2,851 | -20% | 0 | 0 | — |
case-12 | pass→pass | 24,564 | 25,143 | +2% | 1 | 1 | 0% | 3,479 | 4,030 | +16% | 0 | 0 | — |
case-13 | fail→pass | 23,377 | 27,482 | +18% | 1 | 1 | 0% | 3,317 | 3,982 | +20% | 0 | 0 | — |
case-14 | fail→pass | 23,797 | 22,676 | -5% | 1 | 1 | 0% | 3,511 | 3,690 | +5% | 0 | 0 | — |
case-15 | fail→pass | 24,617 | 23,636 | -4% | 1 | 1 | 0% | 3,408 | 3,633 | +7% | 0 | 0 | — |
case-16 | pass→pass | 20,794 | 27,576 | +33% | 1 | 1 | 0% | 2,723 | 4,045 | +49% | 0 | 0 | — |
case-17 | fail→pass | 24,447 | 26,695 | +9% | 1 | 1 | 0% | 3,287 | 4,048 | +23% | 0 | 0 | — |
case-18 | pass→pass | 22,076 | 27,744 | +26% | 1 | 1 | 0% | 2,985 | 3,744 | +25% | 0 | 0 | — |
case-19 | pass→pass | 21,719 | 21,764 | +0% | 1 | 1 | 0% | 2,889 | 3,340 | +16% | 0 | 0 | — |
case-20 | fail→fail | 23,895 | 25,422 | +6% | 1 | 1 | 0% | 3,505 | 4,085 | +17% | 0 | 0 | — |
case-21 | fail→fail | 25,110 | 24,710 | -2% | 1 | 1 | 0% | 3,227 | 4,244 | +32% | 0 | 0 | — |
case-22 | fail→pass | 24,393 | 28,340 | +16% | 1 | 1 | 0% | 3,478 | 4,265 | +23% | 0 | 0 | — |
case-23 | fail→pass | 22,457 | 25,261 | +12% | 1 | 1 | 0% | 2,751 | 3,589 | +30% | 0 | 0 | — |
case-24 | pass→pass | 30,823 | 22,716 | -26% | 1 | 1 | 0% | 3,693 | 3,745 | +1% | 0 | 0 | — |
case-26 | pass→fail | 23,090 | 40,954 | +77% | 1 | 1 | 0% | 2,947 | 3,515 | +19% | 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. 26 cases were attempted. The headline lift of +50 percentage points is the difference between those two pass rates over the 26 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.