shadow-loan-council

# Shadow Loan Council The **pure-compute** verdict layer of Shadow's compliance council. Takes a typed loan dict + policy + optional AML/KYC signals; returns a `{verdict, voices, adverse_action_codes, traceability, attestation, ...}` bundle in milliseconds. NO LLM call inside the tool body — every threshold is hardcoded to a checked-in policy artifact. ## When to use Install this skill in Claude Desktop / Cursor / OpenCode when you want Claude to: - Get an immediate policy-driven verdict on a loan application without waiting for a stochastic LLM call - Verify what Shadow's council would say against your own reasoning (audit path) - Generate the adverse-action code + borrower-readable notice text from a signed reason-code dictionary - Screen an AML/KYC risk with specific rule citations (BSA, OFAC, USA PATRIOT §326, FinCEN CDD, FATF) - Get an Ed25519-signed attestation binding the loan input to the exact verdict, so the bank auditor can verify with the public key ## What the response contains ```json { "final_verdict": "approve" | "escalate" | "block", "confidence_weighted_verdict": "approve" | "escalate" | "block", "aggregated_score": -1.0 to 1.0, "voice_contributions": [{ voice, weight, confidence, score }], "voices": [ { voice, verdict, confidence, rationale, adverse_action_codes, metrics } ], "presentation_order": [3, 0, 1, 5, 2, 4], "risk_packet": { var_95_10d, es_95_10d, concentration, sector_exposure, risk_budget_status }, "adverse_action_codes": [ ...