Imported from jeremylongshore/tons-of-skills-marketplace (plugins/saas-packs/clay-pack/skills/clay-policy-guardrails/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill clay-policy-guardrails. Copyright stays with the author (MIT).

Clay Policy Guardrails

Overview

Policy enforcement and guardrails for Clay data enrichment pipelines. Clay processes personal and business data at scale, requiring strict controls around credit spending, data privacy compliance, input validation, and export restrictions.

Prerequisites

• Clay integration in production or pre-production
• Understanding of GDPR/CCPA requirements
• Credit budget defined by management
• Data classification policy for your organization

Instructions

Step 1: Credit Spending Guardrails

// src/clay/policies/credit-policy.ts
interface CreditPolicy {
  dailyLimit: number;
  perTableLimit: number;
  perBatchLimit: number;
  alertThresholdPct: number;
  hardStopEnabled: boolean;
}

const CREDIT_POLICIES: Record<string, CreditPolicy> = {
  conservative: {
    dailyLimit: 200,
    perTableLimit: 500,
    perBatchLimit: 100,
    alertThresholdPct: 70,
    hardStopEnabled: true,
  },
  standard: {
    dailyLimit: 500,
    perTableLimit: 2000,
    perBatchLimit: 500,
    alertThresholdPct: 80,
    hardStopEnabled: true,
  },
  aggressive: {
    dailyLimit: 2000,
    perTableLimit: 10000,
    perBatchLimit: 2000,
    alertThresholdPct: 90,
    hardStopEnabled: false, // Alert only, don't stop
  },
};

class CreditPolicyEnforcer {
  private dailyUsed = 0;
  private tableUsage = new Map<string, number>();

  constructor(private policy: CreditPolicy) {}

  checkBatch(tableId: string, rowCount: number, creditsPerRow: number): {
    allowed: boolean;
    reason?: string;
  } {
    const estimated = rowCount * creditsPerRow;

    // Batch limit
    if (estimated > this.policy.perBatchLimit) {
      return {
        allowed: false,
        reason: `Batch (${estimated} credits) exceeds per-batch limit (${this.policy.perBatchLimit}). Split into smaller batches.`,
      };
    }

    // Daily limit
    if (this.dailyUsed + estimated > this.policy.dailyLimit) {
      if (this.policy.hardStopEnabled) {
        return {
          allowed: false,
          reason: `Would exceed daily limit: ${this.dailyUsed} + ${estimated} > ${this.policy.dailyLimit}`,
        };
      }
      console.warn(`WARNING: Exceeding daily limit (${this.dailyUsed + estimated}/${this.policy.dailyLimit})`);
    }

    // Per-table limit
    const tableTotal = (this.tableUsage.get(tableId) || 0) + estimated;
    if (tableTotal > this.policy.perTableLimit) {
      return {
        allowed: false,
        reason: `Table ${tableId} would exceed limit: ${tableTotal} > ${this.policy.perTableLimit}`,
      };
    }

    // Alert threshold
    const dailyPct = ((this.dailyUsed + estimated) / this.policy.dailyLimit) * 100;
    if (dailyPct > this.policy.alertThresholdPct) {
      console.warn(`Credit alert: ${dailyPct.toFixed(0)}% of daily limit used`);
    }

    return { allowed: true };
  }

  recordUsage(tableId: string, credits: number) {
    this.dailyUsed += credits;
    this.tableUsage.set(tableId, (this.tableUsage.get(tableId) || 0) + credits);
  }
}

Step 2: Data Privacy Guardrails

// src/clay/policies/privacy-policy.ts

// Fields that should NEVER be enriched or stored
const BLOCKED_ENRICHMENT_FIELDS = new Set([
  'ssn', 'social_security', 'tax_id',
  'date_of_birth', 'dob', 'birthday',
  'home_address', 'home_phone',
  'personal_phone', 'personal_mobile',
  'bank_account', 'credit_card',
  'medical_history', 'health_records',
  'salary', 'compensation',
  'political_affiliation', 'religion',
  'ethnic_origin', 'sexual_orientation',
]);

// Fields that require explicit consent
const CONSENT_REQUIRED_FIELDS = new Set([
  'personal_email', 'phone_number', 'mobile_phone',
]);

interface PrivacyCheckResult {
  allowed: boolean;
  violations: string[];
  warnings: string[];
}

function checkPrivacy(
  fieldsToEnrich: string[],
  hasExplicitConsent: boolean = false,
): PrivacyCheckResult {
  const violations: string[] = [];
  const warnings: string[] = [];

  for (const field of fieldsToEnrich) {
    const normalized = field.toLowerCase().replace(/[\s-]/g, '_');

    if (BLOCKED_ENRICHMENT_FIELDS.has(normalized)) {
      violations.push(`BLOCKED: "${field}" is a restricted field (never enrich)`);
    }

    if (CONSENT_REQUIRED_FIELDS.has(normalized) && !hasExplicitConsent) {
      warnings.push(`CONSENT: "${field}" requires explicit consent to enrich`);
    }
  }

  return {
    allowed: violations.length === 0,
    violations,
    warnings,
  };
}

Step 3: Input Validation Guardrails

// src/clay/policies/input-validation.ts

const PERSONAL_EMAIL_DOMAINS = new Set([
  'gmail.com', 'yahoo.com', 'hotmail.com', 'outlook.com', 'icloud.com',
  'aol.com', 'protonmail.com', 'mail.com', 'yandex.com', 'gmx.com',
]);

const DISPOSABLE_EMAIL_DOMAINS = new Set([
  'tempmail.com', 'guerrillamail.com', 'throwaway.email', 'yopmail.com',
  'mailinator.com', '10minutemail.com', 'trashmail.com',
]);

interface ValidationResult {
  valid: Record<string, unknown>[];
  rejected: { row: Record<string, unknown>; reason: string }[];
  stats: {
    total: number;
    valid: number;
    invalidDomain: number;
    personalDomain: number;
    disposableDomain: number;
    missingRequiredField: number;
    duplicates: number;
  };
}

function validateBatch(
  rows: Record<string, unknown>[],
  requiredFields: string[] = ['domain'],
): ValidationResult {
  const seen = new Set<string>();
  const stats = {
    total: rows.length, valid: 0, invalidDomain: 0,
    personalDomain: 0, disposableDomain: 0, missingRequiredField: 0, duplicates: 0,
  };
  const valid: Record<string, unknown>[] = [];
  const rejected: { row: Record<string, unknown>; reason: string }[] = [];

  for (const row of rows) {
    // Required fields check
    const missing = requiredFields.filter(f => !row[f]);
    if (missing.length > 0) {
      rejected.push({ row, reason: `Missing required: ${missing.join(', ')}` });
      stats.missingRequiredField++;
      continue;
    }

    const domain = String(row.domain || '').toLowerCase().trim();

    // Domain validation
    if (!domain.includes('.') || domain.length < 4) {
      rejected.push({ row, reason: `Invalid domain: "${domain}"` });
      stats.invalidDomain++;
      continue;
    }

    // Personal domain filter
    if (PERSONAL_EMAIL_DOMAINS.has(domain)) {
      rejected.push({ row, reason: `Personal email domain: ${domain}` });
      stats.personalDomain++;
      continue;
    }

    // Disposable domain filter
    if (DISPOSABLE_EMAIL_DOMAINS.has(domain)) {
      rejected.push({ row, reason: `Disposable email domain: ${domain}` });
      stats.disposableDomain++;
      continue;
    }

    // Deduplication
    const key = `${domain}:${String(row.first_name || '').toLowerCase()}:${String(row.last_name || '').toLowerCase()}`;
    if (seen.has(key)) {
      stats.duplicates++;
      continue;
    }
    seen.add(key);

    valid.push({ ...row, domain });
    stats.valid++;
  }

  return { valid, rejected, stats };
}

Step 4: Export Restrictions

// src/clay/policies/export-policy.ts
type ExportDestination = 'crm' | 'outreach' | 'analytics' | 'csv';

const EXPORT_RULES: Record<ExportDestination, {
  allowedFields: string[];
  blockedFields: string[];
  requiresApproval: boolean;
}> = {
  crm: {
    allowedFields: ['email', 'first_name', 'last_name', 'company_name', 'job_title', 'icp_score'],
    blockedFields: ['personal_email', 'home_address'],
    requiresApproval: false,
  },
  outreach: {
    allowedFields: ['email', 'first_name', 'company_name', 'personalized_opener'],
    blockedFields: ['phone_number', 'linkedin_url', 'personal_email'],
    requiresApproval: false,
  },
  analytics: {
    allowedFields: ['company_name', 'industry', 'employee_count', 'icp_score'],
    blockedFields: ['email', 'first_name', 'last_name', 'phone_number'],
    requiresApproval: false,
  },
  csv: {
    allowedFields: ['*'], // All fields
    blockedFields: ['personal_email', 'home_address', 'ssn'],
    requiresApproval: true, // Requires manager approval for CSV export
  },
};

function filterForExport(
  rows: Record<string, unknown>[],
  destination: ExportDestination,
): Record<string, unknown>[] {
  const rules = EXPORT_RULES[destination];

  return rows.map(row => {
    const filtered: Record<string, unknown> = {};
    for (const [key, value] of Object.entries(row)) {
      if (rules.blockedFields.includes(key)) continue;
      if (rules.allowedFields[0] !== '*' && !rules.allowedFields.includes(key)) continue;
      filtered[key] = value;
    }
    return filtered;
  });
}

Error Handling

Output

Create an enforceable policy record with approved purpose, input quality rules, credit limits, blocked/sensitive fields, permitted destinations, approval requirements, audit owner, and exception expiry. Policy code must fail closed: an unknown destination, missing consent/purpose, or absent approval cannot silently export or enrich a record.

Examples

For a CSV export request, apply the destination-specific field filter, require the documented manager approval, and preserve a redacted audit decision. If a row includes a blocked field or has no qualifying purpose, suppress it and notify the requester rather than dropping the policy check to complete a batch.

Resources

• GDPR Official Text
• CCPA Requirements
• Clay Community

Next Steps

For architecture patterns at scale, see clay-architecture-variants.