Adobe Policy & Guardrails

Overview

Automated policy enforcement for Adobe integrations: credential pattern scanning (Adobe OAuth secrets use p8_ prefix), Firefly content policy pre-screening, PDF Services quota guardrails, and OAuth scope validation.

Prerequisites

ESLint configured in project
CI/CD pipeline (GitHub Actions)
Understanding of Adobe credential patterns

Instructions

Guardrail 1: Adobe Credential Pattern Scanner

Adobe OAuth Server-to-Server secrets follow the p8_ prefix pattern:

# .github/workflows/adobe-security.yml
name: Adobe Security Scan
on: [push, pull_request]
jobs:
  credential-scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Scan for Adobe credential patterns
        run: |
          EXIT_CODE=0

          # Adobe OAuth client secrets (p8_ prefix)
          if grep -rE "p8_[A-Za-z0-9_-]{20,}" --include="*.ts" --include="*.js" --include="*.py" --include="*.json" --include="*.yaml" --include="*.yml" . 2>/dev/null | grep -v node_modules | grep -v '.git'; then
            echo "::error::Adobe client_secret pattern (p8_) found in source code"
            EXIT_CODE=1
          fi

          # Adobe IMS access tokens (JWT format)
          if grep -rE "eyJ[A-Za-z0-9_-]{100,}\.[A-Za-z0-9_-]{100,}" --include="*.ts" --include="*.js" . 2>/dev/null | grep -v node_modules | grep -v '.git' | grep -v '\.test\.' | grep -v '__mock'; then
            echo "::warning::Potential Adobe access token found in source (may be test fixture)"
          fi

          # Org IDs (format: HEXSTRING@AdobeOrg)
          if grep -rE "[A-F0-9]{24}@AdobeOrg" --include="*.ts" --include="*.js" --include="*.json" . 2>/dev/null | grep -v node_modules | grep -v '.git' | grep -v '.env.example'; then
            echo "::warning::Adobe Org ID found in source — consider using env var"
          fi

          exit $EXIT_CODE

Guardrail 2: Firefly Content Policy Pre-Screener

// src/adobe/guardrails/content-policy.ts
// Pre-screen prompts before sending to Firefly API to avoid wasted credits

interface ContentPolicyResult {
  allowed: boolean;
  violations: string[];
  suggestions: string[];
}

const CONTENT_RULES = [
  {
    name: 'real-people',
    pattern: /\b(photo of|portrait of|picture of)\s+(a\s+)?(real|actual|specific)\s+(person|man|woman|child)/i,
    message: 'Firefly cannot generate images of specific real people',
    suggestion: 'Use generic descriptions like "a professional in a business suit"',
  },
  {
    name: 'trademarks',
    pattern: /\b(nike|adidas|apple|google|microsoft|disney|marvel|coca.?cola|pepsi|starbucks|mcdonalds)\b/i,
    message: 'Firefly will reject prompts containing brand trademarks',
    suggestion: 'Use generic descriptions like "athletic shoes" or "tech company logo style"',
  },
  {
    name: 'explicit-content',
    pattern: /\b(nude|naked|explicit|pornograph|gore|violent|bloody|graphic death)\b/i,
    message: 'Firefly rejects explicit or violent content',
    suggestion: 'Use appropriate imagery descriptions',
  },
  {
    name: 'celebrity',
    pattern: /\b(celebrity|famous|actor|actress|politician|president|singer|musician)\s+(name|like|resembling)/i,
    message: 'Firefly cannot generate images of identifiable celebrities',
    suggestion: 'Describe the style or aesthetic without naming individuals',
  },
];

export function screenFireflyPrompt(prompt: string): ContentPolicyResult {
  const violations: string[] = [];
  const suggestions: string[] = [];

  for (const rule of CONTENT_RULES) {
    if (rule.pattern.test(prompt)) {
      violations.push(`[${rule.name}] ${rule.message}`);
      suggestions.push(rule.suggestion);
    }
  }

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

// Usage in API layer
export function guardFireflyPrompt(prompt: string): void {
  const result = screenFireflyPrompt(prompt);
  if (!result.allowed) {
    throw new Error(
      `Firefly content policy pre-check failed:\n` +
      result.violations.join('\n') +
      '\n\nSuggestions:\n' +
      result.suggestions.join('\n')
    );
  }
}

Guardrail 3: PDF Services Quota Enforcement

// src/adobe/guardrails/pdf-quota.ts
// Enforce PDF Services monthly transaction limits

class PdfQuotaGuard {
  private monthlyLimit: number;
  private transactionsUsed: number = 0;
  private monthStart: Date;

  constructor(tier: 'free' | 'paid' = 'free') {
    this.monthlyLimit = tier === 'free' ? 500 : Infinity;
    this.monthStart = new Date(new Date().getFullYear(), new Date().getMonth(), 1);
  }

  check(): { allowed: boolean; remaining: number; warning: boolean } {
    // Reset counter on new month
    const now = new Date();
    if (now.getMonth() !== this.monthStart.getMonth()) {
      this.transactionsUsed = 0;
      this.monthStart = new Date(now.getFullYear(), now.getMonth(), 1);
    }

    const remaining = this.monthlyLimit - this.transactionsUsed;
    return {
      allowed: remaining > 0,
      remaining,
      warning: remaining < this.monthlyLimit * 0.2,
    };
  }

  record(): void {
    const status = this.check();
    if (!status.allowed) {
      throw new Error(`PDF Services quota exhausted (${this.monthlyLimit} transactions/month)`);
    }
    this.transactionsUsed++;
    if (status.warning) {
      console.warn(`PDF Services: ${status.remaining - 1} transactions remaining this month`);
    }
  }
}

export const pdfQuota = new PdfQuotaGuard(
  process.env.ADOBE_PDF_TIER === 'paid' ? 'paid' : 'free'
);

Guardrail 4: OAuth Scope Validation

// Verify that the requested scopes match what the environment should use
function validateAdobeScopes(scopes: string, environment: string): void {
  const scopeList = scopes.split(',').map(s => s.trim());

  // Development should only have minimal scopes
  if (environment === 'development') {
    const prodOnlyScopes = ['ff_apis'];
    const violations = scopeList.filter(s => prodOnlyScopes.includes(s));
    if (violations.length > 0) {
      console.warn(`Adobe scope warning: ${violations.join(', ')} should not be in development`);
    }
  }

  // Required scopes that should always be present
  const required = ['openid', 'AdobeID'];
  const missing = required.filter(s => !scopeList.includes(s));
  if (missing.length > 0) {
    throw new Error(`Adobe required scopes missing: ${missing.join(', ')}`);
  }
}

Guardrail 5: Runtime Operation Guard

// Prevent dangerous operations based on environment
const BLOCKED_IN_PROD: Record<string, string> = {
  'delete-all-assets': 'Mass deletion blocked in production',
  'reset-quota-counter': 'Quota reset blocked in production',
  'use-test-credentials': 'Test credentials blocked in production',
};

function guardAdobeOperation(operation: string): void {
  const isProd = process.env.NODE_ENV === 'production';
  if (isProd && BLOCKED_IN_PROD[operation]) {
    throw new Error(`BLOCKED: ${BLOCKED_IN_PROD[operation]}`);
  }
}

Output

CI secret scanning for Adobe credential patterns (p8_, JWTs, Org IDs)
Firefly prompt pre-screening avoiding wasted credits on policy violations
PDF Services quota enforcement with monthly tracking
OAuth scope validation per environment
Runtime operation guards for production safety

Error Handling

Issue	Cause	Solution
Secret scan false positive	Test fixture contains pattern	Exclude test dirs from scan
Prompt wrongly rejected	Pattern too broad	Refine regex; allow legitimate uses
Quota counter reset	Server restart	Persist counter in Redis/DB
Scope validation fails	Wrong env var	Check `NODE_ENV` and `ADOBE_SCOPES`

Resources

Next Steps

For architecture blueprints, see adobe-architecture-variants.

adobe-policy-guardrails