Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Decode a personal, auto, or mortgage loan offer into what it really costs and where the traps are. Use when someone asks 'is this loan a good deal', 'decode my loan offer', 'what am I signing', or 'what will this mortgage actually cost me'. Produces a total-cost-of-loan number, APR vs advertised-rate reconciliation, ranked red flags (prepayment penalties, junk fees, rate-reset exposure), and the three questions that most change the deal.
.claude/skills/mohitagw15856-loan-decoder/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 96% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 108% | 0% |
> loan-decoder 的简体中文翻译 — 英文版本为规范版本。
贷款文件围绕他们想让你看的数字(月供)搭建,还有几个他们宁愿你别看的。这个技能算出你总共实际要付多少,然后把offer里所有能悄悄让这个数字对你不利的东西分级列出。
仅在未提供时索取:
算式永远摆出来:
1. 结论——接受 / 先谈这些条款 / 换一家,总成本数字放在最前。
2. 真实数字——总成本、总利息、总费用、APR vs 广告利率及差距解释;浮动利率则给出最坏重置月供。
3. 解读表
| 条款 / 费用 | 文件怎么写 | 对你意味着什么 | 严重程度 | |---|---|---|---|
4. 🚩 警示信号,分级排序——引用的条款、现实情景下的金额代价、建议争取的修改。
5. 最能改变交易的三个问题——针对这份offer(例如"不带附加品的利率是多少?"、"有无提前还款罚金,书面确认?"、"哪些费用归你们,哪些归第三方?")。
6. 什么可以谈——利率、费用、附加品、罚金移除——以及哪根杠杆最能撬动总成本。
在产出物结尾原文附上:"这是一份大白话解读,不构成法律/财务建议——法律因司法辖区而异;任何关键事项请向合格专业人士确认。"
[待确认],绝不悄悄估算借款人侧贷款审查实践——总成本计算、APR对账、费用审计、重置情景呈现。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 24,954 | 22,494 | -10% | 1 | 1 | 0% | 5,209 | 5,747 | +10% | 0 | 0 | — |
case-11 | fail→pass | 10,606 | 7,391 | -30% | 1 | 1 | 0% | 2,206 | 2,340 | +6% | 0 | 0 | — |
case-02 | fail→pass | 22,469 | 20,501 | -9% | 1 | 1 | 0% | 4,709 | 5,430 | +15% | 0 | 0 | — |
case-03 | fail→fail | 18,856 | 15,515 | -18% | 1 | 1 | 0% | 3,824 | 4,002 | +5% | 0 | 0 | — |
case-04 | pass→pass | 7,002 | 8,499 | +21% | 1 | 1 | 0% | 1,832 | 3,047 | +66% | 0 | 0 | — |
case-05 | pass→pass | 9,459 | 9,456 | -0% | 1 | 1 | 0% | 2,007 | 3,347 | +67% | 0 | 0 | — |
case-06 | pass→pass | 10,984 | 8,875 | -19% | 1 | 1 | 0% | 2,260 | 2,913 | +29% | 0 | 0 | — |
case-07 | fail→pass | 12,121 | 17,614 | +45% | 1 | 1 | 0% | 2,581 | 5,071 | +96% | 0 | 0 | — |
case-08 | pass→pass | 10,819 | 13,186 | +22% | 1 | 1 | 0% | 2,004 | 3,492 | +74% | 0 | 0 | — |
case-09 | fail→pass | 7,800 | 6,631 | -15% | 1 | 1 | 0% | 1,533 | 2,290 | +49% | 0 | 0 | — |
case-10 | pass→pass | 9,746 | 10,514 | +8% | 1 | 1 | 0% | 2,099 | 3,155 | +50% | 0 | 0 | — |
case-12 | pass→pass | 9,575 | 14,417 | +51% | 1 | 1 | 0% | 2,127 | 4,028 | +89% | 0 | 0 | — |
case-13 | fail→pass | 9,391 | 14,951 | +59% | 1 | 1 | 0% | 1,787 | 3,720 | +108% | 0 | 0 | — |
case-14 | pass→pass | 3,961 | 5,138 | +30% | 1 | 1 | 0% | 820 | 2,174 | +165% | 0 | 0 | — |
case-15 | pass→pass | 8,664 | 5,413 | -38% | 1 | 1 | 0% | 1,665 | 2,247 | +35% | 0 | 0 | — |
case-16 | fail→pass | 10,237 | 10,784 | +5% | 1 | 1 | 0% | 1,859 | 3,069 | +65% | 0 | 0 | — |
case-17 | fail→pass | 9,901 | 6,431 | -35% | 1 | 1 | 0% | 1,705 | 2,161 | +27% | 0 | 0 | — |
case-18 | fail→pass | 10,427 | 17,071 | +64% | 1 | 1 | 0% | 2,120 | 4,615 | +118% | 0 | 0 | — |
case-19 | fail→pass | 12,413 | 15,210 | +23% | 1 | 1 | 0% | 2,366 | 4,082 | +73% | 0 | 0 | — |
case-20 | pass→pass | 11,309 | 16,839 | +49% | 1 | 1 | 0% | 2,373 | 4,139 | +74% | 0 | 0 | — |
case-21 | fail→pass | 10,473 | 9,076 | -13% | 1 | 1 | 0% | 1,783 | 2,658 | +49% | 0 | 0 | — |
case-22 | pass→pass | 9,982 | 8,402 | -16% | 1 | 1 | 0% | 1,539 | 2,479 | +61% | 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 +45 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.