SKILL.md
Lending Risk Brief Skill
A loan book fails in patterns before it fails in names. This skill writes the portfolio-level brief that makes the patterns visible: where the book is concentrated, which vintages are misbehaving, which way the grades are migrating, what the macro could do to it, and which names need decisions now.
What This Skill Produces
- Concentration analysis: sector, geography, single-name — each against limits
- Vintage performance comparison
- A migration-matrix narrative (not just the matrix)
- Macro-sensitivity scenarios (base / adverse / severe) with named transmission channels
- A top-10 watch-names table
- Recommended actions with owners
Required Inputs
Ask for what's available; compute what the data supports and mark the rest [data gap — request from portfolio systems]:
- Portfolio snapshot — exposures by borrower, sector, geography, grade, origination vintage
- Concentration limits from the risk appetite statement, if set
- Grade migrations this period (upgrades/downgrades by exposure)
- Delinquency/NPL and provision figures, current and prior periods
- Watch-list candidates already known to the team
Portfolio Framework
1. Concentration. Three cuts, each vs its limit (or vs a stated reference norm if no limit exists — and flag the missing limit as a finding): top sector and top-3 sector share; geographic share; single-name — top-10 and top-20 obligor share, largest single exposure vs capital. Concentration that grew via passive drift (runoff elsewhere) deserves the same flag as active growth — say which it was. Correlated concentrations count together (e.g. construction lending + commercial-real-estate collateral is one bet, not two).
2. Vintage performance. Compare cohorts at the same age on-book (delinquency/default at month 12, 24…), not calendar snapshots — a young book always looks clean. A vintage underperforming its age-matched predecessors signals an underwriting-standards question for that origination period; name the period and what changed in criteria then, if known.
