Credit Risk Committee
Use the shared credit office public API to prepare structured committee-ready JSON answers for branch reviews, allocation decisions, stress tests, and segment posture analysis.
Quick Start
Every task starts the same way:
- Fetch manifest and policies in parallel.
- Read the supplied
answer_template.json to know the required output shape.
- Fetch branch details, loans, metrics, sector-exposures, and benchmarks as needed.
- Apply policy rules to derive ratings, scores, stress results, and decisions.
- Output valid JSON conforming to the template.
API Endpoints
The base URL arrives as <TASK_ENV_BASE_URL>. All endpoints are GET-only. Call them with curl -s and pipe through python3 -m json.tool for readability. Use python3 -c for inline computation.
Always fetch in parallel where possible. The manifest and policies are needed for every task.
Always Available
GET /api/manifest -- record counts, versions, branches list
GET /api/policies -- risk-rating thresholds, CDFI factor scores, stress formulas, CRE weights, concentration rules, enums
GET /api/benchmarks/fdic/q4-2024 -- five FDIC benchmark ratios
GET /api/benchmarks/ncua/q1-2025 -- per-state NCUA benchmarks with US national row
Branch-Scoped
GET /api/branches/{branch_id} -- lending_capacity_q1, sector_ceiling_pct, cre_policy_limit_pct, total_assets, state_code
GET /api/branches/{branch_id}/metrics -- total_loans_outstanding, nonperforming_loans, delinquency_30_plus_pct
GET /api/branches/{branch_id}/loans -- current_rating, dscr, ltv, fico, payment_status, debt_to_asset, liquidity_months, sector, outstanding_balance
GET /api/branches/{branch_id}/sector-exposures -- sector-level exposure, limit_pct, grandfathered
GET /api/branches/{branch_id}/applications -- pending applications with financial inputs
Credit Union Segment
GET /api/credit-union-segments/{segment_id} -- segment profile, quarterly_capacity, peer_states, risk_tolerance, minimum_checklist
Core Business Rules
See references/policies.md for the complete rule reference. Key rules to apply:
Risk Rating Re-derivation
The final rating is the worst (highest numeric) from three independent factors. Apply only where data exists:
- DSCR: >=1.50 -> 3, >=1.25 -> 4, >=1.05 -> 5, >=1.00 -> 6, <1.00 -> 7
- LTV: <=0.65 -> 3, <=0.75 -> 4, <=0.85 -> 5, <=1.00 -> 6, >1.00 -> 7
- Delinquency: 30 Days -> 4, 60 Days -> 5, 90+ Days -> 7, Nonaccrual -> 8 (Current -> no factor)
Null dscr/ltv omit that factor. final_rating = max(available factor ratings).
CDFI Factor Scoring
Sum scores from 4 factors (0, 1, 3, or 5 each; see references/policies.md for full table). Classify: Prime 0-5, Desirable 6-9, Satisfactory 10-13, Watch 14-18, Doubtful >=19, Projected Loss >=19 with ltv>1.0.
CRE Weighted Scoring
0.45*capacity + 0.03*capital + 0.05*character + 0.36*collateral_exposure + 0.11*conditions. Each component 1-5. <=2.0 approve_quality, <=3.0 conditional, >3.0 weak.
Stress Formulas
- Watch-list +200bp:
stressed_dscr = dscr / 1.18 (breach < 1.0)
- CRE dual-stress:
stressed_dscr = dscr * 0.85 / 1.18 (breach < 1.0)
Concentration
Sector exposure / branch total_loans_outstanding. Compare to sector_ceiling_pct (general) or cre_policy_limit_pct (CRE). Grandfathered sectors may already exceed; new approvals may not worsen without mitigation. Variance bps: (actual - limit) * 10000.
NPA Benchmark
branch_npa_ratio = nonperforming_loans / total_loans_outstanding. Variance: ratio - fdic_benchmark. BPS: variance * 10000.
Watch-List Actions
Assign: monitor, watchlist, special_assets, workout, partial_chargeoff_review, legal_referral. Ratings >=7 with nonaccrual or breached DSCR warrant special_assets or stronger.
Material Downgrades
final_rating - current_rating >= 2. List sorted by loan_id.
Task Patterns
Rating Migration Review
- Fetch branch, metrics, loans, sector-exposures, policies, FDIC benchmark.
- Filter loans where
current_rating >= target_min.
- Re-derive each loan rating. Sum exposure by final_rating.
- Track migration for loans originally at target_min.
- Assign watch-list actions to loans with final_rating >=6 or breached DSCR.
- Compute NPA benchmark variance.
- List material downgrades (>=2 notches).
- Top problem credit: highest final_rating, then largest exposure.
Lending Allocation
- Fetch branch, metrics, sector-exposures, applications, policies.
- Prioritize: DSCR, relationship_years, sector headroom, credit quality.
- Allocate from lending_capacity_q1; track cumulative usage.
- Decline reasons: weak_dscr when <1.05, high_ltv when >0.85, low_fico when <580, startup_risk when years_in_business <2, capacity_limit when exhausted, recent_bankruptcy, sector_breach when post-approval exceeds limit.
- Flag sectors at or over limit_pct.
- Post-approval concentrations: existing + approved.
Watch-List Stress
- Fetch branch, loans, policies.
- Filter loans with
current_rating >= 6.
- Score CDFI risk class per adverse loan.
- Apply +200bp DSCR stress. Breach when stressed_dscr < 1.0.
- Workout queue sorted by descending exposure.
- Assign recommended_action by severity.
- Severe bucket counts: group by current_rating + payment_status.
Competing CRE Decision
- Fetch branch, metrics, loans, sector-exposures, target applications, policies, FDIC benchmark.
- CRE weighted score per application.
- CRE dual-stress formula.
- CRE exposure and post-approval concentration vs cre_policy_limit_pct.
- FDIC delinquency benchmark vs branch delinquency.
- Recommend stronger credit; decline/defer unselected with reason codes.
- List conditions for selected credit.
Segment Posture
- Fetch segment data, NCUA benchmarks.
- Compare state to US national and peer medians.
- Posture: continue, continue_with_tighter_conditions, or pause.
- Escalation triggers from internal_context.
- Interpretation: capacity_status, external_risk_status, risk_tolerance, committee_message.
Guardrails
- Round currency to 2 decimals, ratios to 4 decimals.
- Sort as specified: ascending loan_id, application_id, rating, sector.
- Use only enum values from policies or template. No invented codes.
- Null factors are omitted, not guessed.
- Branch-scoped data only; do not merge across branches.
- Output only valid JSON matching the template. No narrative text.
1---2name: credit-risk-committee-23description: Prepare credit risk committee packets using a shared credit office REST API. Use when Codex needs to produce branch rating migration reviews, lending-committee allocation decisions, watch-list stress packets, competing CRE decisions, or credit-union segment posture pages from the task environment API.4metadata:5 source: https://skillmd.com/skills/prism-shadow/credit-risk-committee-26---78# Credit Risk Committee910Use the shared credit office public API to prepare structured committee-ready JSON answers for branch reviews, allocation decisions, stress tests, and segment posture analysis.1112## Quick Start1314Every task starts the same way:15161. Fetch manifest and policies in parallel.172. Read the supplied `answer_template.json` to know the required output shape.183. Fetch branch details, loans, metrics, sector-exposures, and benchmarks as needed.194. Apply policy rules to derive ratings, scores, stress results, and decisions.205. Output valid JSON conforming to the template.2122## API Endpoints2324The base URL arrives as `<TASK_ENV_BASE_URL>`. All endpoints are GET-only. Call them with `curl -s` and pipe through `python3 -m json.tool` for readability. Use `python3 -c` for inline computation.2526Always fetch in parallel where possible. The manifest and policies are needed for every task.2728### Always Available2930- `GET /api/manifest` -- record counts, versions, branches list31- `GET /api/policies` -- risk-rating thresholds, CDFI factor scores, stress formulas, CRE weights, concentration rules, enums32- `GET /api/benchmarks/fdic/q4-2024` -- five FDIC benchmark ratios33- `GET /api/benchmarks/ncua/q1-2025` -- per-state NCUA benchmarks with US national row3435### Branch-Scoped3637- `GET /api/branches/{branch_id}` -- lending_capacity_q1, sector_ceiling_pct, cre_policy_limit_pct, total_assets, state_code38- `GET /api/branches/{branch_id}/metrics` -- total_loans_outstanding, nonperforming_loans, delinquency_30_plus_pct39- `GET /api/branches/{branch_id}/loans` -- current_rating, dscr, ltv, fico, payment_status, debt_to_asset, liquidity_months, sector, outstanding_balance40- `GET /api/branches/{branch_id}/sector-exposures` -- sector-level exposure, limit_pct, grandfathered41- `GET /api/branches/{branch_id}/applications` -- pending applications with financial inputs4243### Credit Union Segment4445- `GET /api/credit-union-segments/{segment_id}` -- segment profile, quarterly_capacity, peer_states, risk_tolerance, minimum_checklist4647## Core Business Rules4849See [references/policies.md](references/policies.md) for the complete rule reference. Key rules to apply:5051### Risk Rating Re-derivation5253The final rating is the **worst (highest numeric)** from three independent factors. Apply only where data exists:5455- **DSCR**: >=1.50 -> 3, >=1.25 -> 4, >=1.05 -> 5, >=1.00 -> 6, <1.00 -> 756- **LTV**: <=0.65 -> 3, <=0.75 -> 4, <=0.85 -> 5, <=1.00 -> 6, >1.00 -> 757- **Delinquency**: 30 Days -> 4, 60 Days -> 5, 90+ Days -> 7, Nonaccrual -> 8 (Current -> no factor)5859Null dscr/ltv omit that factor. `final_rating = max(available factor ratings)`.6061### CDFI Factor Scoring6263Sum scores from 4 factors (0, 1, 3, or 5 each; see [references/policies.md](references/policies.md) for full table). Classify: Prime 0-5, Desirable 6-9, Satisfactory 10-13, Watch 14-18, Doubtful >=19, Projected Loss >=19 with ltv>1.0.6465### CRE Weighted Scoring6667`0.45*capacity + 0.03*capital + 0.05*character + 0.36*collateral_exposure + 0.11*conditions`. Each component 1-5. <=2.0 approve_quality, <=3.0 conditional, >3.0 weak.6869### Stress Formulas7071- **Watch-list +200bp**: `stressed_dscr = dscr / 1.18` (breach < 1.0)72- **CRE dual-stress**: `stressed_dscr = dscr * 0.85 / 1.18` (breach < 1.0)7374### Concentration7576Sector exposure / branch total_loans_outstanding. Compare to sector_ceiling_pct (general) or cre_policy_limit_pct (CRE). Grandfathered sectors may already exceed; new approvals may not worsen without mitigation. Variance bps: `(actual - limit) * 10000`.7778### NPA Benchmark7980`branch_npa_ratio = nonperforming_loans / total_loans_outstanding`. Variance: `ratio - fdic_benchmark`. BPS: `variance * 10000`.8182### Watch-List Actions8384Assign: monitor, watchlist, special_assets, workout, partial_chargeoff_review, legal_referral. Ratings >=7 with nonaccrual or breached DSCR warrant special_assets or stronger.8586### Material Downgrades8788`final_rating - current_rating >= 2`. List sorted by loan_id.8990## Task Patterns9192### Rating Migration Review93941. Fetch branch, metrics, loans, sector-exposures, policies, FDIC benchmark.952. Filter loans where `current_rating >= target_min`.963. Re-derive each loan rating. Sum exposure by final_rating.974. Track migration for loans originally at target_min.985. Assign watch-list actions to loans with final_rating >=6 or breached DSCR.996. Compute NPA benchmark variance.1007. List material downgrades (>=2 notches).1018. Top problem credit: highest final_rating, then largest exposure.102103### Lending Allocation1041051. Fetch branch, metrics, sector-exposures, applications, policies.1062. Prioritize: DSCR, relationship_years, sector headroom, credit quality.1073. Allocate from lending_capacity_q1; track cumulative usage.1084. Decline reasons: weak_dscr when <1.05, high_ltv when >0.85, low_fico when <580, startup_risk when years_in_business <2, capacity_limit when exhausted, recent_bankruptcy, sector_breach when post-approval exceeds limit.1095. Flag sectors at or over limit_pct.1106. Post-approval concentrations: existing + approved.111112### Watch-List Stress1131141. Fetch branch, loans, policies.1152. Filter loans with `current_rating >= 6`.1163. Score CDFI risk class per adverse loan.1174. Apply +200bp DSCR stress. Breach when stressed_dscr < 1.0.1185. Workout queue sorted by descending exposure.1196. Assign recommended_action by severity.1207. Severe bucket counts: group by current_rating + payment_status.121122### Competing CRE Decision1231241. Fetch branch, metrics, loans, sector-exposures, target applications, policies, FDIC benchmark.1252. CRE weighted score per application.1263. CRE dual-stress formula.1274. CRE exposure and post-approval concentration vs cre_policy_limit_pct.1285. FDIC delinquency benchmark vs branch delinquency.1296. Recommend stronger credit; decline/defer unselected with reason codes.1307. List conditions for selected credit.131132### Segment Posture1331341. Fetch segment data, NCUA benchmarks.1352. Compare state to US national and peer medians.1363. Posture: continue, continue_with_tighter_conditions, or pause.1374. Escalation triggers from internal_context.1385. Interpretation: capacity_status, external_risk_status, risk_tolerance, committee_message.139140## Guardrails141142- Round currency to 2 decimals, ratios to 4 decimals.143- Sort as specified: ascending loan_id, application_id, rating, sector.144- Use only enum values from policies or template. No invented codes.145- Null factors are omitted, not guessed.146- Branch-scoped data only; do not merge across branches.147- Output only valid JSON matching the template. No narrative text.