KI-AGENT: ZUGFeRD- und E-Rechnungserkennung ergänzen

backend/src/modules/cron/prepareIncomingInvoices.ts

@@ -1,6 +1,8 @@
 import { FastifyInstance } from "fastify"
 import dayjs from "dayjs"
 import { getInvoiceDataFromGPT } from "../../utils/gpt"
+import { loadFileBuffer } from "../../utils/fileBuffer"
+import { detectElectronicInvoice } from "../einvoice/detectElectronicInvoice"
 
 // Drizzle schema
 import {
@@ -90,7 +92,32 @@ export function prepareIncomingInvoices(server: FastifyInstance) {
             for (const file of filesRes) {
                 console.log(`Processing file ${file.id} for tenant ${tenantId}`)
 
-                const data = await getInvoiceDataFromGPT(server,file, tenantId)
+                let fileData: Buffer
+
+                try {
+                    fileData = await loadFileBuffer(file.path!)
+                } catch (error) {
+                    server.log.error(error, `Datei ${file.id} konnte nicht aus S3 geladen werden.`)
+                    continue
+                }
+
+                const electronicInvoice = await detectElectronicInvoice(fileData, file)
+
+                if (electronicInvoice) {
+                    server.log.info({
+                        fileId: file.id,
+                        container: electronicInvoice.container,
+                        syntax: electronicInvoice.syntax,
+                        attachmentName: electronicInvoice.attachmentName,
+                    }, "Strukturierte E-Rechnung erkannt.")
+
+                    if (electronicInvoice.container === "xml") {
+                        server.log.warn({ fileId: file.id }, "Die Auswertung eigenständiger E-Rechnungs-XML folgt in Etappe 2.")
+                        continue
+                    }
+                }
+
+                const data = await getInvoiceDataFromGPT(server, file, tenantId, fileData)
 
                 if (!data) {
                     server.log.warn(`GPT returned no data for file ${file.id}`)

backend/src/modules/einvoice/detectElectronicInvoice.ts

@@ -0,0 +1,141 @@
+import {
+    decodePDFRawStream,
+    PDFDict,
+    PDFDocument,
+    PDFHexString,
+    PDFName,
+    PDFRawStream,
+    PDFString,
+} from "pdf-lib"
+
+export type ElectronicInvoiceSyntax = "cii" | "ubl-invoice" | "ubl-credit-note"
+
+export type ElectronicInvoiceDetection = {
+    container: "pdf" | "xml"
+    syntax: ElectronicInvoiceSyntax
+    xml: Buffer
+    attachmentName: string | null
+}
+
+type FileLike = {
+    name?: string | null
+    path?: string | null
+    mimeType?: string | null
+}
+
+const decodePdfText = (value: unknown): string | null => {
+    if (value instanceof PDFString || value instanceof PDFHexString) {
+        return value.decodeText()
+    }
+
+    return null
+}
+
+const detectXmlSyntax = (xml: Buffer): ElectronicInvoiceSyntax | null => {
+    const sample = xml.subarray(0, Math.min(xml.length, 64 * 1024)).toString("utf8")
+
+    if (/<(?:[A-Za-z_][\w.-]*:)?CrossIndustryInvoice(?:\s|>)/i.test(sample)) {
+        return "cii"
+    }
+
+    if (
+        /<Invoice(?:\s|>)/i.test(sample)
+        && /urn:oasis:names:specification:ubl:schema:xsd:Invoice-2/i.test(sample)
+    ) {
+        return "ubl-invoice"
+    }
+
+    if (
+        /<CreditNote(?:\s|>)/i.test(sample)
+        && /urn:oasis:names:specification:ubl:schema:xsd:CreditNote-2/i.test(sample)
+    ) {
+        return "ubl-credit-note"
+    }
+
+    return null
+}
+
+const extractXmlFromFileSpec = (fileSpec: PDFDict): { name: string | null, xml: Buffer } | null => {
+    const context = fileSpec.context
+    const embeddedFiles = context.lookupMaybe(fileSpec.get(PDFName.of("EF")), PDFDict)
+
+    if (!embeddedFiles) return null
+
+    const stream = context.lookup(
+        embeddedFiles.get(PDFName.of("UF")) || embeddedFiles.get(PDFName.of("F")),
+    )
+
+    if (!(stream instanceof PDFRawStream)) return null
+
+    const name = decodePdfText(
+        context.lookup(fileSpec.get(PDFName.of("UF")) || fileSpec.get(PDFName.of("F"))),
+    )
+    const contents = Buffer.from(decodePDFRawStream(stream).decode())
+
+    return { name, xml: contents }
+}
+
+const detectEmbeddedInvoice = async (pdf: Buffer): Promise<ElectronicInvoiceDetection | null> => {
+    let document: PDFDocument
+
+    try {
+        document = await PDFDocument.load(pdf, {
+            ignoreEncryption: true,
+            updateMetadata: false,
+        })
+    } catch {
+        return null
+    }
+
+    for (const [, object] of document.context.enumerateIndirectObjects()) {
+        if (!(object instanceof PDFDict)) continue
+
+        const type = object.get(PDFName.of("Type"))
+        if (!(type instanceof PDFName) || type.asString() !== "/Filespec") continue
+
+        const embedded = extractXmlFromFileSpec(object)
+        if (!embedded) continue
+
+        const syntax = detectXmlSyntax(embedded.xml)
+        if (!syntax) continue
+
+        return {
+            container: "pdf",
+            syntax,
+            xml: embedded.xml,
+            attachmentName: embedded.name,
+        }
+    }
+
+    return null
+}
+
+const hasPdfFileType = (file: FileLike, data: Buffer) => {
+    const filename = String(file.name || file.path || "").toLowerCase()
+    const mimeType = String(file.mimeType || "").toLowerCase().split(";")[0].trim()
+
+    return filename.endsWith(".pdf") || mimeType === "application/pdf" || data.subarray(0, 5).toString("ascii") === "%PDF-"
+}
+
+export const detectElectronicInvoice = async (
+    data: Buffer,
+    file: FileLike,
+): Promise<ElectronicInvoiceDetection | null> => {
+    if (hasPdfFileType(file, data)) {
+        return detectEmbeddedInvoice(data)
+    }
+
+    const beginsLikeXml = data.subarray(0, 1024).toString("utf8").replace(/^\uFEFF/, "").trimStart().startsWith("<")
+    const syntax = beginsLikeXml ? detectXmlSyntax(data) : null
+
+    if (syntax) {
+        return {
+            container: "xml",
+            syntax,
+            xml: data,
+            attachmentName: file.name || file.path?.split("/").pop() || null,
+        }
+    }
+
+    return null
+}