Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write a credit memo for a lending decision: borrower story, facility structure, repayment sources, financial-ratio spread with covenant headroom, risk factors with mitigants, risk-rating rationale, and a recommendation. Use when asked to write a credit memo, credit application, credit paper, loan write-up, or prepare a deal for credit committee. Produces a complete credit memo ready for committee review.
.claude/skills/mohitagw15856-credit-memo/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -4% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 23% | 0% |
A credit memo has one job: let a committee member who has never met the borrower decide whether the bank gets its money back. This skill writes that document — the story, the structure, the numbers, and the risks — with the discipline that every risk carries a mitigant or an explicit acceptance, and every rating has a stated rationale.
Ask for what's missing; from a thin brief, draft with every inferred figure labelled [assumed — verify]:
Borrower story. What the business does, who owns and runs it, and why they need the money now — growth, refinance, working-capital cycle, or distress dressed as growth. The purpose must match the tenor and structure (don't fund long-term assets with short-term debt).
Repayment sources — the core of the memo:
Ratio spread. Show at least: leverage (Debt/EBITDA), DSCR, interest coverage, current ratio, and any sector-critical metric — 2–3 years of trend, not a snapshot. Frame thresholds as conventional reference points and calibrate to the institution's grid. For each proposed covenant, compute day-one headroom: (actual − required) ÷ required.
Risks and mitigants. Every risk gets a structural mitigant (covenant, security, guarantee, pricing) or an explicit acceptance with rationale ("accepted: single-customer concentration, mitigated partially by 3-year contract; residual risk accepted given…"). A mitigant-free risk list is a memo that hasn't finished its job.
Rating and recommendation. State the rating driver in one sentence (cash-flow strength, leverage, collateral quality, or sector) and what would move it a notch either way. Recommend: approve / approve with conditions (name them) / decline.
1. Recommendation & rating — up front: approve/conditions/decline, proposed rating, one-paragraph why. 2. Borrower story — business, ownership, management, purpose. 3. Facility structure — table: facility | amount | tenor | pricing | security | guarantees. 4. Repayment sources — primary / secondary / tertiary, each tested with numbers. 5. Financial spread — table: metric | FY-2 | FY-1 | current | proj | covenant | headroom. 6. Risks & mitigants — table: risk | mitigant or explicit acceptance | residual. 7. Conditions & covenants — with day-one headroom.
End with: "This memo is analytical support, not a credit decision. Approval authority, rating, and terms follow your institution's credit policy and applicable regulation."
[assumed — verify]| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 33,576 | 27,677 | -18% | 1 | 1 | 0% | 6,113 | 5,817 | -5% | 0 | 0 | — |
case-02 | fail→pass | 31,382 | 25,492 | -19% | 1 | 1 | 0% | 6,245 | 6,102 | -2% | 0 | 0 | — |
case-03 | fail→pass | 32,453 | 26,927 | -17% | 1 | 1 | 0% | 6,223 | 5,966 | -4% | 0 | 0 | — |
case-04 | pass→pass | 8,929 | 13,751 | +54% | 1 | 1 | 0% | 1,799 | 3,853 | +114% | 0 | 0 | — |
case-05 | pass→pass | 65,586 | 8,001 | -88% | 1 | 1 | 0% | 1,094 | 2,586 | +136% | 0 | 0 | — |
case-06 | pass→pass | 10,823 | 18,919 | +75% | 1 | 1 | 0% | 1,946 | 4,775 | +145% | 0 | 0 | — |
case-07 | fail→pass | 81,091 | 20,281 | -75% | 1 | 1 | 0% | 3,830 | 4,856 | +27% | 0 | 0 | — |
case-08 | pass→pass | 16,103 | 17,691 | +10% | 1 | 1 | 0% | 2,498 | 4,522 | +81% | 0 | 0 | — |
case-09 | pass→pass | 10,622 | 13,819 | +30% | 1 | 1 | 0% | 2,089 | 3,658 | +75% | 0 | 0 | — |
case-10 | pass→pass | 33,376 | 8,651 | -74% | 1 | 1 | 0% | 2,032 | 2,476 | +22% | 0 | 0 | — |
case-11 | pass→pass | 17,924 | 12,598 | -30% | 1 | 1 | 0% | 2,733 | 3,274 | +20% | 0 | 0 | — |
case-17 | fail→pass | 11,819 | 7,753 | -34% | 1 | 1 | 0% | 2,213 | 2,718 | +23% | 0 | 0 | — |
case-12 | pass→pass | 13,253 | 18,273 | +38% | 1 | 1 | 0% | 2,587 | 3,963 | +53% | 0 | 0 | — |
case-13 | pass→pass | 16,802 | 74,704 | +345% | 1 | 1 | 0% | 2,757 | 3,746 | +36% | 0 | 0 | — |
case-14 | pass→pass | 16,332 | 18,622 | +14% | 1 | 1 | 0% | 3,045 | 5,035 | +65% | 0 | 0 | — |
case-15 | fail→fail | 15,252 | 13,201 | -13% | 1 | 1 | 0% | 2,822 | 3,918 | +39% | 0 | 0 | — |
case-16 | pass→pass | 13,342 | 6,437 | -52% | 1 | 1 | 0% | 2,100 | 2,339 | +11% | 0 | 0 | — |
case-18 | pass→pass | 16,980 | 11,191 | -34% | 1 | 1 | 0% | 3,057 | 3,161 | +3% | 0 | 0 | — |
case-19 | pass→pass | 17,389 | 19,728 | +13% | 1 | 1 | 0% | 2,634 | 4,413 | +68% | 0 | 0 | — |
case-20 | fail→pass | 12,980 | 10,468 | -19% | 1 | 1 | 0% | 2,181 | 2,862 | +31% | 0 | 0 | — |
case-21 | pass→pass | 15,438 | 15,171 | -2% | 1 | 1 | 0% | 2,876 | 4,201 | +46% | 0 | 0 | — |
case-22 | pass→pass | 2,852 | 4,268 | +50% | 1 | 1 | 0% | 591 | 1,908 | +223% | 0 | 0 | — |
case-23 | pass→pass | 20,610 | 19,153 | -7% | 1 | 1 | 0% | 3,665 | 4,910 | +34% | 0 | 0 | — |
case-24 | pass→pass | 11,366 | 10,152 | -11% | 1 | 1 | 0% | 1,792 | 2,607 | +45% | 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. 24 cases were attempted. The headline lift of +25 percentage points is the difference between those two pass rates over the 24 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.