Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →银行业务流程编排器，将多个银行数据分析Skills按业务链路串联执行。当用户提出跨多个分析环节的复合业务需求时（如"帮我做一次完整的AML排查"、"做一套客户经营分析"、"全面分析存贷款经营情况"），自动识别所需的Skill链路、确定执行顺序、在节点间传递数据，并在关键节点请求用户确认。
.claude/skills/aifinlab-banking-workflow-orchestrator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 129% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 89% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 90% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 80% | 0% |
银行数据分析体系包含25个专项分析Skills，分布在经营分析、客户经营、渠道交易、风险合规四大领域。单个Skill解决单点问题，但实际业务场景往往需要多个Skill协同——例如一次完整的反洗钱排查需要依次完成初筛、异常识别、聚类分析；一次客户经营闭环需要从客户画像出发，经过分层、预警、渗透分析，最终输出机会清单。
当前痛点：用户需要自己记住Skill之间的依赖关系和数据传递逻辑，手动逐个调用并搬运中间结果。本编排器解决这个问题——用户只需描述业务目标，编排器自动规划执行链路、按序调用、传递数据、输出最终结果。
| 输入项 | 说明 | 是否必须 | |--------|------|----------| | 业务目标描述 | 用户用自然语言描述的分析需求 | 必须 | | 原始数据/文件 | 链路起点Skill所需的输入数据 | 必须 | | 链路选择 | 指定执行哪条预定义链路（可选，不指定则自动识别） | 可选 | | 执行模式 | 全自动 / 逐步确认（默认逐步确认） | 可选 | | 中间结果调整 | 用户对某个节点输出的修正或补充 | 可选 |
以下是从25个Skills的依赖关系中提取的标准业务链路。
suspicious-transaction-screening → transaction-flow-anomaly-detection → high-risk-transaction-clustering
初筛命中交易 异常模式识别 聚类分析输出复核清单bank-customer-360 → customer-segmentation → ┬─ customer-churn-alert
├─ customer-product-penetration
├─ dormant-account-analysis
└─ early-repayment-churn-analysis
↓ （全部汇入）
customer-opportunity-list-generation┬─ business-metrics-attribution
├─ branch-performance-benchmarking
├─ deposit-growth-attribution → business-analysis-summary
├─ loan-structure-analysis
└─ bank-calc-utils（计算支撑）fund-flow-analysis → large-fund-movement-tracking ↔ account-volatility-monitoring
资金流向分析 大额资金追踪 账户波动监测payment-failure-attribution → channel-transaction-performance
支付失败归因 渠道交易表现分析用户可以指定任意Skill组合构建自定义链路，编排器负责：
收到用户请求后：
执行前检查起点Skill所需的输入数据：
按拓扑顺序执行链路中的每个Skill节点：
对于可并行的节点（如客户链路中的4个分析节点），同时启动执行以提高效率。
节点间数据传递遵循以下规则：
链路执行完毕后：
本Skill依赖所有25个银行数据分析Skills，按需调用：
| 领域 | 可调用Skills | |------|-------------| | 经营分析 | business-metrics-attribution, business-analysis-summary, branch-performance-benchmarking, deposit-growth-attribution, loan-structure-analysis, bank-calc-utils | | 客户经营 | bank-customer-360, customer-segmentation, customer-churn-alert, customer-product-penetration, customer-opportunity-list-generation, dormant-account-analysis, early-repayment-churn-analysis, segment-credit-performance | | 渠道交易 | channel-transaction-performance, payment-failure-attribution, fund-flow-analysis, large-fund-movement-tracking, account-volatility-monitoring | | 风险合规 | suspicious-transaction-screening, transaction-flow-anomaly-detection, high-risk-transaction-clustering, risk-customer-behavior-change, watchlist-match-result-analysis, regulatory-reporting-quality-validation |
无（顶层编排Skill，输出直接面向用户）
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 36,258 | 64,298 | +77% | 1 | 1 | 0% | 6,147 | 8,171 | +33% | 0 | 0 | — |
case-02 | fail→fail | 33,310 | 32,398 | -3% | 1 | 1 | 0% | 5,677 | 7,852 | +38% | 0 | 0 | — |
case-03 | fail→fail | 29,548 | 29,177 | -1% | 1 | 1 | 0% | 3,938 | 7,124 | +81% | 0 | 0 | — |
case-04 | pass→pass | 28,410 | 13,443 | -53% | 1 | 1 | 0% | 2,838 | 4,426 | +56% | 0 | 0 | — |
case-05 | pass→pass | 13,532 | 10,347 | -24% | 1 | 1 | 0% | 2,141 | 3,986 | +86% | 0 | 0 | — |
case-06 | fail→pass | 11,657 | 9,681 | -17% | 1 | 1 | 0% | 1,634 | 3,742 | +129% | 0 | 0 | — |
case-07 | fail→pass | 18,943 | 17,988 | -5% | 1 | 1 | 0% | 2,781 | 5,261 | +89% | 0 | 0 | — |
case-08 | pass→pass | 20,551 | 16,732 | -19% | 1 | 1 | 0% | 2,537 | 4,856 | +91% | 0 | 0 | — |
case-09 | fail→fail | 17,717 | 26,738 | +51% | 1 | 1 | 0% | 2,709 | 6,652 | +146% | 0 | 0 | — |
case-10 | fail→pass | 19,026 | 11,443 | -40% | 1 | 1 | 0% | 3,218 | 4,208 | +31% | 0 | 0 | — |
case-11 | pass→pass | 17,206 | 10,188 | -41% | 1 | 1 | 0% | 2,358 | 3,972 | +68% | 0 | 0 | — |
case-12 | pass→pass | 16,215 | 14,473 | -11% | 1 | 1 | 0% | 2,345 | 4,839 | +106% | 0 | 0 | — |
case-13 | pass→pass | 15,332 | 8,811 | -43% | 1 | 1 | 0% | 2,080 | 3,616 | +74% | 0 | 0 | — |
case-14 | fail→pass | 16,241 | 12,548 | -23% | 1 | 1 | 0% | 2,361 | 4,494 | +90% | 0 | 0 | — |
case-15 | pass→pass | 17,033 | 8,639 | -49% | 1 | 1 | 0% | 2,440 | 3,774 | +55% | 0 | 0 | — |
case-16 | pass→pass | 18,349 | 13,087 | -29% | 1 | 1 | 0% | 2,542 | 4,365 | +72% | 0 | 0 | — |
case-17 | fail→pass | 16,981 | 12,319 | -27% | 1 | 1 | 0% | 2,369 | 4,272 | +80% | 0 | 0 | — |
case-18 | pass→pass | 19,388 | 17,355 | -10% | 1 | 1 | 0% | 2,907 | 5,179 | +78% | 0 | 0 | — |
case-19 | pass→pass | 11,886 | 12,846 | +8% | 1 | 1 | 0% | 1,745 | 4,202 | +141% | 0 | 0 | — |
case-20 | pass→pass | 15,900 | 11,602 | -27% | 1 | 1 | 0% | 2,468 | 4,128 | +67% | 0 | 0 | — |
case-21 | fail→pass | 11,329 | 14,688 | +30% | 1 | 1 | 0% | 1,558 | 4,690 | +201% | 0 | 0 | — |
case-22 | pass→pass | 14,435 | 11,293 | -22% | 1 | 1 | 0% | 2,093 | 4,062 | +94% | 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 +27 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.