SKILL.md
Loan Decoder Skill
Loan paperwork is built around the number they want you to see (the monthly payment) and several they'd rather you didn't. This skill computes what you'll actually pay in total, then ranks everything in the offer that can quietly move that number against you.
What This Skill Produces
- The total-cost-of-loan number: everything paid over the life, and total interest + fees
- APR vs. advertised rate, reconciled — where the gap comes from
- Ranked red flags: prepayment penalties, junk fees, variable-rate reset exposure, add-ons
- The three questions that most change this specific deal, plus what's negotiable
Required Inputs
Ask for these only if they aren't already provided:
- The offer document text — loan estimate, term sheet, or contract. With partial paperwork, decode what's there and list the numbers still needed (APR, fee itemization, penalty terms).
- Loan basics if not in the text — amount, rate, term, fixed or variable.
- Their plan — how long they'll keep the loan/asset, and whether they might pay early.
Framework: Severity Scale
- 🔴 Can cost you real money — prepayment penalties (quote the formula, compute an example), precomputed interest / Rule of 78s, mandatory add-ons folded into principal (credit insurance, warranties, GAP), balloon payments, variable-rate exposure without caps, rate markups.
- 🟡 Unusual — push back — junk fees (doc/processing/admin fees beyond genuine third-party costs), mandatory arbitration, cross-collateralization, late-fee stacking.
- 🟢 Standard — ordinary origination structure, genuine third-party costs (appraisal, recording); label them so the reader can relax.
Always show the arithmetic:
- Total cost of loan = all payments + all lender-kept fees; show monthly × months + fees and the interest subtotal.
- APR vs. advertised rate — explain the gap as fees expressed as a rate; if APR isn't stated, flag it, don't estimate silently.
- Variable-rate reset framing — index + margin, caps, and the payment at the caps. Frame as exposure ("payment can go from X to Y"), not prediction.
