Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when you need a bank post-loan visit follow-up assistant to turn visit notes/materials into a structured memo (摘要/关键更新/风险观察/用途核验/行动项). Trigger this skill for贷后回访纪要、现场走访记录整理、催办跟踪与行动项台账生成，并可用脚本从结构化输入生成标准纪要。
.claude/skills/aifinlab-bank-t142-corporate-finance-task-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 128% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 89% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 129% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 112% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 311% | 0% |
本技能用于把“贷后回访/现场走访/电话回访”的零散信息，整理成可流转、可追责、可复查的结构化纪要。
它解决的不是“写得更长”，而是：
最低必需（详见 references/input-schema.md）：
company.namevisit.visit_datevisit.purpose建议提供（能显著提升输出质量）：
visit.key_updates[]：事实更新（可复查）visit.risk_observations[]：异常/风险观察visit.loan_use_check：资金用途核验结论（四流合一/替代证据）visit.repayment_source_update：还款来源更新visit.action_items[]：行动项（责任到人 + 截止）脚本输出字段见 references/output-schema.md。
bashpython scripts/run_skill.py --input assets/example-input.json --format markdown python scripts/run_skill.py --input assets/example-input.json --format json
scripts/run_skill.py：命令行入口scripts/post_loan_visit_summary.py：t142 场景封装（调用共享引擎）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | pass→pass | 10,555 | 13,248 | +26% | 1 | 1 | 0% | 1,682 | 3,273 | +95% | 0 | 0 | — |
case-01 | fail→pass | 10,526 | 47,343 | +350% | 1 | 1 | 0% | 1,754 | 3,997 | +128% | 0 | 0 | — |
case-02 | fail→pass | 14,871 | 18,596 | +25% | 1 | 1 | 0% | 1,956 | 3,706 | +89% | 0 | 0 | — |
case-04 | fail→pass | 15,795 | 25,861 | +64% | 1 | 1 | 0% | 2,030 | 4,644 | +129% | 0 | 0 | — |
case-05 | pass→pass | 16,871 | 23,599 | +40% | 1 | 1 | 0% | 2,683 | 4,643 | +73% | 0 | 0 | — |
case-06 | pass→pass | 6,190 | 19,518 | +215% | 1 | 1 | 0% | 976 | 4,177 | +328% | 0 | 0 | — |
case-07 | pass→pass | 4,925 | 15,836 | +222% | 1 | 1 | 0% | 710 | 3,540 | +399% | 0 | 0 | — |
case-08 | fail→pass | 14,291 | 24,318 | +70% | 1 | 1 | 0% | 2,075 | 4,405 | +112% | 0 | 0 | — |
case-09 | pass→pass | 14,596 | 19,886 | +36% | 1 | 1 | 0% | 1,885 | 3,799 | +102% | 0 | 0 | — |
case-10 | pass→pass | 12,584 | 24,920 | +98% | 1 | 1 | 0% | 1,985 | 4,417 | +123% | 0 | 0 | — |
case-11 | fail→pass | 7,234 | 16,472 | +128% | 1 | 1 | 0% | 888 | 3,646 | +311% | 0 | 0 | — |
case-12 | pass→pass | 13,749 | 23,071 | +68% | 1 | 1 | 0% | 2,024 | 3,813 | +88% | 0 | 0 | — |
case-13 | pass→pass | 8,427 | 21,865 | +159% | 1 | 1 | 0% | 1,404 | 4,691 | +234% | 0 | 0 | — |
case-14 | fail→pass | 11,701 | 19,420 | +66% | 1 | 1 | 0% | 1,580 | 3,926 | +148% | 0 | 0 | — |
case-15 | pass→pass | 22,844 | 17,953 | -21% | 1 | 1 | 0% | 1,665 | 3,576 | +115% | 0 | 0 | — |
case-16 | pass→pass | 11,620 | 25,031 | +115% | 1 | 1 | 0% | 1,898 | 5,343 | +182% | 0 | 0 | — |
case-17 | pass→pass | 15,826 | 50,213 | +217% | 1 | 1 | 0% | 2,204 | 4,354 | +98% | 0 | 0 | — |
case-18 | pass→pass | 18,370 | 24,566 | +34% | 1 | 1 | 0% | 2,737 | 4,160 | +52% | 0 | 0 | — |
case-19 | pass→pass | 10,875 | 22,991 | +111% | 1 | 1 | 0% | 1,505 | 4,668 | +210% | 0 | 0 | — |
case-20 | fail→pass | 18,590 | 21,284 | +14% | 1 | 1 | 0% | 2,567 | 4,388 | +71% | 0 | 0 | — |
case-21 | fail→pass | 16,772 | 20,846 | +24% | 1 | 1 | 0% | 2,173 | 4,119 | +90% | 0 | 0 | — |
case-22 | fail→pass | 27,645 | 25,489 | -8% | 1 | 1 | 0% | 4,594 | 4,845 | +5% | 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 +41 percentage points is the difference between those two pass rates over the 22 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.