Skip to main content

loan-file-readiness-reviewer

Reviews lending and mortgage files for document completeness, extraction provenance, cross-source consistency, freshness, and policy-defined exceptions. Use when processors need a traceable readiness package before human underwriting or approval.

Ir para a instalação

Informações da origem

Repositório
aiappsgbb/agentic-loop
Última atividade na origem
11 de agosto de 2026 às 14:09
Idioma detectado do SKILL.md
inglês
Estrelas
1
Forks
0

Opções de instalação

Por padrão, está selecionado o prompt que primeiro revisa a origem. Você pode mudar para um comando direto ou baixar uma cópia local.

Revise os arquivos de origem

Leia o SKILL.md e os arquivos complementares exibidos pelo SkillsMP antes de decidir se vai instalar.

Exibindo SKILL.md

SKILL.md
Instruções da origem · Visualização somente leitura
name
loan-file-readiness-reviewer
description
Reviews lending and mortgage files for document completeness, extraction provenance, cross-source consistency, freshness, and policy-defined exceptions. Use when processors need a traceable readiness package before human underwriting or approval.
license
MIT
metadata
{"author":"aiappsgbb","version":"1.1"}
# Loan-File Readiness Reviewer Create a processor-ready, auditable inventory of what a lending file contains, what it lacks, and which inconsistencies require human review. This skill prepares evidence; it does not assess creditworthiness or make a lending decision. ## When to use - A new or updated loan file needs completeness review - Extracted document data needs provenance and reconciliation - A processor needs a minimal, prioritized evidence request - A file needs readiness status before underwriting **Not for:** approval, denial, pricing, credit-risk ranking, fraud determination, legal advice, or requesting documents outside the applicable checklist and policy. ## Required inputs - Application and product-specific document checklist - Authorized borrower documents and system-of-record data - Document validity, recency, and signature requirements - Current processing policy and exception-routing rules - Source identifiers and extraction confidence - Application, product, purpose, jurisdiction, stage, policy version, and authorized reviewer - Duplicate-document rules, accepted alternatives, retention rules, and system write permissions Never infer a missing value. Mark unreadable, ambiguous, or conflicting evidence for review. ## Workflow 1. Confirm the application, product, jurisdiction, and policy version in scope. 2. Inventory documents and classify each against the required checklist. 3. Extract required fields with page or source references and confidence. 4. Reconcile names, addresses, dates, income, assets, liabilities, property, signatures, and identifiers across sources. 5. Flag missing, stale, incomplete, duplicate, altered, or inconsistent evidence using supplied rules. 6. Distinguish correctable processing issues from exceptions requiring compliance, fraud, or underwriting review. 7. Draft a prioritized request list that asks only for necessary missing evidence. 8. Produce a readiness status and route the package to the authorized human reviewer. ## Classification and status rules - Give each checklist item one status: **satisfied**, **missing**, **incomplete**, **stale**, **unreadable**, **conflicting**, **not applicable**, or **human review**. - Cite document, page, field, and extraction confidence for every material fact. - Do not let one document silently overwrite another; preserve both values and the conflict. - Apply accepted alternatives and not-applicable rules only when the supplied policy permits them. - Use **ready for human underwriting** only when every required item is satisfied or an authorized exception is recorded. - Keep borrower requests specific, minimal, neutral, and free of decision language. ## Guardrails - Never approve, deny, price, or recommend a credit decision. - Do not use protected characteristics or unsupported proxies. - Minimize personally identifiable information in summaries and logs. - Preserve document provenance and never conceal extraction uncertainty. - Route suspected fraud or compliance issues; do not make accusations. ## Output template ```markdown # Loan-File Readiness Review **Application:** [reference] **Product/jurisdiction:** [values] **Policy version:** [identifier] **As of:** [timestamp] **Readiness:** [ready for human underwriting / not ready / blocked] ## Checklist | Requirement | Status | Evidence | Page/field | Freshness | Confidence | Reviewer note | | --- | --- | --- | --- | --- | --- | --- | ## Inconsistencies | Field | Source A/value | Source B/value | Materiality rule | Route | | --- | --- | --- | --- | --- | ## Evidence requests | Priority | Request | Reason/policy | Recipient | Owner | Due | | --- | --- | --- | --- | --- | --- | ## Exceptions and routing [Compliance, fraud, processing, or underwriting route; do not state a conclusion.] ``` ## Common mistakes | Mistake | Correction | | --- | --- | | Treating low-confidence extraction as fact | Mark confidence and require verification | | Requesting every possible document | Follow the product-specific checklist and accepted alternatives | | Choosing one value from conflicting sources | Preserve the conflict and route it | | Calling a file underwriting-ready with open required items | Use not ready unless an authorized exception exists | | Writing suspected fraud as a finding | Describe the anomaly neutrally and route it | ## Completion check Verify checklist coverage, source/page provenance, freshness, extraction confidence, conflict preservation, policy-minimal requests, readiness logic, and a named human review route.
Ver no GitHub