Merge pull request 'feat(selfhost): show sanitized update logs in admin UI' (#231) from dev into main

Reviewed-on: #231

agents/fedeo-device-agent/.env.example

@@ -10,6 +10,9 @@ FEDEO_SCAN_MODE=Color
 FEDEO_SCAN_SOURCE=
 FEDEO_SCAN_TIMEOUT_SECONDS=900
 FEDEO_SCAN_POSTPROCESS=false
+FEDEO_SCAN_BACKGROUND_CLEANUP=true
+FEDEO_SCAN_OCR=true
+FEDEO_SCAN_OCR_LANGUAGE=deu+eng
 FEDEO_SCAN_POSTPROCESS_PROFILE=document
 FEDEO_SCAN_POSTPROCESS_PYTHON=
 FEDEO_SCAN_POSTPROCESS_STRICT=false

agents/fedeo-device-agent/Dockerfile

@@ -13,6 +13,9 @@ FROM node:20-bookworm-slim AS runtime
 ENV NODE_ENV=production
 ENV FEDEO_WORK_DIR=/work
 ENV FEDEO_SCAN_POSTPROCESS=false
+ENV FEDEO_SCAN_BACKGROUND_CLEANUP=true
+ENV FEDEO_SCAN_OCR=true
+ENV FEDEO_SCAN_OCR_LANGUAGE=deu+eng
 ENV FEDEO_SCAN_POSTPROCESS_PROFILE=document
 ENV FEDEO_SCAN_POSTPROCESS_PYTHON=/opt/fedeo-device-agent/.venv-opencv/bin/python
 ENV FEDEO_SCAN_POSTPROCESS_STRICT=false
@@ -28,6 +31,9 @@ RUN apt-get update \
         python3-pip \
         python3-venv \
         sane-utils \
+        tesseract-ocr \
+        tesseract-ocr-deu \
+        tesseract-ocr-eng \
     && rm -rf /var/lib/apt/lists/*
 
 COPY requirements-opencv.txt ./

agents/fedeo-device-agent/README.md

@@ -44,11 +44,11 @@ FEDEO_TEST_POSTGRES_IMAGE=postgres:16-alpine node --import tsx tests/instanceAge
 ### macOS
 
 ```bash
-brew install node sane-backends
+brew install node sane-backends tesseract tesseract-lang
 scanimage -L
 ```
 
-Für die Dateierstellung wird Python mit Pillow benötigt, beispielsweise über `npm run setup:opencv` (siehe unten).
+Für die Dateierstellung und Hintergrundbereinigung wird Python mit Pillow und OpenCV benötigt, beispielsweise über `npm run setup:opencv` (siehe unten). `tesseract --list-langs` muss für die Standardkonfiguration `deu` und `eng` enthalten.
 
 Drucken nutzt später das macOS-Drucksystem/CUPS:
 
@@ -60,7 +60,7 @@ lpstat -p
 
 ```bash
 sudo apt update
-sudo apt install -y nodejs npm sane-utils python3-pil
+sudo apt install -y nodejs npm sane-utils python3-pil python3-opencv tesseract-ocr tesseract-ocr-deu tesseract-ocr-eng
 scanimage -L
 ```
 
@@ -90,6 +90,9 @@ Der iX1600 wird vom SANE-Backend `fujitsu` per USB unterstützt. Die USB-ID ist
    FEDEO_SCAN_RESOLUTION=300
    FEDEO_SCAN_MODE=Color
    FEDEO_SCAN_POSTPROCESS=false
+   FEDEO_SCAN_BACKGROUND_CLEANUP=true
+   FEDEO_SCAN_OCR=true
+   FEDEO_SCAN_OCR_LANGUAGE=deu+eng
    FEDEO_SCAN_TIMEOUT_SECONDS=900
    ```
 
@@ -139,7 +142,24 @@ npm run build:installer
 
 `npm test` prüft mit `check:installer`, dass das Paket dem aktuellen Quellcode entspricht. Das Paket enthält keine Zugangsdaten; diese ergänzt erst der Browser beim Download. Die Installer-Tests verwenden temporäre Verzeichnisse und simulierte Systembefehle.
 
-## OpenCV-Nachbearbeitung
+## Scannerhintergrund und OCR
+
+Hintergrundbereinigung und OCR sind standardmäßig aktiv und lassen sich in **Administration → Scanner** als Vorgabe oder im einzelnen Scan-Dialog einstellen:
+
+- **Scannerhintergrund weiß machen:** Bei einem kleinen Bon oder Dokument werden sicher erkannte Flächen außerhalb des Papiers weiß. Seitengröße, Position und Inhalt des Papiers bleiben erhalten. Unsichere Papiergrenzen werden nicht verändert. Diese Verarbeitung ist unabhängig vom zusätzlichen Zuschnitt.
+- **Durchsuchbares PDF (OCR):** Der Agent erkennt nach der Bildverarbeitung Text mit Tesseract und bettet ihn direkt in das PDF ein, bevor er es hochlädt. Das gilt auch für normale Dateien ohne Eingangsbeleg. Deutsch und Englisch sind vorausgewählt; PNG und TIFF erhalten keine OCR-Textschicht. Die Darstellung bleibt das gescannte Bild, ergänzt um auswählbaren und durchsuchbaren Text ([Tesseract-Dokumentation](https://tesseract-ocr.github.io/tessdoc/Command-Line-Usage.html)).
+
+```env
+FEDEO_SCAN_BACKGROUND_CLEANUP=true
+FEDEO_SCAN_OCR=true
+FEDEO_SCAN_OCR_LANGUAGE=deu+eng
+```
+
+Nach einem Update von FEDEO bestehende Agenten über **Einrichtung öffnen** mit dem neuen Linux-Installer aktualisieren. Dieser installiert auch OpenCV, Tesseract und die Sprachpakete. Bei manueller Installation die oben genannten Pakete und die aktuelle Agent-Software installieren und den Dienst neu starten. Der Agent meldet die verfügbare Verarbeitung beim Start an FEDEO; fehlende Fähigkeiten werden vor dem Auftrag gemeldet. Zusätzlich prüft er die benötigten Pakete und die gewählte OCR-Sprache vor jedem Papiereinzug. Ein OCR-Fehler erzeugt keinen erfolgreichen Scan ohne Textschicht; die Rohseiten bleiben zur Wiederherstellung erhalten.
+
+Die OCR läuft innerhalb des Scan-Auftrags nach dem Einzug. Bei langen Stapeln kann dieser Schritt länger dauern; das Zeitlimit für die Dateierstellung wird ebenfalls über `FEDEO_SCAN_TIMEOUT_SECONDS` gesetzt. Leere Rückseiten und Seitenreihenfolge bleiben erhalten.
+
+## Zusätzliche OpenCV-Nachbearbeitung
 
 Für automatischen Zuschnitt, leichte Entzerrung, Rotation und Kontrastkorrektur kann die OpenCV-Pipeline aktiviert werden.
 
@@ -156,7 +176,7 @@ FEDEO_SCAN_POSTPROCESS_PYTHON=/pfad/zum/agent/.venv-opencv/bin/python
 FEDEO_SCAN_POSTPROCESS_STRICT=false
 ```
 
-Wenn `FEDEO_SCAN_POSTPROCESS_PYTHON` leer bleibt, verwendet der Agent automatisch `.venv-opencv/bin/python`, sofern diese Umgebung existiert, sonst `python3`. Pillow ist auch ohne Bildkorrektur für die Dateierstellung erforderlich. Falls OpenCV nicht installiert ist und `FEDEO_SCAN_POSTPROCESS_STRICT=false` gesetzt ist, werden die Rohseiten im gewünschten Ausgabeformat zusammengeführt. Eine fehlgeschlagene Korrektur verliert keine Seiten. Die Warnung erscheint im Scan-Ergebnis. Beim iX1600-Profil ist die Bildkorrektur zunächst ausgeschaltet, damit der vollständige A4-Scan erhalten bleibt.
+Wenn `FEDEO_SCAN_POSTPROCESS_PYTHON` leer bleibt, verwendet der Agent automatisch `.venv-opencv/bin/python`, sofern diese Umgebung existiert, sonst `python3`. Pillow ist auch ohne Bildkorrektur für die Dateierstellung erforderlich. Falls die zusätzliche Korrektur fehlschlägt und `FEDEO_SCAN_POSTPROCESS_STRICT=false` gesetzt ist, werden die betroffenen Rohseiten weiterverarbeitet; die Warnung erscheint im Scan-Ergebnis. Fehlendes OpenCV kann nur bei ausgeschalteter Hintergrundbereinigung toleriert werden. Beim iX1600-Profil ist der zusätzliche Zuschnitt ausgeschaltet, damit der vollständige A4-Scan erhalten bleibt; Hintergrundbereinigung und OCR sind eingeschaltet.
 
 Profile:
 
@@ -166,7 +186,7 @@ Profile:
 
 ## Container-Betrieb
 
-Auf Linux und Raspberry Pi OS kann der Agent komplett im Container laufen. Dadurch bleiben Node.js, Python, OpenCV und SANE im Image. Auf dem Host werden dann nur Docker und Zugriff auf den USB-Scanner benötigt.
+Auf Linux und Raspberry Pi OS kann der Agent komplett im Container laufen. Dadurch bleiben Node.js, Python, OpenCV, Tesseract einschließlich Deutsch/Englisch und SANE im Image. Auf dem Host werden dann nur Docker und Zugriff auf den USB-Scanner benötigt.
 
 ```bash
 cp .env.example .env
@@ -202,7 +222,7 @@ npm test
 npm run test:python
 ```
 
-Die Tests benötigen Python 3 mit Pillow. Sie simulieren `scanimage`, erstellen echte PDFs und prüfen Duplex-/Simplex-Stapel, Seitenreihenfolge, Auflösung, leeren Einzug, Papierstau, beschädigte Seiten, Dateinamen und OpenCV-Ausfälle. Ein Test mit angeschlossenem iX1600 bleibt zusätzlich erforderlich.
+Die Tests benötigen Python 3 mit Pillow, NumPy und OpenCV. Für den OCR-Integrationstest sind Tesseract mit `deu`/`eng` und `pdftotext` aus Poppler erforderlich; ohne diese Programme wird er übersprungen. Bei Verwendung der virtuellen Umgebung zuerst `. .venv-opencv/bin/activate` ausführen. Die Tests simulieren `scanimage`, erstellen echte PDFs und prüfen Duplex-/Simplex-Stapel, Seitenreihenfolge, Auflösung, leeren Einzug, Papierstau, beschädigte Seiten, Dateinamen und OpenCV-Ausfälle. Bildtests prüfen die Hintergrundbereinigung bei geraden und schrägen Bons sowie den Erhalt von Text, farbigen Logos und vollständigen Seiten. Der OCR-Test liest die tatsächliche PDF-Textschicht einschließlich leerer Rückseiten aus. Ein Test mit angeschlossenem iX1600 bleibt zusätzlich erforderlich.
 
 Hardwareprüfung am 17.09.2026: ScanSnap iX1600 per USB unter Linux Mint 22.3 mit SANE 1.2.1. Zwei A4-Blätter wurden mit 300 dpi in Farbe beidseitig gescannt; das erzeugte PDF enthält vier A4-Seiten in der Reihenfolge Vorderseite/Rückseite. Alle vier Seiten wurden visuell geprüft. Der Test umfasste den lokalen Agenten bis zur PDF-Erstellung; der FEDEO-Upload wird separat mit einem simulierten Dateispeicher getestet.
 

agents/fedeo-device-agent/scripts/build-installer.mjs

@@ -14,7 +14,7 @@ async function filesUnder(directory) {
 const packageJson = JSON.parse(await readFile(path.join(root, 'package.json'), 'utf8'))
 const paths = [
   ...(await filesUnder('dist')).filter(file => file.endsWith('.js')),
-  'scripts/scan_document.py', 'scripts/opencv_postprocess.py', 'scripts/setup-opencv.sh',
+  'scripts/scan_document.py', 'scripts/scan_background.py', 'scripts/opencv_postprocess.py', 'scripts/setup-opencv.sh',
   'requirements-opencv.txt',
 ]
 const files = Object.fromEntries(await Promise.all(paths.map(async file => [file, (await readFile(path.join(root, file))).toString('base64')])))

agents/fedeo-device-agent/scripts/install-linux.sh

@@ -25,17 +25,22 @@ FEDEO_SETUP_UNIT="fedeo-device-agent-$FEDEO_SETUP_ID.service"
 FEDEO_SETUP_UNITS="${XDG_CONFIG_HOME:-$HOME/.config}/systemd/user"
 
 echo 'FEDEO Scan-Agent wird für diesen Benutzer eingerichtet.'
-if ! command -v node >/dev/null || ! command -v scanimage >/dev/null || ! python3 -c 'import PIL' >/dev/null 2>&1; then
+fedeo_processing_ready() {
+  python3 -c 'import PIL, cv2, numpy' >/dev/null 2>&1 &&
+    tesseract --list-langs 2>/dev/null | grep -qx deu &&
+    tesseract --list-langs 2>/dev/null | grep -qx eng
+}
+if ! command -v node >/dev/null || ! command -v scanimage >/dev/null || ! fedeo_processing_ready; then
   if ! command -v apt-get >/dev/null; then
-    echo 'Bitte Node.js >= 20, SANE (scanimage), Python 3 und Pillow installieren und das Skript erneut starten.' >&2
+    echo 'Bitte Node.js >= 20, SANE, Python/Pillow/OpenCV sowie Tesseract mit Deutsch/Englisch installieren und das Skript erneut starten.' >&2
     exit 1
   fi
   echo 'Fehlende Systempakete werden über apt installiert (sudo benötigt).'
   sudo apt-get update
-  sudo apt-get install -y nodejs sane-utils python3-pil
+  sudo apt-get install -y nodejs sane-utils python3-pil python3-opencv tesseract-ocr tesseract-ocr-deu tesseract-ocr-eng
 fi
 node -e 'if (Number(process.versions.node.split(".")[0]) < 20) { console.error("Node.js >= 20 wird benötigt. Bitte Node.js aktualisieren und die Installation erneut starten."); process.exit(1) }'
-python3 -c 'import PIL'
+fedeo_processing_ready || { echo 'Die benötigte Bildverarbeitung oder die OCR-Sprachpakete fehlen weiterhin.' >&2; exit 1; }
 
 FEDEO_SETUP_NODE="$(command -v node)"
 FEDEO_SETUP_HAS_SYSTEMD=false
@@ -81,7 +86,9 @@ def unit_quote(value):
 unit_path.write_text('\n'.join([
     '[Unit]', 'Description=FEDEO Scan-Agent', 'After=network-online.target', '',
     '[Service]', 'Type=simple',
-    'WorkingDirectory=' + unit_quote(agent_dir),
+    # WorkingDirectory takes a literal path, unlike Environment/ExecStart.
+    # Quotes would become part of the path; only systemd specifiers need escaping.
+    'WorkingDirectory=' + str(agent_dir).replace('%', '%%'),
     'Environment=' + unit_quote('FEDEO_AGENT_ENV=' + str(config_path)),
     'ExecStart=' + unit_quote(node) + ' ' + unit_quote(agent_dir / 'dist/main.js'),
     'Restart=on-failure', 'RestartSec=5', 'UMask=0077', '',

agents/fedeo-device-agent/scripts/scan_background.py

@@ -0,0 +1,74 @@
+"""Conservative paper masking. Keep the canvas and every pixel inside the paper."""
+from pathlib import Path
+
+import cv2
+import numpy as np
+
+
+def whiten_background(image):
+    height, width = image.shape[:2]
+    scale = min(1.0, 1000.0 / max(height, width))
+    small = cv2.resize(image, None, fx=scale, fy=scale, interpolation=cv2.INTER_AREA)
+    gray = cv2.cvtColor(small, cv2.COLOR_BGR2GRAY)
+    border = np.concatenate((gray[0], gray[-1], gray[:, 0], gray[:, -1]))
+    background = float(np.median(border))
+    # White surroundings already need no correction. Never search inside a
+    # white document for an isolated text box or logo to keep instead.
+    if background >= 235 or np.percentile(border, 75) - np.percentile(border, 25) > 55:
+        return image
+
+    threshold, _ = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
+    mask = np.uint8(gray > max(threshold, background + 20)) * 255
+    mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, np.ones((5, 5), np.uint8))
+    contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+    candidates = []
+    area = gray.size
+    for contour in contours:
+        contour_area = cv2.contourArea(contour)
+        if not 0.015 * area < contour_area < 0.985 * area:
+            continue
+        perimeter = cv2.arcLength(contour, True)
+        polygon = cv2.approxPolyDP(contour, 0.018 * perimeter, True)
+        if len(polygon) != 4 or not cv2.isContourConvex(polygon):
+            continue
+        # Use the quadrilateral only to recognize a sheet, then keep its full
+        # contour hull so rounded or slightly uneven edges cannot lose content.
+        polygon = cv2.convexHull(contour)
+        paper = np.zeros_like(gray)
+        cv2.fillPoly(paper, [polygon], 255)
+        inside = cv2.erode(paper, np.ones((7, 7), np.uint8)) > 0
+        ring = (cv2.dilate(paper, np.ones((9, 9), np.uint8)) > 0) & (paper == 0)
+        if inside.sum() < 100 or ring.sum() < 50:
+            continue
+        paper_light = float(np.median(gray[inside]))
+        # A paper edge must separate a mostly light sheet from darker scanner
+        # backing all around it. Ambiguous photos/pages are left untouched.
+        if paper_light < background + 30 or np.mean(gray[ring] < paper_light - 20) < 0.85:
+            continue
+        candidates.append(polygon)
+
+    if not candidates:
+        return image
+    keep_small = np.zeros_like(gray)
+    cv2.fillPoly(keep_small, candidates, 255)
+    # Do not erase another significant object that we could not classify.
+    outside = (keep_small == 0)
+    if np.mean(gray[outside] > background + 30) > 0.04:
+        return image
+    keep = np.zeros((height, width), dtype=np.uint8)
+    polygons = [np.rint(polygon / scale).astype(np.int32) for polygon in candidates]
+    cv2.fillPoly(keep, polygons, 255)
+    # Rounding/downsampling uncertainty grows with the original scan DPI.
+    padding = max(2, int(np.ceil(2 / scale)))
+    keep = cv2.dilate(keep, np.ones((2 * padding + 1, 2 * padding + 1), np.uint8))
+    result = image.copy()
+    result[keep == 0] = 255
+    return result
+
+
+def clean_scan_background(input_path: Path, output_path: Path):
+    image = cv2.imread(str(input_path), cv2.IMREAD_COLOR)
+    if image is None:
+        raise RuntimeError("Scan-Bild konnte nicht gelesen werden")
+    if not cv2.imwrite(str(output_path), whiten_background(image)):
+        raise RuntimeError("Bereinigtes Scan-Bild konnte nicht gespeichert werden")

agents/fedeo-device-agent/scripts/scan_document.py

@@ -1,21 +1,56 @@
 #!/usr/bin/env python3
-"""Assemble ordered scanner pages; OpenCV is optional, PDF creation is not."""
+"""Prepare ordered scanner pages and publish an optionally searchable PDF atomically."""
 import argparse
 import json
+import math
 from pathlib import Path
+import re
+import signal
+import subprocess
+import tempfile
 
-from PIL import Image
+from PIL import Image, PdfParser
 
 
-def create_document(inputs, output, resolution, use_postprocess=False, profile="document", strict=False):
+def run_ocr_command(arguments):
+    process = subprocess.Popen(arguments, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
+    try:
+        stdout, stderr = process.communicate()
+    except BaseException:
+        process.kill()
+        process.wait()
+        raise
+    if process.returncode != 0:
+        raise RuntimeError("OCR fehlgeschlagen: " + stderr.strip())
+    return stdout
+
+
+def check_ocr(language):
+    if not re.fullmatch(r"[a-zA-Z0-9_]+(?:\+[a-zA-Z0-9_]+)*", language):
+        raise ValueError("Ungültige OCR-Sprache")
+    try:
+        languages = run_ocr_command(["tesseract", "--list-langs"]).splitlines()
+    except FileNotFoundError as error:
+        raise RuntimeError("Tesseract fehlt. Bitte tesseract-ocr und die Sprachpakete tesseract-ocr-deu/tesseract-ocr-eng installieren.") from error
+    missing = set(language.split("+")) - {line.strip() for line in languages}
+    if missing:
+        raise RuntimeError("OCR-Sprachpakete fehlen: " + ", ".join(sorted(missing)))
+
+
+def create_document(inputs, output, resolution, use_postprocess=False, profile="document", strict=False,
+                    ocr=False, ocr_language="deu+eng", background_cleanup=False):
     if not inputs:
         raise ValueError("Keine Seiten vorhanden")
+    if not math.isfinite(resolution) or resolution <= 0:
+        raise ValueError("Ungültige Scan-Auflösung")
     output = Path(output)
     file_format = {".pdf": "PDF", ".png": "PNG", ".tiff": "TIFF"}.get(output.suffix.lower())
     if not file_format:
         raise ValueError("Unbekanntes Ausgabeformat")
-    if file_format != "PDF" and len(inputs) != 1:
-        raise ValueError("Mehrseitige Scans benötigen PDF als Ausgabeformat")
+    if (len(inputs) != 1 or ocr) and file_format != "PDF":
+        raise ValueError("Mehrseitige Scans und OCR benötigen PDF als Ausgabeformat")
+    if ocr:
+        check_ocr(ocr_language)
 
     warnings = []
     processor = None
@@ -27,47 +62,92 @@ def create_document(inputs, output, resolution, use_postprocess=False, profile="
             if strict:
                 raise
             warnings.append("OpenCV fehlt; alle Seiten wurden ohne Bildkorrektur übernommen.")
+    if background_cleanup:
+        from scan_background import clean_scan_background
 
     partial = output.with_name(output.name + ".partial")
     try:
+        with tempfile.TemporaryDirectory(prefix="prepared-", dir=output.parent) as directory:
+            prepared = []
             for index, input_path in enumerate(inputs):
-            input_path = Path(input_path)
-            image_path = input_path
-            corrected = input_path.with_name(input_path.stem + ".corrected.png")
+                image_path = Path(input_path)
+                corrected = Path(directory) / f"corrected-{index:06}.png"
                 if processor:
                     try:
-                    processor(input_path, corrected, profile)
+                        processor(image_path, corrected, profile)
                         image_path = corrected
                     except Exception:
                         if strict:
                             raise
                         warnings.append(f"Seite {index + 1}: Bildkorrektur fehlgeschlagen; Rohscan übernommen.")
+                if background_cleanup:
+                    cleaned = Path(directory) / f"cleaned-{index:06}.png"
+                    clean_scan_background(image_path, cleaned)
+                    image_path = cleaned
 
-            # Only one decoded page is kept in memory, even for large duplex stacks.
+                # Only one decoded page is kept in memory, even for duplex stacks.
                 with Image.open(image_path) as image:
                     with image.convert("RGB") as rgb:
-                    if file_format == "PDF":
+                        if ocr:
+                            page = Path(directory).resolve() / f"page-{index:06}.png"
+                            rgb.save(page, "PNG", dpi=(resolution, resolution))
+                            prepared.append(page)
+                        elif file_format == "PDF":
                             rgb.save(partial, "PDF", resolution=resolution, quality=90, append=index > 0)
                         else:
                             rgb.save(partial, file_format, dpi=(resolution, resolution))
                 corrected.unlink(missing_ok=True)
+                if background_cleanup:
+                    cleaned.unlink(missing_ok=True)
+            if ocr:
+                page_list = Path(directory) / "pages.txt"
+                if any("\n" in str(page) or "\r" in str(page) for page in prepared):
+                    raise ValueError("Das Scan-Arbeitsverzeichnis darf keinen Zeilenumbruch enthalten")
+                page_list.write_text("\n".join(map(str, prepared)) + "\n", encoding="utf8")
+                prefix = Path(directory) / "searchable"
+                run_ocr_command(["tesseract", str(page_list), str(prefix), "-l", ocr_language,
+                                 "--dpi", str(round(resolution)), "pdf"])
+                searchable = prefix.with_suffix(".pdf")
+                with PdfParser.PdfParser(filename=str(searchable)) as pdf:
+                    if len(pdf.pages) != len(inputs):
+                        raise RuntimeError("OCR-PDF ist unvollständig; Rohscans bleiben erhalten")
+                searchable.replace(partial)
             partial.replace(output)
     finally:
         partial.unlink(missing_ok=True)
-
     return {"pageCount": len(inputs), "warnings": warnings}
 
 
+def interrupted(*_):
+    raise SystemExit("Scan-Verarbeitung abgebrochen")
+
+
 def main():
+    # Let finally blocks stop Tesseract and remove incomplete PDFs on agent timeout.
+    signal.signal(signal.SIGTERM, interrupted)
     parser = argparse.ArgumentParser(description="Scan-Seiten zu einer Datei zusammenführen")
-    parser.add_argument("--inputs", nargs="+", required=True)
-    parser.add_argument("--output", required=True)
-    parser.add_argument("--resolution", type=float, required=True)
+    parser.add_argument("--inputs", nargs="+")
+    parser.add_argument("--output")
+    parser.add_argument("--resolution", type=float, default=300)
     parser.add_argument("--profile", choices=["document", "receipt", "raw"], default="document")
     parser.add_argument("--postprocess", action="store_true")
     parser.add_argument("--strict", action="store_true")
+    parser.add_argument("--background-cleanup", action="store_true")
+    parser.add_argument("--ocr", action="store_true")
+    parser.add_argument("--ocr-language", default="deu+eng")
+    parser.add_argument("--check-runtime", action="store_true")
     args = parser.parse_args()
-    print(json.dumps(create_document(args.inputs, args.output, args.resolution, args.postprocess, args.profile, args.strict)))
+    if args.check_runtime:
+        if args.background_cleanup or (args.postprocess and args.strict):
+            import cv2, numpy
+        if args.ocr:
+            check_ocr(args.ocr_language)
+        print(json.dumps({"ready": True}))
+        return
+    if not args.inputs or not args.output:
+        parser.error("--inputs und --output sind erforderlich")
+    print(json.dumps(create_document(args.inputs, args.output, args.resolution, args.postprocess, args.profile,
+                                    args.strict, args.ocr, args.ocr_language, args.background_cleanup)))
 
 
 if __name__ == "__main__":

agents/fedeo-device-agent/src/config.ts

@@ -62,6 +62,9 @@ export const loadConfig = (): AgentConfig => {
         scanSource: optional(process.env.FEDEO_SCAN_SOURCE),
         scanTimeoutSeconds: numberFromEnv(process.env.FEDEO_SCAN_TIMEOUT_SECONDS, 900),
         scanPostprocess: booleanFromEnv(process.env.FEDEO_SCAN_POSTPROCESS, false),
+        scanBackgroundCleanup: booleanFromEnv(process.env.FEDEO_SCAN_BACKGROUND_CLEANUP, true),
+        scanOcr: booleanFromEnv(process.env.FEDEO_SCAN_OCR, true),
+        scanOcrLanguage: optional(process.env.FEDEO_SCAN_OCR_LANGUAGE) || "deu+eng",
         postprocessProfile: postprocessProfileFromEnv(process.env.FEDEO_SCAN_POSTPROCESS_PROFILE),
         postprocessPython: optional(process.env.FEDEO_SCAN_POSTPROCESS_PYTHON) || defaultPostprocessPython(),
         postprocessStrict: booleanFromEnv(process.env.FEDEO_SCAN_POSTPROCESS_STRICT, false),

agents/fedeo-device-agent/src/main.ts

@@ -7,6 +7,7 @@ import { loadConfig } from "./config.js"
 import { log } from "./logger.js"
 import { listPrinters } from "./print/cups.js"
 import { hasSane, listScanners, runScan } from "./scan/sane.js"
+import { processingCapabilities } from "./scan/postprocess.js"
 import { AgentHeartbeat } from "./types.js"
 
 const sleep = (ms: number) => new Promise((resolve) => setTimeout(resolve, ms))
@@ -19,6 +20,7 @@ const stringifyError = (error: unknown) => {
 const main = async () => {
     const config = loadConfig()
     const api = new FedeoApi(config)
+    const processing = await processingCapabilities(config)
 
     log.info("FEDEO Geräte-Agent startet", {
         platform: process.platform,
@@ -38,6 +40,7 @@ const main = async () => {
             const heartbeatPayload: AgentHeartbeat = {
                 capabilities: {
                     scan: scanAvailable,
+                    ...processing,
                     print: printerNames.length > 0,
                     platform: process.platform,
                 },

agents/fedeo-device-agent/src/scan/postprocess.ts

@@ -6,15 +6,34 @@ import { runCommand } from "../commands.js"
 const agentRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "../..")
 const documentScript = path.join(agentRoot, "scripts/scan_document.py")
 
-export const checkScanRuntime = async (config: AgentConfig, postprocess: boolean) => {
-    const imports = postprocess && config.postprocessStrict ? "import PIL, cv2, numpy" : "import PIL"
+type ProcessingOptions = { ocr: boolean; ocrLanguage: string; backgroundCleanup: boolean }
+
+export const checkScanRuntime = async (config: AgentConfig, postprocess: boolean, options: ProcessingOptions) => {
+    const args = [documentScript, "--check-runtime", "--ocr-language", options.ocrLanguage]
+    if (postprocess) args.push("--postprocess")
+    if (config.postprocessStrict) args.push("--strict")
+    if (options.backgroundCleanup) args.push("--background-cleanup")
+    if (options.ocr) args.push("--ocr")
     try {
-        const result = await runCommand(config.postprocessPython, ["-c", imports], { timeoutMs: 10_000 })
+        const result = await runCommand(config.postprocessPython, args, { timeoutMs: 10_000 })
         if (result.code === 0) return
-    } catch {
+        throw new Error(result.stderr.trim())
+    } catch (error) {
         // Report missing runtimes before the scanner consumes any paper.
+        const detail = error instanceof Error ? error.message : String(error)
+        throw new Error("Scan-Verarbeitung nicht verfügbar. Bitte den Agenten mit Python/Pillow, OpenCV und Tesseract samt OCR-Sprachpaketen installieren. " + detail)
     }
-    throw new Error("Python/Pillow (bei strikter Nachbearbeitung zusätzlich OpenCV) fehlt. Bitte npm run setup:opencv ausführen oder python3-pil installieren.")
+}
+
+export const processingCapabilities = async (config: AgentConfig) => {
+    const available = async (options: ProcessingOptions) => {
+        try { await checkScanRuntime(config, false, options); return true } catch { return false }
+    }
+    const [ocr, backgroundCleanup] = await Promise.all([
+        available({ ocr: true, ocrLanguage: config.scanOcrLanguage, backgroundCleanup: false }),
+        available({ ocr: false, ocrLanguage: config.scanOcrLanguage, backgroundCleanup: true }),
+    ])
+    return { ocr, backgroundCleanup }
 }
 
 export const createScanDocument = async (
@@ -23,11 +42,14 @@ export const createScanDocument = async (
     outputPath: string,
     resolution: number,
     postprocess: boolean,
-    profile: AgentConfig["postprocessProfile"]
+    profile: AgentConfig["postprocessProfile"],
+    options: ProcessingOptions
 ): Promise<ScanResult> => {
     const args = [documentScript, "--output", outputPath, "--resolution", String(resolution), "--profile", profile]
     if (postprocess) args.push("--postprocess")
     if (config.postprocessStrict) args.push("--strict")
+    if (options.backgroundCleanup) args.push("--background-cleanup")
+    if (options.ocr) args.push("--ocr", "--ocr-language", options.ocrLanguage)
     args.push("--inputs", ...pages)
 
     const result = await runCommand(config.postprocessPython, args, {

agents/fedeo-device-agent/src/scan/sane.ts

@@ -20,7 +20,9 @@ export const hasSane = () => commandExists("scanimage")
 
 export const listScanners = async () => {
     if (!await hasSane()) return []
-    const result = await runCommand("scanimage", ["-L"], { timeoutMs: 10_000 })
+    // SANE also probes network backends; these can take more than ten seconds
+    // even when the USB scanner is already available.
+    const result = await runCommand("scanimage", ["-L"], { timeoutMs: 30_000 })
     if (result.code !== 0) return []
     return result.stdout.split(/\r?\n/)
         .map((line) => line.trim().match(/^device `([^']+)'/)?.[1])
@@ -51,7 +53,12 @@ export const runScan = async (
     const postprocess = booleanSetting(settings, "postprocess", config.scanPostprocess)
     const profile = stringSetting(settings, "postprocessProfile") || config.postprocessProfile
     if (!["document", "receipt", "raw"].includes(profile)) throw new Error("Unbekanntes Nachbearbeitungsprofil")
-    await checkScanRuntime(config, postprocess)
+    const processing = {
+        backgroundCleanup: booleanSetting(settings, "backgroundCleanup", config.scanBackgroundCleanup),
+        ocr: format === "pdf" && booleanSetting(settings, "ocr", config.scanOcr),
+        ocrLanguage: stringSetting(settings, "ocrLanguage") || config.scanOcrLanguage,
+    }
+    await checkScanRuntime(config, postprocess, processing)
 
     // Each physical scan has its own directory. A filename cannot escape it or overwrite another job.
     await mkdir(config.workDir, { recursive: true, mode: 0o700 })
@@ -109,8 +116,8 @@ export const runScan = async (
             throw new Error("Der Duplex-Scan ist unvollständig: Zu einer Vorderseite fehlt die Rückseite.")
         }
 
-        await reportProgress(`${pages.length} Seiten gescannt. Datei wird erstellt...`)
-        return await createScanDocument(config, pages, outputPath, resolution, postprocess, profile as AgentConfig["postprocessProfile"])
+        await reportProgress(`${pages.length} Seiten gescannt. ${processing.ocr ? "Durchsuchbares PDF mit OCR wird erstellt..." : "Datei wird erstellt..."}`)
+        return await createScanDocument(config, pages, outputPath, resolution, postprocess, profile as AgentConfig["postprocessProfile"], processing)
     } catch (error) {
         const message = error instanceof Error ? error.message : String(error)
         throw new Error(`${message} Rohscans bleiben zur Wiederherstellung in ${jobDir}.`)

agents/fedeo-device-agent/src/types.ts

@@ -11,6 +11,9 @@ export type AgentConfig = {
     scanSource?: string
     scanTimeoutSeconds: number
     scanPostprocess: boolean
+    scanBackgroundCleanup: boolean
+    scanOcr: boolean
+    scanOcrLanguage: string
     postprocessProfile: "document" | "receipt" | "raw"
     postprocessPython: string
     postprocessStrict: boolean
@@ -20,6 +23,8 @@ export type AgentHeartbeat = {
     capabilities: {
         scan: boolean
         print: boolean
+        ocr: boolean
+        backgroundCleanup: boolean
         platform: NodeJS.Platform
     }
     scannerNames: string[]

agents/fedeo-device-agent/tests/installer.test.mjs

@@ -14,6 +14,8 @@ const agentId = 'b48c0e33-1abd-4b11-89fa-1bb7b39a3057'
 const token = 'fedeo_agent_' + 'a'.repeat(64)
 const options = { agentId, token, backendUrl: 'https://fedeo.example.com/backend', autostart: true }
 const execute = promisify(execFile)
+const hasSystemdAnalyze = process.platform === 'linux' && await execute('systemd-analyze', ['--version'])
+  .then(() => true, error => { if (error.code === 'ENOENT') return false; throw error })
 
 test('backend URLs preserve reverse-proxy paths and reject credentials or non-HTTP protocols', () => {
   assert.equal(agentBackendUrl('/backend/', 'https://fedeo.example.com'), options.backendUrl)
@@ -50,11 +52,15 @@ for (const start of [false, true]) {
     await writeFile(path.join(bin, 'scanimage'), '#!/bin/sh\nexit 0\n', { mode: 0o755 })
     await writeFile(path.join(bin, 'id'), '#!/bin/sh\nprintf "1000\\n"\n', { mode: 0o755 })
     await writeFile(path.join(bin, 'systemctl'), '#!/bin/sh\nprintf "%s\\n" "$*" >> "$FEDEO_TEST_SYSTEMD_LOG"\n', { mode: 0o755 })
+    const realPython = (await execute('sh', ['-c', 'command -v python3'])).stdout.trim()
+    await writeFile(path.join(bin, 'python3'), '#!/bin/sh\nif [ "$1" = "-c" ] && [ "$2" = "import PIL, cv2, numpy" ]; then exit 0; fi\nexec "$FEDEO_TEST_PYTHON" "$@"\n', { mode: 0o755 })
+    await writeFile(path.join(bin, 'tesseract'), '#!/bin/sh\nprintf "deu\\neng\\n"\n', { mode: 0o755 })
     const env = {
       ...process.env, PATH: bin + ':' + process.env.PATH,
       XDG_DATA_HOME: path.join(temp, 'data with spaces %'),
       XDG_CONFIG_HOME: path.join(temp, 'config'), XDG_STATE_HOME: path.join(temp, 'state'),
       FEDEO_TEST_SYSTEMD_LOG: path.join(temp, 'systemd.log'),
+      FEDEO_TEST_PYTHON: realPython,
     }
     const scriptPath = path.join(temp, 'install.sh')
     await writeFile(scriptPath, renderAgentInstaller(bundle, options))
@@ -70,9 +76,13 @@ for (const start of [false, true]) {
     for (const [file, encoded] of Object.entries(bundle.files)) {
       assert.equal((await readFile(path.join(agentDir, file))).toString('base64'), encoded)
     }
-    const unit = await readFile(path.join(env.XDG_CONFIG_HOME, 'systemd/user', 'fedeo-device-agent-' + agentId + '.service'), 'utf8')
+    const unitPath = path.join(env.XDG_CONFIG_HOME, 'systemd/user', 'fedeo-device-agent-' + agentId + '.service')
+    const unit = await readFile(unitPath, 'utf8')
     assert.ok(unit.includes('data with spaces %%'))
     assert.ok(!unit.includes(token))
+    await t.test('systemd accepts the generated service, including a working directory with spaces and percent signs', { skip: !hasSystemdAnalyze }, async () => {
+      await execute('systemd-analyze', ['--user', 'verify', unitPath])
+    })
     const calls = await readFile(env.FEDEO_TEST_SYSTEMD_LOG, 'utf8')
     assert.equal(calls.includes('--user enable fedeo-device-agent-'), start)
     assert.equal(calls.includes('--user restart fedeo-device-agent-'), start)

agents/fedeo-device-agent/tests/scan.test.ts

@@ -3,16 +3,21 @@ import test, { TestContext } from "node:test"
 import os from "node:os"
 import path from "node:path"
 import { chmod, mkdtemp, readFile, readdir, rm, writeFile } from "node:fs/promises"
-import { runScan } from "../src/scan/sane.js"
+import { listScanners, runScan } from "../src/scan/sane.js"
 import { commandExists, runCommand } from "../src/commands.js"
 import { AgentConfig, ScanJob } from "../src/types.js"
 
 const fakeScanner = `#!/usr/bin/env python3
-import json, os, sys
+import json, os, sys, time
 from pathlib import Path
 from PIL import Image
 args = sys.argv[1:]
 Path(os.environ['SCAN_TEST_CALL']).write_text(json.dumps(args))
+if args == ['-L']:
+    time.sleep(float(os.environ.get('SCAN_TEST_DISCOVERY_DELAY', '0')))
+    print("device " + chr(96) + "fujitsu:ScanSnap iX1600:test' is a FUJITSU scanner")
+    print("device " + chr(96) + "airscan:test' is a network scanner")
+    sys.exit(int(os.environ.get('SCAN_TEST_EXIT', '0')))
 batch = next((arg.split('=', 1)[1] for arg in args if arg.startswith('--batch=')), None)
 single = args[args.index('--output-file') + 1] if '--output-file' in args else None
 count = int(os.environ.get('SCAN_TEST_PAGES', '4')) if batch else 1
@@ -46,12 +51,25 @@ const setup = async (t: TestContext) => {
         workDir: path.join(dir, "scans with 100% space"), scannerName: "fujitsu:ScanSnap iX1600:test",
         scanFormat: "pdf", scanResolution: 300, scanMode: "Color", scanSource: "ADF Duplex",
         scanTimeoutSeconds: 10, scanPostprocess: false, postprocessProfile: "document",
+        scanBackgroundCleanup: false, scanOcr: false, scanOcrLanguage: "deu+eng",
         postprocessPython: "python3", postprocessStrict: false,
     }
     const job: ScanJob = { id: "test-job", tenantId: 1, agentId: "test-agent", status: "running", settings: { paperSize: "A4" } }
     return { config, job, dir, args: async () => JSON.parse(await readFile(process.env.SCAN_TEST_CALL!, "utf8")) as string[] }
 }
 
+test("scanner discovery waits for network probes beyond the former ten-second limit", async (t) => {
+    await setup(t)
+    process.env.SCAN_TEST_DISCOVERY_DELAY = "10.2"
+    assert.deepEqual(await listScanners(), ["fujitsu:ScanSnap iX1600:test", "airscan:test"])
+})
+
+test("failed scanner discovery does not advertise devices from an unsuccessful command", async (t) => {
+    await setup(t)
+    process.env.SCAN_TEST_EXIT = "1"
+    assert.deepEqual(await listScanners(), [])
+})
+
 test("duplex batch saves all pages as PDF, accepts NO_DOCS and passes A4 geometry", async (t) => {
     const { config, job, args } = await setup(t)
     process.env.SCAN_TEST_EMPTY_FINAL = "1"
@@ -126,6 +144,18 @@ test("missing image runtime is detected before feeding paper", async (t) => {
     await assert.rejects(args(), /ENOENT/)
 })
 
+test("missing requested OCR language fails before the scanner consumes paper", async (t) => {
+    const { config, job, dir, args } = await setup(t)
+    const tesseract = path.join(dir, "tesseract")
+    await writeFile(tesseract, "#!/bin/sh\nprintf 'List of available languages (1):\\neng\\n'\n")
+    await chmod(tesseract, 0o755)
+    job.settings!.ocr = true
+    job.settings!.ocrLanguage = "deu+eng"
+    await assert.rejects(runScan(config, job), /OCR-Sprachpakete fehlen: deu/)
+    await assert.rejects(args(), /ENOENT/)
+    await assert.rejects(readdir(config.workDir), /ENOENT/)
+})
+
 test("single-page scans work without batch and requested paths cannot escape the job directory", async (t) => {
     const { config, job, args } = await setup(t)
     job.settings = { source: "Flatbed", format: "png" }

agents/fedeo-device-agent/tests/scan_background_test.py

@@ -0,0 +1,64 @@
+import sys
+import unittest
+from pathlib import Path
+
+import cv2
+import numpy as np
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts"))
+from scan_background import whiten_background
+
+
+class ScanBackgroundTest(unittest.TestCase):
+    def receipt(self, background=45, paper=245):
+        image = np.full((1600, 1200, 3), background, dtype=np.uint8)
+        image[120:1440, 360:820] = paper
+        for y in range(220, 1350, 70):
+            cv2.putText(image, "BON 12,34 EUR", (375, y), cv2.FONT_HERSHEY_SIMPLEX, .7, (20, 20, 20), 2)
+        cv2.rectangle(image, (400, 140), (780, 180), (40, 80, 210), -1)
+        return image
+
+    def test_receipt_on_dark_or_gray_backing_keeps_canvas_text_and_colored_logo(self):
+        for background in (35, 140):
+            image = self.receipt(background)
+            result = whiten_background(image)
+            self.assertEqual(result.shape, image.shape)
+            self.assertTrue(np.all(result[:80] == 255))
+            self.assertTrue(np.all(result[:, :250] == 255))
+            self.assertTrue(np.array_equal(result[120:1440, 360:820], image[120:1440, 360:820]))
+
+    def test_slanted_receipt_is_whitened_without_rotating_or_cropping(self):
+        image = np.full((1600, 1200, 3), 55, dtype=np.uint8)
+        polygon = np.array([[450, 100], [890, 155], [720, 1450], [280, 1390]])
+        cv2.fillPoly(image, [polygon], (245, 245, 245))
+        cv2.putText(image, "FEDEO", (470, 700), cv2.FONT_HERSHEY_SIMPLEX, 1.5, (0, 0, 0), 3)
+        result = whiten_background(image)
+        self.assertEqual(result.shape, image.shape)
+        inside = np.zeros(image.shape[:2], dtype=np.uint8)
+        cv2.fillPoly(inside, [polygon], 255)
+        self.assertTrue(np.array_equal(result[inside > 0], image[inside > 0]))
+        self.assertTrue(np.all(result[:50] == 255))
+
+    def test_white_full_page_is_not_replaced_by_its_largest_text_box(self):
+        image = np.full((1600, 1200, 3), 255, dtype=np.uint8)
+        cv2.rectangle(image, (300, 400), (800, 1200), (30, 30, 30), 4)
+        cv2.putText(image, "Footer", (20, 1550), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 2)
+        self.assertTrue(np.array_equal(whiten_background(image), image))
+
+    def test_slanted_receipt_retains_every_paper_edge_pixel_at_300_dpi(self):
+        image = np.full((3508, 2480, 3), 45, dtype=np.uint8)
+        polygon = np.array([[820, 200], [1700, 340], [1250, 3300], [380, 3160]])
+        cv2.fillPoly(image, [polygon], (245, 245, 245))
+        inside = np.zeros(image.shape[:2], dtype=np.uint8)
+        cv2.fillPoly(inside, [polygon], 255)
+        result = whiten_background(image)
+        self.assertTrue(np.array_equal(result[inside > 0], image[inside > 0]))
+        self.assertTrue(np.all(result[:50] == 255))
+
+    def test_blank_and_ambiguous_low_contrast_pages_remain_unchanged(self):
+        for image in [np.full((300, 200, 3), 245, dtype=np.uint8), self.receipt(215, 230)]:
+            self.assertTrue(np.array_equal(whiten_background(image), image))
+
+
+if __name__ == "__main__":
+    unittest.main()

agents/fedeo-device-agent/tests/scan_ocr_test.py

@@ -0,0 +1,71 @@
+import shutil
+import subprocess
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+from unittest.mock import patch
+
+from PIL import Image, ImageDraw, ImageFont, PdfParser
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts"))
+from scan_document import create_document, check_ocr
+
+
+class ScanOcrTest(unittest.TestCase):
+    def test_missing_language_is_reported_before_processing(self):
+        with patch("scan_document.run_ocr_command", return_value="List of available languages (1):\neng\n"):
+            with self.assertRaisesRegex(RuntimeError, "deu"):
+                check_ocr("deu+eng")
+        with self.assertRaises(ValueError):
+            check_ocr("--bad-language")
+
+    @unittest.skipUnless(shutil.which("tesseract") and shutil.which("pdftotext"), "Tesseract and Poppler required")
+    def test_ocr_keeps_receipt_canvas_blank_back_and_page_order_with_real_extractable_text(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            pages = []
+            for index, text in enumerate(["FEDEO Bon 12345", "", "Zweite Seite 24680"]):
+                page = root / f"page-{index}.png"
+                image = Image.new("RGB", (1240, 1754), (45, 45, 45) if index == 0 else "white")
+                draw = ImageDraw.Draw(image)
+                if index == 0:
+                    draw.rectangle((260, 150, 1000, 1550), fill="white")
+                if text:
+                    draw.text((300, 300), text, fill="black", font=ImageFont.load_default(size=44))
+                    draw.text((300, 390), "Summe EUR 12,34", fill="black", font=ImageFont.load_default(size=36))
+                image.save(page)
+                pages.append(page)
+            output = root / "searchable.pdf"
+            result = create_document(pages, output, 150, ocr=True, background_cleanup=True)
+            self.assertEqual(result["pageCount"], 3)
+            with PdfParser.PdfParser(filename=str(output)) as pdf:
+                self.assertEqual(len(pdf.pages), 3)
+                for reference in pdf.pages:
+                    box = pdf.read_indirect(reference)[b"MediaBox"]
+                    self.assertAlmostEqual(box[2], 1240 * 72 / 150, delta=.1)
+                    self.assertAlmostEqual(box[3], 1754 * 72 / 150, delta=.1)
+            text = subprocess.check_output(["pdftotext", str(output), "-"], text=True)
+            extracted = text.split("\f")
+            self.assertIn("12345", extracted[0])
+            self.assertEqual(extracted[1].strip(), "")
+            self.assertIn("24680", extracted[2])
+            self.assertTrue(all(page.exists() for page in pages))
+
+    def test_ocr_failure_never_publishes_a_partial_document_or_removes_raw_pages(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            page = root / "raw.png"
+            Image.new("RGB", (400, 600), "white").save(page)
+            output = root / "failed.pdf"
+            with patch("scan_document.check_ocr"), patch("scan_document.run_ocr_command", side_effect=RuntimeError("OCR failed")):
+                with self.assertRaisesRegex(RuntimeError, "OCR failed"):
+                    create_document([page], output, 300, ocr=True)
+            self.assertFalse(output.exists())
+            self.assertFalse(output.with_name("failed.pdf.partial").exists())
+            self.assertTrue(page.exists())
+            self.assertEqual(list(root.glob("prepared-*")), [])
+
+
+if __name__ == "__main__":
+    unittest.main()

backend/src/modules/selfhost-updater.ts

@@ -13,7 +13,8 @@ const errors = {
     COMPOSE_REVIEW_REQUIRED: "Die Compose-Datei muss zuerst auf dem Host geprüft und übernommen werden.",
     RECOVERY_REQUIRED: "Ein vorheriges Update benötigt Wiederherstellung auf dem Host.",
     PLAN_CHANGED: "Release oder Konfiguration haben sich geändert. Bitte erneut prüfen.",
-    UPDATE_FAILED: "Auftrag fehlgeschlagen. Geschützte Protokolle auf dem Host prüfen.",
+    UPDATE_FAILED: "Auftrag fehlgeschlagen.",
+    LOGS_UNAVAILABLE: "Das Update-Protokoll kann gerade nicht gelesen werden.",
     INTERRUPTED: "Der Auftrag wurde unterbrochen. Status auf dem Host prüfen.",
     UPDATER_UNAVAILABLE: "Der Host-Update-Dienst ist nicht erreichbar. Dienst und Socket-Einbindung prüfen.",
     UPDATER_PROTOCOL: "Der Host-Update-Dienst hat eine ungültige Antwort geliefert.",
@@ -46,12 +47,17 @@ const job = state.extend({ plan: plan.optional(), progress: state.optional(),
 });
 const status = z.object({ apiVersion: z.literal(1), acceptingRequests: z.boolean(), engine: state,
     installed: state.pick({ version: true, revision: true }), request: job.optional(),
+    capabilities: z.object({ requestLogs: z.boolean() }).optional(),
+});
+const logs = z.object({ requestId, logs: z.array(z.object({
+    source: z.enum(["service", "events", "commands"]), content: z.string().max(6000), truncated: z.boolean(),
+})).max(3),
 });
 
 // No caller-controlled host, URL, socket, headers or command can reach the privileged service.
 export function createUpdaterClient(socketPath = "/run/fedeo-updater/control.sock", timeoutMs = 5000) {
     return (method: "GET" | "POST", path: string, body?: unknown): Promise<{ status: number; body: unknown }> => {
-        if (!((method === "GET" && (path === "/v1/status" || /^\/v1\/requests\/[A-Za-z0-9][A-Za-z0-9_-]{7,79}$/.test(path)))
+        if (!((method === "GET" && (path === "/v1/status" || /^\/v1\/requests\/[A-Za-z0-9][A-Za-z0-9_-]{7,79}(?:\/logs)?$/.test(path)))
             || (method === "POST" && ["/v1/check", "/v1/apply"].includes(path)))) {
             return Promise.reject(new UpdaterError("INVALID_REQUEST", 400));
         }
@@ -76,7 +82,12 @@ export function createUpdaterClient(socketPath = "/run/fedeo-updater/control.soc
                             throw knownError(code) && [400, 404, 409, 503].includes(res.statusCode!)
                                 ? new UpdaterError(code, res.statusCode) : new UpdaterError("UPDATER_PROTOCOL", 502);
                         }
-                        resolve({ status: res.statusCode!, body: (path === "/v1/status" ? status : job).parse(data) });
+                        const schema = path === "/v1/status" ? status : path.endsWith("/logs") ? logs : job;
+                        const parsed = schema.parse(data);
+                        if (path.endsWith("/logs") && data.requestId !== path.split("/")[3]) {
+                            throw new UpdaterError("UPDATER_PROTOCOL", 502);
+                        }
+                        resolve({ status: res.statusCode!, body: parsed });
                     } catch (error) {
                         reject(error instanceof UpdaterError ? error : new UpdaterError("UPDATER_PROTOCOL", 502));
                     }

backend/src/routes/instanceAgents.ts

@@ -152,6 +152,7 @@ export default async function instanceAgentRoutes(server: FastifyInstance) {
                 active: instanceAgents.active,
                 preferredScannerName: instanceAgents.preferredScannerName,
                 scanDefaults: instanceAgents.scanDefaults,
+                capabilities: instanceAgents.capabilities,
             })
             .from(instanceAgents)
             .where(eq(instanceAgents.id, body.agentId))
@@ -161,6 +162,14 @@ export default async function instanceAgentRoutes(server: FastifyInstance) {
             return reply.code(404).send({ error: "Active agent not found" })
         }
 
+        const settings = { ...((agent.scanDefaults || {}) as Record<string, any>), ...(body.settings || {}) }
+        const enabled = (value: unknown) => value === true || (typeof value === "string" && ["1", "true", "yes", "ja", "on"].includes(value.toLowerCase().trim()))
+        const capabilities = agent.capabilities as Record<string, any>
+        if (((settings.format || "pdf") === "pdf" && enabled(settings.ocr) && !capabilities?.ocr) ||
+            (enabled(settings.backgroundCleanup) && !capabilities?.backgroundCleanup)) {
+            return reply.code(409).send({ error: "Bitte den Scanner-Agenten mit Tesseract und OpenCV aktualisieren. Die angeforderte Scan-Verarbeitung ist dort noch nicht verfügbar." })
+        }
+
         const [job] = await server.db
             .insert(instanceAgentScanJobs)
             .values({
@@ -169,10 +178,7 @@ export default async function instanceAgentRoutes(server: FastifyInstance) {
                 requestedBy: req.user?.user_id,
                 scannerName: body.scannerName || agent.preferredScannerName,
                 requestedFilename: body.requestedFilename,
-                settings: {
-                    ...((agent.scanDefaults || {}) as Record<string, any>),
-                    ...(body.settings || {}),
-                },
+                settings,
                 target: body.target || {},
             })
             .returning()

backend/src/routes/selfhostUpdates.ts

@@ -24,6 +24,10 @@ export default async function selfhostUpdateRoutes(server: FastifyInstance, opti
         if (!requestId.safeParse(req.params.id).success) return reply.code(400).send({ error: { code: "INVALID_REQUEST" } });
         return forward(reply, "GET", `/v1/requests/${req.params.id}`);
     });
+    server.get<{ Params: { id: string } }>("/admin/updates/requests/:id/logs", async (req, reply) => {
+        if (!requestId.safeParse(req.params.id).success) return reply.code(400).send({ error: { code: "INVALID_REQUEST" } });
+        return forward(reply, "GET", `/v1/requests/${req.params.id}/logs`);
+    });
     for (const command of ["check", "apply"] as const) {
         server.post(`/admin/updates/${command}`, { bodyLimit: 4096 }, async (req, reply) => {
             // The auth plugin verifies this JWT. Cookies alone cannot trigger host operations (CSRF).

backend/tests/instanceAgentSetup.test.ts

@@ -55,3 +55,30 @@ test("unknown agents cannot receive setup credentials", async (t) => {
     assert.equal(response.statusCode, 404)
     assert.equal(f.updates.length, 0)
 })
+
+test("OCR and background cleanup require an agent that supports the requested processing", async t => {
+    const app = Fastify()
+    t.after(() => app.close())
+    const jobs: any[] = []
+    let capabilities: Record<string, boolean> = { scan: true }
+    app.addHook("onRequest", async request => { request.user = { is_admin: true, tenant_id: 1 } as any })
+    app.decorate("db", {
+        select: () => ({ from: () => ({ where: () => ({ limit: async () => [{ id, active: true, scanDefaults: {}, capabilities }] }) }) }),
+        insert: () => ({ values: (data: any) => ({ returning: async () => { jobs.push(data); return [{ id: "job", ...data }] } }) }),
+    })
+    await app.register(routes)
+    for (const settings of [{ ocr: true }, { ocr: "true" }, { backgroundCleanup: true }]) {
+        const response = await app.inject({ method: "POST", url: "/scan-jobs", payload: { agentId: id, settings } })
+        assert.equal(response.statusCode, 409)
+        assert.equal(jobs.length, 0)
+    }
+    capabilities = { scan: true, ocr: true, backgroundCleanup: true }
+    const response = await app.inject({
+        method: "POST", url: "/scan-jobs",
+        payload: { agentId: id, settings: { ocr: true, ocrLanguage: "deu+eng", backgroundCleanup: true }, target: { folder: null, type: null } },
+    })
+    assert.equal(response.statusCode, 200)
+    assert.equal(jobs.length, 1)
+    assert.equal(jobs[0].settings.ocr, true)
+    assert.equal(jobs[0].target.type, null)
+})

backend/tests/selfhostUpdates.test.ts

@@ -30,11 +30,22 @@ test("all update endpoints require a current global admin, including after revoc
     assert.equal((await f.app.inject({ url: "/admin/updates/status" })).statusCode, 202);
     f.revoke();
     for (const [method, url] of [["GET", "/admin/updates/status"], ["GET", `/admin/updates/requests/${id}`],
+        ["GET", `/admin/updates/requests/${id}/logs`],
         ["POST", "/admin/updates/check"], ["POST", "/admin/updates/apply"]] as const) {
         assert.equal((await f.app.inject({ method, url, headers, payload: method === "POST" ? { requestId: id } : undefined })).statusCode, 403);
     }
     assert.equal(f.calls.length, 1);
 });
+test("admin log access uses the fixed request route and does not cache responses", async t => {
+    const f = await fixture(t);
+    const response = await f.app.inject({ url: `/admin/updates/requests/${id}/logs` });
+    assert.equal(response.statusCode, 202);
+    assert.equal(response.headers["cache-control"], "no-store");
+    assert.deepEqual(f.calls[0], ["GET", `/v1/requests/${id}/logs`, undefined]);
+    assert.equal((await f.app.inject({ url: "/admin/updates/requests/bad.id/logs" })).statusCode, 400);
+    const anonymous = await fixture(t, true, false);
+    assert.equal((await anonymous.app.inject({ url: `/admin/updates/requests/${id}/logs` })).statusCode, 401);
+});
 test("unauthenticated and cookie-only requests cannot start jobs", async t => {
     const anonymous = await fixture(t, true, false);
     assert.equal((await anonymous.app.inject({ url: "/admin/updates/status" })).statusCode, 401);
@@ -117,3 +128,19 @@ test("Unix client bounds response size, deadline, and protocol shape", async t =
         });
     }
 });
+test("Unix client validates bounded log responses and their request association", async t => {
+    let payload: any = { requestId: id, logs: [{ source: "commands", content: "pull access denied", truncated: true,
+        path: "/private/commands.log" }], private: "omitted" };
+    const client = await socketFixture(t, (_req, res) => res.end(JSON.stringify(payload)));
+    const response = await client("GET", `/v1/requests/${id}/logs`);
+    assert.deepEqual(response.body, { requestId: id, logs: [{ source: "commands", content: "pull access denied", truncated: true }] });
+    for (const invalid of [
+        { ...payload, requestId: "different-request" },
+        { ...payload, logs: [{ source: "env", content: "secret", truncated: false }] },
+        { ...payload, logs: [{ source: "commands", content: "x".repeat(6001), truncated: false }] },
+    ]) {
+        payload = invalid;
+        await assert.rejects(client("GET", `/v1/requests/${id}/logs`), { code: "UPDATER_PROTOCOL" });
+    }
+    await assert.rejects(client("GET", `/v1/requests/${id}/logs?path=/etc/passwd`), { code: "INVALID_REQUEST" });
+});

docs/selfhost-updates.md

@@ -101,7 +101,15 @@ Der aktuelle Updater erlaubt dabei gezielt das zusätzliche, schreibgeschützte
 
 Danach laufen weitere Anwendungs-Releases einschließlich Compose und Migrationen über den Button. Änderungen am Host-Dienst selbst benötigen weiterhin die gemeinsame Aktualisierung der Host-Dateien und den Dienstinstaller.
 
-Die Backend-Routen `/api/admin/updates/status`, `/api/admin/updates/requests/:id`, `/api/admin/updates/check` und `/api/admin/updates/apply` prüfen bei jeder Anfrage die aktuellen globalen Administratorrechte aus der Datenbank. Mandantenadministratoren haben keinen Zugriff. Startanfragen benötigen zusätzlich den bereits von der Authentifizierung geprüften Bearer-JWT; Cookie-Anmeldung allein kann keine Hostaktion auslösen. Das Backend verwendet ausschließlich den festen Unix-Socket, begrenzt Antwortgröße und Wartezeit und liefert nur erlaubte Status-/Planfelder aus. CLI-Wiederherstellung, Übernahme fremder Compose-Dateien, Shell-Befehle und frei wählbare Download-Adressen bleiben auf dem Host.
+Die Backend-Routen `/api/admin/updates/status`, `/api/admin/updates/requests/:id`, `/api/admin/updates/requests/:id/logs`, `/api/admin/updates/check` und `/api/admin/updates/apply` prüfen bei jeder Anfrage die aktuellen globalen Administratorrechte aus der Datenbank. Mandantenadministratoren haben keinen Zugriff. Startanfragen benötigen zusätzlich den bereits von der Authentifizierung geprüften Bearer-JWT; Cookie-Anmeldung allein kann keine Hostaktion auslösen. Das Backend verwendet ausschließlich den festen Unix-Socket, begrenzt Antwortgröße und Wartezeit und liefert nur erlaubte Status-/Planfelder und bereinigte Protokollauszüge aus. CLI-Wiederherstellung, Übernahme fremder Compose-Dateien, Shell-Befehle und frei wählbare Download-Adressen bleiben auf dem Host.
+
+### Update-Protokoll in der Oberfläche
+
+Unter **FEDEO-Updates → Update-Protokoll anzeigen** sehen Systemadministratoren die Fehlerdetails des Host-Dienstes, den Update-Verlauf und die Werkzeugausgaben (beispielsweise Docker-Pull- oder Migrationsfehler). Bei fehlgeschlagenen oder unterbrochenen Aufträgen öffnet sich die Ansicht automatisch. Solange sie geöffnet ist, wird sie zusammen mit dem Auftragsstatus aktualisiert; „Protokoll aktualisieren“ fragt sie zusätzlich sofort ab. Während der Wartungsunterbrechung bleibt die letzte geladene Ausgabe sichtbar.
+
+Pro Quelle werden höchstens die letzten 6000 UTF-8-Bytes angezeigt; gekürzte Ausgaben sind markiert. Die Bereinigung erfolgt auf dem Host vor der Übertragung: erkannte Kennwörter, Tokens, Authentifizierungsheader und URL-Zugangsdaten sowie bekannte geheime Werte aus der Umgebung und den aktuellen/gesicherten Konfigurationsdateien werden ersetzt. Die Originalprotokolle bleiben unverändert geschützt auf dem Host. Nicht als Zugangsdaten erkennbare Anwendungsdaten können weiterhin enthalten sein.
+
+Bestehende Installationen benötigen für diese Ansicht **auch einen aktualisierten Host-Dienst**, zusätzlich zu Backend und Frontend. Die vier zusammengehörigen Host-Dateien aus einem geprüften Stand bereitstellen und `selfhost-update-service-install.sh` erneut ausführen, wie unter „Host-Dienst installieren“ beschrieben. Ein Anwendungsupdate allein ersetzt den Host-Dienst nicht. Die Oberfläche erkennt ältere Dienste und zeigt dafür einen konkreten Hinweis. Bereits vorhandene Aufträge bleiben lesbar; ältere Protokolle werden anhand der gespeicherten Auftrags-ID zugeordnet. Nach Aktualisierung des Host-Dienstes lässt sich auch vor einem Anwendungsupdate direkt `GET /v1/requests/<requestId>/logs` über den Unix-Socket verwenden.
 
 ## Lokale Dienst-API
 
@@ -112,6 +120,7 @@ Alle Antworten sind JSON mit `Cache-Control: no-store`. Mutationen benötigen `C
 | `GET /v1/status` | Installierter Stand, Zustand der Update-Engine und aktueller Dienstauftrag |
 | `POST /v1/check` | Asynchrone Prüfung; `requestId`, optional `version` ohne `v` |
 | `GET /v1/requests/<requestId>` | Auftrag, bereinigter Prüfplan und aktueller Fortschritt |
+| `GET /v1/requests/<requestId>/logs` | Bereinigte, begrenzte Auszüge aus Dienst-, Ablauf- und Werkzeugprotokoll; nur für diesen Auftrag |
 | `POST /v1/apply` | Asynchrones Update; `requestId`, feste `version` und erfolgreiche `checkRequestId` |
 
 Beispiel für eine Prüfung des neuesten Releases:
@@ -151,7 +160,7 @@ Aufträge liegen dauerhaft und root-lesbar unter `.updates/service/`. Nach einem
 1. Exklusive Host-Sperre erwerben; offene Wiederherstellungsfälle blockieren neue Updates.
 2. Release samt Prüfsummen laden, Manifest/Commit prüfen und die FEDEO-Images auf die exakten Digests setzen.
 3. Aktuelle und neue Compose-Konfiguration mit Docker Compose validieren. Projektname, Container-Identitäten, Mounts, Netzwerke und Zugangsdaten erhalten. Die Shell-Umgebung überschreibt dabei keine Werte aus der Instanz-`.env`.
-4. Speicherplatz prüfen: mindestens 5 GiB frei am Docker-Datenverzeichnis vor dem Pull sowie geschätzte Backup-Größe plus Reserve im Betriebsverzeichnis. Images laden, Backup-Platz erneut prüfen.
+4. Speicherplatz prüfen: mindestens 5 GiB frei am Docker-Datenverzeichnis vor dem Pull sowie geschätzte Backup-Größe plus Reserve im Betriebsverzeichnis. Images mit exakt gepinntem SHA-256-Digest nur laden, wenn sie lokal fehlen (`pull --policy missing`); vorhandene Tags ohne passenden Digest reichen dafür nicht aus. Bewegliche Infrastruktur-Tags weiterhin aktualisieren (`pull --policy always`). Ein fehlendes, nicht herunterladbares Image beendet den Auftrag vor der Wartung. Backup-Platz erneut prüfen. Während der anschließenden Infrastrukturstarts und Migrationen erfolgt kein weiterer Pull (`up`/`run --pull never`), auch nicht bei einer abweichenden Pull-Policy in einer Override-Datei.
 5. Schreibende Dienste und Proxy anhalten. PostgreSQL bleibt für `pg_dump` aktiv.
 6. Konfiguration, Container-Metadaten, PostgreSQL-Dump im Custom-Format, globale PostgreSQL-Rollen, gestoppten MinIO-Dateispeicher und Zertifikate sichern. Dump und TAR auf Lesbarkeit prüfen und SHA-256-Prüfsummen schreiben. `backup/complete.json` markiert ein vollständig geschriebenes Backup; ein Restore-Probelauf auf einer getrennten Instanz wird dadurch nicht ersetzt.
    Beim Übergang vom alten Matrix-Stack enthält das TAR zusätzlich `matrix/`. Container-Metadaten enthalten die ursprünglichen Image-IDs, Mounts und Neustartrichtlinien; `complete.json` nennt die stillgelegten Dienste und erhaltenen Container. Vor dem Dateiaustausch wird deren automatischer Neustart deaktiviert. Das Zielrelease enthält keine Matrix-Dienste und keine Backend-Abhängigkeit auf Synapse mehr.
@@ -163,6 +172,8 @@ Status und Phasen werden atomar in `.updates/status.json` und pro Auftrag in `st
 
 ## Fehler und Wiederherstellung
 
+Bei einer nicht erreichbaren Registry können bereits lokal vorhandene, exakt passende Digest-Images weiterverwendet werden. Das hilft beispielsweise bei einem `401 Unauthorized` der MinIO-Registry, wenn sowohl MinIO als auch `mc` mit den Ziel-Digests bereits auf diesem Host liegen. Der Host-Updater muss diese Pull-Logik bereits enthalten; bestehende Dienstinstallationen nach Austausch des Hostskripts mit dem Dienstinstaller aktualisieren. Veröffentlichte Release-Assets werden nicht nachträglich verändert. Für Instanzen ohne diese Images bleibt eine geprüfte Bezugsquelle notwendig. Der lokale Cache ersetzt weder Sicherheitsupdates noch eine dauerhafte Versorgung mit gepflegten Images.
+
 - **Download, Konfiguration oder Pull fehlgeschlagen:** Der laufende Stack bleibt bestehen.
 - **Backup oder Stop fehlgeschlagen:** Der Updater versucht, die zuvor laufenden Container wieder zu starten. Gelingt das nicht, bleibt der Wiederherstellungsstatus gesetzt.
 - **Ab Austausch der Dateien/Infrastruktur:** Keine automatische Rückkehr zu älteren Images. Bei einem Fehler werden Backend, Frontend und Proxy nach Möglichkeit angehalten. Das Datenbankschema oder das Speicherformat könnten bereits verändert sein.

frontend/components/AgentSetupWizard.vue

@@ -38,7 +38,14 @@ const installCommand = computed(() => 'bash ' + filename.value)
 const tokenReady = computed(() => /^fedeo_agent_[a-f0-9]{64}$/.test(token.value.trim()))
 const online = computed(() => agentSetupReady(agent.value, checkingSince.value, now.value))
 const scannerOptions = computed(() => (agent.value?.scannerNames || []).map(value => ({ label: value, value })))
-const readyToSave = computed(() => online.value && agent.value?.capabilities?.scan && scannerOptions.value.some(item => item.value === selectedScanner.value))
+const setupScanDefaults = computed(() => profile.value === 'ix1600' ? { ...ix1600ScanDefaults } : {
+  format: 'pdf', resolution: 300, mode: 'Color', paperSize: 'auto', postprocess: false,
+  backgroundCleanup: true, ocr: true, ocrLanguage: 'deu+eng', ...agent.value?.scanDefaults,
+})
+const processingMissing = computed(() =>
+  (setupScanDefaults.value.format === 'pdf' && setupScanDefaults.value.ocr && !agent.value?.capabilities?.ocr) ||
+  (setupScanDefaults.value.backgroundCleanup && !agent.value?.capabilities?.backgroundCleanup))
+const readyToSave = computed(() => online.value && agent.value?.capabilities?.scan && !processingMissing.value && scannerOptions.value.some(item => item.value === selectedScanner.value))
 const endpointPreview = computed(() => {
   try { return agentBackendUrl(backendUrl.value) + '/instance-agent/heartbeat' } catch { return '' }
 })
@@ -129,10 +136,7 @@ const finishSetup = async () => {
   if (!agent.value || !readyToSave.value || busy.value) return
   busy.value = true
   try {
-    const scanDefaults = profile.value === 'ix1600' ? { ...ix1600ScanDefaults } : {
-      format: 'pdf', resolution: 300, mode: 'Color', paperSize: 'auto', postprocess: false, ...agent.value.scanDefaults,
-    }
-    await admin.updateInstanceAgent(agent.value.id, { preferredScannerName: selectedScanner.value, scanDefaults })
+    await admin.updateInstanceAgent(agent.value.id, { preferredScannerName: selectedScanner.value, scanDefaults: setupScanDefaults.value })
     saved.value = true
     token.value = ''
     stopPolling()
@@ -212,7 +216,7 @@ onBeforeUnmount(() => {
       </div>
       <div class="space-y-4 rounded-lg bg-muted/40 p-5">
         <h3 class="font-semibold text-highlighted">Installation herunterladen</h3>
-        <p class="text-sm text-muted">Die Datei enthält die Agent-Software und deine Konfiguration. Sie prüft Node.js ab Version 20, SANE und Python/Pillow und installiert fehlende Systempakete mit sudo.</p>
+        <p class="text-sm text-muted">Die Datei enthält die Agent-Software und deine Konfiguration. Sie installiert bei Bedarf Node.js ab Version 20, SANE, Bildverarbeitung und OCR für Deutsch/Englisch mit sudo.</p>
         <p class="text-sm text-muted">Die Installation erfolgt in deinem Benutzerverzeichnis. Starte die Datei auf dem Rechner mit dem Scanner.</p>
         <UAlert v-if="configurationError" color="error" variant="soft" :title="configurationError" />
         <UButton icon="i-heroicons-arrow-down-tray" :disabled="!tokenReady || busy" @click="downloadInstaller">Linux-Installation herunterladen</UButton>
@@ -253,6 +257,8 @@ onBeforeUnmount(() => {
       <UAlert v-if="!online" color="warning" variant="soft" title="Agent ist derzeit nicht verbunden" description="Agent starten und die Verbindung erneut prüfen." />
       <UAlert v-else-if="!agent?.capabilities?.scan || !scannerOptions.length" color="warning" variant="soft" title="Noch kein Scanner verfügbar"
               description="USB-Verbindung und Stromversorgung prüfen. scanimage -L muss den Scanner für denselben Linux-Benutzer anzeigen." />
+      <UAlert v-else-if="processingMissing" color="warning" variant="soft" title="Scan-Verarbeitung noch nicht bereit"
+              description="Für dieses Profil fehlen OCR oder Hintergrundbereinigung. Die aktuelle Linux-Installation aus Schritt 2 ausführen und anschließend den Agenten neu starten." />
       <UAlert v-if="connectionError" color="error" variant="soft" title="Status konnte nicht geprüft werden" :description="connectionError" />
       <UFormField label="Scanner" required>
         <USelectMenu v-model="selectedScanner" :items="scannerOptions" value-key="value" label-key="label" class="w-full" placeholder="Erkannten Scanner auswählen" />
@@ -263,7 +269,7 @@ onBeforeUnmount(() => {
           { label: 'Bisherige Einstellungen / allgemeiner Scanner', value: 'existing' }
         ]" value-key="value" label-key="label" class="w-full" />
       </UFormField>
-      <p class="text-sm text-muted">Das iX1600-Profil speichert den gesamten Stapel mit Vorder- und Rückseiten in einem PDF. Weitere Einstellungen kannst du anschließend in der Scannerverwaltung ändern.</p>
+      <p class="text-sm text-muted">Das iX1600-Profil speichert Vorder- und Rückseiten als durchsuchbares PDF. Erkannter Scannerhintergrund außerhalb des Papiers wird weiß; das A4-Format bleibt erhalten.</p>
       <div class="flex flex-wrap gap-3">
         <UButton color="neutral" variant="outline" :loading="checking" @click="checkConnection">Scanner erneut suchen</UButton>
         <UButton :disabled="!readyToSave" :loading="busy" @click="finishSetup">Einrichtung abschließen</UButton>

frontend/components/FileScanModal.vue

@@ -34,6 +34,9 @@ const scanForm = reactive({
   source: "ADF Duplex",
   paperSize: "A4",
   postprocess: false,
+  backgroundCleanup: true,
+  ocr: true,
+  ocrLanguage: "deu+eng",
   postprocessProfile: "document"
 })
 
@@ -49,7 +52,8 @@ watch(() => scanForm.format, (format) => {
 const applyIx1600Preset = () => {
   Object.assign(scanForm, {
     format: "pdf", resolution: 300, mode: "Color", source: "ADF Duplex",
-    paperSize: "A4", postprocess: false, postprocessProfile: "document"
+    paperSize: "A4", postprocess: false, postprocessProfile: "document",
+    backgroundCleanup: true, ocr: true, ocrLanguage: "deu+eng"
   })
 }
 
@@ -83,6 +87,9 @@ const applyAgentDefaults = (agent) => {
   scanForm.source = agent.scanDefaults?.source || "ADF Duplex"
   scanForm.paperSize = agent.scanDefaults?.paperSize || "A4"
   scanForm.postprocess = Boolean(agent.scanDefaults?.postprocess)
+  scanForm.backgroundCleanup = agent.scanDefaults?.backgroundCleanup !== false
+  scanForm.ocr = agent.scanDefaults?.ocr !== false
+  scanForm.ocrLanguage = agent.scanDefaults?.ocrLanguage || "deu+eng"
   scanForm.postprocessProfile = agent.scanDefaults?.postprocessProfile || "document"
   if (isFeederScan.value) scanForm.format = "pdf"
 
@@ -147,6 +154,11 @@ const waitForJob = async (jobId) => {
 
 const startScan = async () => {
   if (!selectedAgentId.value || scanInProgress.value) return
+  if ((scanForm.format === "pdf" && scanForm.ocr && !selectedAgent.value?.capabilities?.ocr) ||
+      (scanForm.backgroundCleanup && !selectedAgent.value?.capabilities?.backgroundCleanup)) {
+    toast.add({ title: "Agent aktualisieren", description: "Für OCR und Hintergrundbereinigung benötigt der Scanner-Rechner die aktuelle Agent-Installation mit Tesseract und OpenCV.", color: "error" })
+    return
+  }
 
   scanInProgress.value = true
   statusMessage.value = "Scan-Auftrag wird erstellt..."
@@ -166,6 +178,9 @@ const startScan = async () => {
           source: scanForm.source || null,
           paperSize: scanForm.paperSize,
           postprocess: scanForm.postprocess,
+          backgroundCleanup: scanForm.backgroundCleanup,
+          ocr: scanForm.format === "pdf" && scanForm.ocr,
+          ocrLanguage: scanForm.ocrLanguage,
           postprocessProfile: scanForm.postprocessProfile
         },
         target: {
@@ -319,10 +334,23 @@ loadAgents()
             </UFormField>
           </div>
 
+          <div class="space-y-3">
+            <UCheckbox v-model="scanForm.backgroundCleanup" label="Scannerhintergrund weiß machen" :disabled="scanInProgress" />
+            <p class="text-xs text-muted">Erkannte Flächen außerhalb von Bon oder Dokument werden weiß. Das Seitenformat bleibt erhalten; bei unsicheren Rändern bleibt der Scan unverändert.</p>
+            <UCheckbox v-model="scanForm.ocr" label="Durchsuchbares PDF (OCR)" :disabled="scanForm.format !== 'pdf' || scanInProgress" />
+            <UFormField v-if="scanForm.ocr && scanForm.format === 'pdf'" label="OCR-Sprache">
+              <USelectMenu v-model="scanForm.ocrLanguage" :items="[
+                { label: 'Deutsch und Englisch', value: 'deu+eng' },
+                { label: 'Deutsch', value: 'deu' }, { label: 'Englisch', value: 'eng' }
+              ]" value-key="value" label-key="label" :disabled="scanInProgress" />
+            </UFormField>
+            <p class="text-xs text-muted">Texterkennung erfolgt vor dem Upload auf dem Scanner-Rechner, auch für normale Dateien ohne Eingangsbeleg.</p>
+          </div>
+
           <div class="grid gap-3 sm:grid-cols-[auto_minmax(0,1fr)]">
             <UCheckbox
                 v-model="scanForm.postprocess"
-                label="OpenCV-Korrektur"
+                label="Zuschnitt und Ausrichtung zusätzlich korrigieren"
                 :disabled="scanInProgress"
             />
             <UFormField label="Profil">

frontend/components/SelfhostUpdateCard.vue

@@ -1,6 +1,6 @@
 <script setup lang="ts">
 const { overview, job, pending, connected, submitting, refreshing, message, accessDenied, recovery, busy, available, canApply,
-  check, apply, refresh, sendPending } = useSelfhostUpdates()
+  logsOpen, logsLoading, logsMessage, logEntries, refreshLogs, toggleLogs, check, apply, refresh, sendPending } = useSelfhostUpdates()
 const build = useRuntimeConfig().public
 const confirmOpen = ref(false)
 const confirmed = ref(false)
@@ -16,6 +16,7 @@ const statusLabels: Record<string, string> = {
   succeeded: 'Update erfolgreich abgeschlossen', failed: 'Auftrag fehlgeschlagen', interrupted: 'Auftrag unterbrochen',
   idle: 'Bereit', recovered: 'Wiederherstellung bestätigt', unknown: 'Auftragsstatus wird geladen',
 }
+const logLabels = { service: 'Fehlerdetails', events: 'Update-Verlauf', commands: 'Werkzeugausgaben' }
 const progress = computed(() => job.value?.progress || (busy.value ? overview.value?.engine : null))
 const statusLabel = computed(() => phaseLabels[progress.value?.phase || ''] || statusLabels[job.value?.status || 'idle'])
 const backupSize = computed(() => {
@@ -59,6 +60,24 @@ const reload = () => window.location.reload()
       </dl>
       <p v-if="job?.plan?.retiredServices.length" class="text-sm text-muted">Alte Matrix-Dienste werden stillgelegt; Daten und Container bleiben erhalten: {{ job.plan.retiredServices.join(', ') }}.</p>
       <p v-if="job?.requestId" class="break-all text-xs text-muted">Auftrag: {{ job.requestId }}</p>
+      <section v-if="job?.requestId && !accessDenied" aria-label="Update-Protokoll" class="space-y-3 rounded-lg border border-default p-3">
+        <div class="flex flex-wrap items-center justify-between gap-2">
+          <UButton color="neutral" variant="ghost" icon="i-heroicons-document-text" :aria-expanded="logsOpen" @click="toggleLogs">
+            {{ logsOpen ? 'Update-Protokoll ausblenden' : 'Update-Protokoll anzeigen' }}
+          </UButton>
+          <UButton v-if="logsOpen && overview?.capabilities?.requestLogs" color="neutral" variant="outline" size="xs" :loading="logsLoading" :disabled="!connected" @click="refreshLogs">Protokoll aktualisieren</UButton>
+        </div>
+        <div v-if="logsOpen" class="space-y-3">
+          <p class="text-xs text-muted">Die letzten Protokolleinträge werden automatisch aktualisiert. Erkannte Zugangsdaten sind ausgeblendet.</p>
+          <UAlert v-if="logsMessage" color="warning" variant="soft" :title="logsMessage" />
+          <p v-else-if="!logEntries.length" class="text-sm text-muted">{{ logsLoading ? 'Protokoll wird geladen …' : 'Für diesen Auftrag liegen noch keine Protokolleinträge vor.' }}</p>
+          <div v-for="entry in logEntries" :key="entry.source" class="space-y-1">
+            <h3 class="text-sm font-medium">{{ logLabels[entry.source] }}</h3>
+            <p v-if="entry.truncated" class="text-xs text-muted">Gekürzte Anzeige der letzten Einträge. Das vollständige Protokoll liegt auf dem Host.</p>
+            <pre tabindex="0" :aria-label="logLabels[entry.source]" class="max-h-80 overflow-auto whitespace-pre-wrap break-words rounded-md bg-elevated p-3 text-xs">{{ entry.content }}</pre>
+          </div>
+        </div>
+      </section>
       <div class="flex flex-wrap gap-2">
         <UButton icon="i-heroicons-magnifying-glass" :disabled="!available" :loading="submitting && pending?.command === 'check'" @click="check">Nach Updates suchen</UButton>
         <UButton v-if="canApply" icon="i-heroicons-arrow-down-tray" @click="openConfirmation">Update auf {{ job?.plan?.version }} installieren</UButton>

frontend/composables/useAdmin.ts

@@ -144,6 +144,9 @@ export type InstanceAgent = {
     resolution?: number
     mode?: string
     source?: string | null
+    backgroundCleanup?: boolean
+    ocr?: boolean
+    ocrLanguage?: string
     [key: string]: any
   }
   lastSeenAt?: string | null

frontend/composables/useSelfhostUpdates.ts

@@ -7,9 +7,11 @@ export type UpdateJob = UpdateState & {
   plan?: { version: string; revision: string; previousVersion?: string; updateAvailable: boolean;
     requiresComposeAdoption: boolean; estimatedBackupBytes?: number; retiredServices: string[] };
 }
+export type UpdateLogs = { requestId: string; logs: { source: 'service' | 'events' | 'commands'; content: string; truncated: boolean }[] }
 type Overview = {
   acceptingRequests: boolean; engine: UpdateState; request?: UpdateJob;
   installed: { version?: string }; backend: { version: string; revision: string };
+  capabilities?: { requestLogs: boolean };
 }
 type Submission = { command: 'check' | 'apply'; body: { requestId: string; version?: string; checkRequestId?: string } }
 
@@ -24,6 +26,12 @@ export const useSelfhostUpdates = () => {
   const refreshing = ref(false)
   const message = ref('')
   const accessDenied = ref(false)
+  const logs = ref<UpdateLogs | null>(null)
+  const logsOpen = ref(false)
+  const logsLoading = ref(false)
+  const logsMessage = ref('')
+  const logEntries = computed(() => logs.value?.requestId === job.value?.requestId ? logs.value?.logs || [] : [])
+  let openedFailureId: string | undefined
   let stopped = false
   let generation = 0
   let timer: ReturnType<typeof setTimeout> | undefined
@@ -55,6 +63,28 @@ export const useSelfhostUpdates = () => {
       message.value = error?.data?.error?.message || 'Der Update-Dienst ist nicht erreichbar. Dienst und Socket-Einbindung auf dem Host prüfen.'
     }
   }
+  const refreshLogs = async () => {
+    const id = job.value?.requestId
+    if (!logsOpen.value || !id || stopped || accessDenied.value || logsLoading.value) return
+    if (logs.value?.requestId !== id) { logs.value = null; logsMessage.value = '' }
+    if (!overview.value?.capabilities?.requestLogs) {
+      logsMessage.value = 'Dieser Host-Update-Dienst unterstützt die Protokollansicht noch nicht. Die zusammengehörigen Host-Update-Dateien aktualisieren und den Dienstinstaller erneut ausführen. Bis dahin liegen die Protokolle auf dem Host unter .updates/.'
+      return
+    }
+    const currentGeneration = generation
+    logsLoading.value = true
+    try {
+      const result = await api<UpdateLogs>(`requests/${encodeURIComponent(id)}/logs`)
+      if (stopped || currentGeneration !== generation || job.value?.requestId !== id) return
+      logs.value = result
+      logsMessage.value = ''
+    } catch (error: any) {
+      if (stopped || currentGeneration !== generation || job.value?.requestId !== id) return
+      logsMessage.value = error?.data?.error?.message || 'Das Update-Protokoll konnte nicht geladen werden. Die Abfrage wird automatisch wiederholt.'
+      if ([401, 403].includes(errorStatus(error))) { logs.value = null; report(error) }
+    } finally { logsLoading.value = false }
+  }
+  const toggleLogs = async () => { logsOpen.value = !logsOpen.value; await refreshLogs() }
   const refresh = async () => {
     if (refreshing.value || stopped || accessDenied.value) return
     refreshing.value = true
@@ -81,6 +111,11 @@ export const useSelfhostUpdates = () => {
       connected.value = true
       message.value = pending.value ? 'Start noch nicht bestätigt. Du kannst denselben Auftrag erneut senden.' : ''
       persist()
+      if (['failed', 'interrupted'].includes(job.value?.status || '') && openedFailureId !== job.value?.requestId) {
+        openedFailureId = job.value?.requestId
+        logsOpen.value = true
+      }
+      await refreshLogs()
     } catch (error) {
       if (currentGeneration !== generation || stopped) return
       connected.value = false
@@ -90,6 +125,8 @@ export const useSelfhostUpdates = () => {
   const sendPending = async () => {
     if (!pending.value || submitting.value || accessDenied.value) return
     generation++
+    logs.value = null
+    logsMessage.value = ''
     submitting.value = true
     const submission = pending.value
     try {
@@ -109,6 +146,7 @@ export const useSelfhostUpdates = () => {
         report(error)
       }
     } finally { submitting.value = false; persist() }
+    await refreshLogs()
   }
   const check = async () => {
     if (!available.value) return
@@ -145,5 +183,5 @@ export const useSelfhostUpdates = () => {
   })
   onBeforeUnmount(() => { stopped = true; clearTimeout(timer) })
   return { overview, job, pending, connected, submitting, refreshing, message, accessDenied, recovery, busy, available, canApply,
-    check, apply, refresh, sendPending }
+    logsOpen, logsLoading, logsMessage, logEntries, refreshLogs, toggleLogs, check, apply, refresh, sendPending }
 }

frontend/pages/administration/scanners.vue

@@ -31,6 +31,9 @@ const editState = reactive({
   source: "",
   paperSize: "A4",
   postprocess: false,
+  backgroundCleanup: true,
+  ocr: true,
+  ocrLanguage: "deu+eng",
   postprocessProfile: "document",
 })
 
@@ -69,13 +72,17 @@ const applyAgentToForm = (agent: InstanceAgent | null) => {
   editState.source = agent.scanDefaults?.source || ""
   editState.paperSize = agent.scanDefaults?.paperSize || "A4"
   editState.postprocess = Boolean(agent.scanDefaults?.postprocess)
+  editState.backgroundCleanup = agent.scanDefaults?.backgroundCleanup !== false
+  editState.ocr = agent.scanDefaults?.ocr !== false
+  editState.ocrLanguage = agent.scanDefaults?.ocrLanguage || "deu+eng"
   editState.postprocessProfile = agent.scanDefaults?.postprocessProfile || "document"
 }
 
 const applyIx1600Preset = () => {
   Object.assign(editState, {
     format: "pdf", resolution: 300, mode: "Color", source: "ADF Duplex",
-    paperSize: "A4", postprocess: false, postprocessProfile: "document"
+    paperSize: "A4", postprocess: false, postprocessProfile: "document",
+    backgroundCleanup: true, ocr: true, ocrLanguage: "deu+eng"
   })
 }
 
@@ -132,6 +139,9 @@ const saveAgent = async () => {
         source: editState.source || null,
         paperSize: editState.paperSize,
         postprocess: editState.postprocess,
+        backgroundCleanup: editState.backgroundCleanup,
+        ocr: editState.format === "pdf" && editState.ocr,
+        ocrLanguage: editState.ocrLanguage,
         postprocessProfile: editState.postprocessProfile,
       },
     })
@@ -312,8 +322,21 @@ onMounted(async () => {
                 ]" value-key="value" label-key="label" />
               </UFormField>
 
+              <div class="space-y-3">
+                <UCheckbox v-model="editState.backgroundCleanup" label="Scannerhintergrund weiß machen" />
+                <p class="text-xs text-muted">Erkannte Flächen außerhalb von Bon oder Dokument werden weiß. Das Seitenformat bleibt erhalten; bei unsicheren Rändern bleibt der Scan unverändert.</p>
+                <UCheckbox v-model="editState.ocr" label="Durchsuchbares PDF (OCR)" :disabled="editState.format !== 'pdf'" />
+                <UFormField v-if="editState.ocr && editState.format === 'pdf'" label="OCR-Sprache">
+                  <USelectMenu v-model="editState.ocrLanguage" :items="[
+                    { label: 'Deutsch und Englisch', value: 'deu+eng' },
+                    { label: 'Deutsch', value: 'deu' }, { label: 'Englisch', value: 'eng' }
+                  ]" value-key="value" label-key="label" />
+                </UFormField>
+                <p class="text-xs text-muted">Texterkennung erfolgt vor dem Upload auf dem Scanner-Rechner, auch für normale Dateien ohne Eingangsbeleg.</p>
+              </div>
+
               <div class="grid gap-3 sm:grid-cols-[auto_minmax(0,1fr)]">
-                <UCheckbox v-model="editState.postprocess" label="OpenCV-Nachbearbeitung" />
+                <UCheckbox v-model="editState.postprocess" label="Zuschnitt und Ausrichtung zusätzlich korrigieren" />
                 <UFormField label="Profil">
                   <USelectMenu
                       v-model="editState.postprocessProfile"

frontend/pages/createDocument/edit/[[id]].vue

@@ -35,6 +35,7 @@ const guard = ref(true)
 
 const itemInfo = ref({
   type: "invoices",
+  state: "Entwurf",
   taxType: "Standard",
   customer: null,
   contact: null,
@@ -332,6 +333,13 @@ const normalizeEntityId = (value) => {
   }
   return typeof value === "object" ? (value.id ?? null) : value
 }
+const normalizeUnitId = (unit) => {
+  const value = unit && typeof unit === "object" ? unit.id : unit
+  if (typeof value !== "number" && typeof value !== "string") return null
+
+  const id = Number(value)
+  return Number.isSafeInteger(id) && id > 0 ? id : null
+}
 const normalizeCreatedDocumentRow = (row) => {
   let normalizedRow = row
 
@@ -358,6 +366,7 @@ const normalizeCreatedDocumentRow = (row) => {
   return {
     ...normalizedRow,
     id: normalizedRow.id || uuidv4(),
+    unit: normalizeUnitId(normalizedRow.unit),
     costCentre: normalizedRow.costCentre || normalizedRow.costcentre || null,
     purchasePrice: Number(normalizedRow.purchasePrice || 0),
     linkedEntitys: Array.isArray(normalizedRow.linkedEntitys) ? normalizedRow.linkedEntitys : [],
@@ -688,6 +697,7 @@ const setupPage = async () => {
   }
 
 
+  itemInfo.value.rows = normalizeCreatedDocumentRows(itemInfo.value.rows)
   loaded.value = true
 
 }
@@ -2034,19 +2044,8 @@ const setRowData = async (row, service = {sellingPriceComposed: {}}, product = {
 
   if (row.service) {
     console.log(service)
-    if(service.unit) {
-      row.unit = service.unit
-    } else {
-      let selectedService = services.value.find(i => i.id === row.service)
-      console.log(selectedService)
-      if(selectedService.unit?.id) {
-        row.unit = selectedService.unit.id
-      } else {
-        row.unit = selectedService.unit
-      }
-    }
+    row.unit = normalizeUnitId(service?.unit ?? services.value.find(i => i.id === row.service)?.unit)
 
-    //row.unit = service.unit ? service.unit : services.value.find(i => i.id === row.service).unit
     row.inputPrice = ((service.sellingPriceComposed.total || service.sellingPrice) ? (service.sellingPriceComposed.total || service.sellingPrice) : (services.value.find(i => i.id === row.service).sellingPriceComposed.total || services.value.find(i => i.id === row.service).sellingPrice))
     row.description = service.description ? service.description : (services.value.find(i => i.id === row.service) ? services.value.find(i => i.id === row.service).description : "")
     row.costCentre = service.costcentre || services.value.find(i => i.id === row.service)?.costcentre || itemInfo.value.costcentre
@@ -2060,7 +2059,7 @@ const setRowData = async (row, service = {sellingPriceComposed: {}}, product = {
 
   if (row.product) {
     console.log("Product Detected")
-    row.unit = product.unit ? product.unit : products.value.find(i => i.id === row.product).unit
+    row.unit = normalizeUnitId(product?.unit ?? products.value.find(i => i.id === row.product)?.unit)
     row.inputPrice = (product.selling_price ? product.selling_price : products.value.find(i => i.id === row.product).selling_price)
     row.costCentre = row.costCentre || itemInfo.value.costcentre
     //row.price = Number((row.originalPrice * (1 + itemInfo.value.customSurchargePercentage /100)).toFixed(2))
@@ -2094,7 +2093,7 @@ const setRowData = async (row, service = {sellingPriceComposed: {}}, product = {
       <ArchiveButton
           color="error"
           type="createddocuments"
-          v-if="!isTemplateMode && (itemInfo.state === 'Entwurf' || itemInfo.type === 'serialInvoices')"
+          v-if="itemInfo.id && !isTemplateMode && (itemInfo.state === 'Entwurf' || itemInfo.type === 'serialInvoices')"
           variant="outline"
           @confirmed="useEntities('createddocuments').update(itemInfo.id,{archived: true}),
                       router.push('/')"

frontend/tests/agent-setup.test.cjs

@@ -85,13 +85,15 @@ test('stale or inactive agents cannot finish setup; fresh reports save the chose
   await ui.run('finishSetup()')
   assert.equal(ui.run('online.value'), false)
   assert.equal(ui.requests.length, 0)
-  ui.report({ ...baseAgent, lastSeenAt: new Date().toISOString(), scannerNames: [scanner], capabilities: { scan: true } })
+  ui.report({ ...baseAgent, lastSeenAt: new Date().toISOString(), scannerNames: [scanner], capabilities: { scan: true, ocr: true, backgroundCleanup: true } })
   await ui.run('checkConnection()')
   ui.run('profile.value="ix1600"; step.value=3')
   await ui.run('finishSetup()')
   assert.equal(ui.run('saved.value'), true)
   assert.equal(ui.requests[0][2].preferredScannerName, scanner)
   assert.equal(ui.requests[0][2].scanDefaults.source, 'ADF Duplex')
+  assert.equal(ui.requests[0][2].scanDefaults.ocr, true)
+  assert.equal(ui.requests[0][2].scanDefaults.backgroundCleanup, true)
   assert.equal(ui.run('token.value'), '')
   assert.equal(ui.timers.size, 0)
   ui.run('openTestScan()')
@@ -108,6 +110,16 @@ test('online agent without a detected scanner remains incomplete', async t => {
   assert.equal(ui.requests.length, 0)
 })
 
+test('setup cannot finish with an agent missing the processing required by its profile', async t => {
+  const ui = await setup(t, { ...baseAgent })
+  ui.report({ ...baseAgent, lastSeenAt: new Date().toISOString(), scannerNames: [scanner], capabilities: { scan: true } })
+  await ui.run('checkConnection()')
+  ui.run('profile.value="ix1600"; step.value=3')
+  await ui.run('finishSetup()')
+  assert.equal(ui.run('processingMissing.value'), true)
+  assert.equal(ui.requests.length, 0)
+})
+
 test('connection errors stop automatic requests and are visible for retry', async t => {
   const ui = await setup(t, { ...baseAgent }, { getInstanceAgents: async () => { throw Error('offline') } })
   ui.run('step.value=2')

frontend/tests/document-editor-units.test.cjs

@@ -0,0 +1,97 @@
+const { test } = require('node:test')
+const assert = require('node:assert/strict')
+const fs = require('node:fs')
+const vm = require('node:vm')
+const ts = require('typescript')
+const dayjs = require('dayjs')
+const { parse } = require('@vue/compiler-sfc')
+
+const { descriptor } = parse(fs.readFileSync(`${__dirname}/../pages/createDocument/edit/[[id]].vue`, 'utf8'))
+const script = ts.createSourceFile('document-editor.js', descriptor.scriptSetup.content, ts.ScriptTarget.Latest, true, ts.ScriptKind.JS)
+const handlers = new Set(['normalizeEntityId', 'normalizeUnitId', 'normalizeCreatedDocumentRow', 'normalizeCreatedDocumentRows', 'normalizeTaxTypeValue', 'setupPage', 'setRowData'])
+const source = script.statements
+  .filter(statement => ts.isVariableStatement(statement) && statement.declarationList.declarations.some(declaration => handlers.has(declaration.name.getText(script))))
+  .map(statement => statement.getText(script)).join('\n')
+
+function fixture(unit) {
+  const ref = value => ({ value })
+  const product = { id: 20, name: 'Artikel', unit, selling_price: 10, tax_percentage: 19 }
+  const service = { id: 30, name: 'Leistung', unit, sellingPrice: 20, sellingPriceComposed: {}, taxPercentage: 19 }
+  const context = vm.createContext({
+    dayjs, uuidv4: () => 'row-id', console: { log() {} },
+    products: ref([product]), services: ref([service]),
+    productcategories: ref([]), servicecategories: ref([]),
+    itemInfo: ref({ type: 'invoices', taxType: 'Standard', rows: [], agriculture: {} }),
+    route: { params: {}, query: {} }, loaded: ref(false), templateName: ref(''),
+    taxTypeItems: [{ key: 'Standard' }],
+    dataStore: { documentTypesForCreation: { invoices: { labelSingle: 'Rechnung' } } },
+    setupData: async () => {}, setDocumentTypeConfig() {}, setContactPersonData: async () => {},
+    checkCompatibilityWithInputPrice() {}, updateCustomSurcharge() {},
+    useEntities: relation => ({
+      select: async () => relation === 'products' ? [product] : [service],
+      selectSingle: async () => {
+        if (relation === 'documenttemplates') return { name: 'Vorlage', documentType: 'invoices', templateData: { rows: [{ id: 'saved-row', mode: 'normal', product: 20, unit }] } }
+        return { id: 40, type: 'invoices', taxType: 'Standard', rows: [{ id: 'saved-row', mode: 'normal', product: 20, unit }] }
+      },
+    }),
+  })
+  vm.runInContext(source, context)
+  return { context, run: code => vm.runInContext(code, context) }
+}
+
+const formats = [1, { id: 1, name: 'Stück' }, '1', { id: '1', name: 'Stück' }]
+
+test('article and service selection resolve relation objects and legacy IDs to a selectable unit', async () => {
+  for (const unit of formats) {
+    const { run } = fixture(unit)
+    for (const entity of ['product', 'service']) {
+      run(`itemInfo.value.rows = [{ ${entity}: ${entity === 'product' ? 20 : 30} }]`)
+      await run('setRowData(itemInfo.value.rows[0])')
+      assert.equal(run('itemInfo.value.rows[0].unit'), 1, `${entity}: ${JSON.stringify(unit)}`)
+    }
+  }
+})
+
+test('article and service modal results retain their selected unit', async () => {
+  for (const unit of formats) {
+    const { run } = fixture(unit)
+    run('itemInfo.value.rows = [{}]')
+    await run('setRowData(itemInfo.value.rows[0], null, products.value[0])')
+    assert.equal(run('itemInfo.value.rows[0].unit'), 1)
+    run('itemInfo.value.rows = [{}]')
+    await run('setRowData(itemInfo.value.rows[0], services.value[0], null)')
+    assert.equal(run('itemInfo.value.rows[0].unit'), 1)
+  }
+})
+
+test('saved documents and both template entry points restore units without overwriting the stored choice', async () => {
+  for (const route of [{ params: { id: 40 }, query: {} }, { params: {}, query: { templateId: 50 } }, { params: {}, query: { templateFromDocument: 40 } }]) {
+    for (const unit of formats) {
+      const { context, run } = fixture(unit)
+      context.route = route
+      context.products.value[0].unit = { id: 2, name: 'Meter' }
+      await run('setupPage()')
+      assert.equal(run('loaded.value'), true)
+      assert.equal(run('itemInfo.value.rows[0].unit'), 1, JSON.stringify(route))
+      assert.equal(run('itemInfo.value.rows[0].id'), 'saved-row')
+    }
+  }
+})
+
+test('imported JSON rows preserve quantity, prices and the stored unit', () => {
+  const { run, context } = fixture(1)
+  context.rows = [JSON.stringify({ id: 'imported', mode: 'normal', unit: { id: '1' }, quantity: 2.5, price: 12.34 })]
+  const row = run('normalizeCreatedDocumentRows(rows)[0]')
+  assert.equal(row.unit, 1)
+  assert.equal(row.id, 'imported')
+  assert.equal(row.quantity, 2.5)
+  assert.equal(row.price, 12.34)
+})
+
+test('missing and malformed units remain unselected', () => {
+  const { run, context } = fixture(1)
+  for (const unit of [null, undefined, '', {}, { id: null }, 'invalid', 0, -1, true]) {
+    context.row = { unit }
+    assert.equal(run('normalizeCreatedDocumentRow(row).unit'), null)
+  }
+})

frontend/tests/file-scan.test.cjs

@@ -4,7 +4,7 @@ const fs = require('node:fs')
 const vm = require('node:vm')
 const vue = require('vue')
 
-async function setup(getJob, scanData = {}, agents = [{ id: 'agent', active: true, capabilities: { scan: true }, scannerNames: ['fujitsu:ScanSnap iX1600:test'] }]) {
+async function setup(getJob, scanData = {}, agents = [{ id: 'agent', active: true, capabilities: { scan: true, ocr: true, backgroundCleanup: true }, scannerNames: ['fujitsu:ScanSnap iX1600:test'] }]) {
   const notices = []
   const requests = []
   let unmount
@@ -38,6 +38,9 @@ test('duplex scanning continues past the former three-minute polling limit', asy
   const body = ui.requests.find(request => request.url === '/api/scan-jobs').options.body
   assert.equal(body.settings.source, 'ADF Duplex')
   assert.equal(body.settings.paperSize, 'A4')
+  assert.equal(body.settings.ocr, true)
+  assert.equal(body.settings.ocrLanguage, 'deu+eng')
+  assert.equal(body.settings.backgroundCleanup, true)
   assert.equal(body.target.folder, 'folder-1')
 })
 
@@ -72,7 +75,7 @@ test('iX1600 preset resets image settings and normalizes the output filename to
 })
 
 test('setup test scan uses the requested agent even when another agent is first', async () => {
-  const agents = ['first', 'configured'].map(id => ({ id, active: true, capabilities: { scan: true }, scannerNames: ['scanner-' + id] }))
+  const agents = ['first', 'configured'].map(id => ({ id, active: true, capabilities: { scan: true, ocr: true, backgroundCleanup: true }, scannerNames: ['scanner-' + id] }))
   const ui = await setup(() => ({ job: { status: 'completed' } }), { agentId: 'configured' }, agents)
   await ui.run('startScan()')
   assert.equal(ui.requests.find(request => request.url === '/api/scan-jobs').options.body.agentId, 'configured')
@@ -83,3 +86,10 @@ test('an unavailable setup agent never silently scans with another device', asyn
   await ui.run('startScan()')
   assert.equal(ui.requests.filter(request => request.url === '/api/scan-jobs').length, 0)
 })
+
+test('an old agent cannot silently produce an image-only PDF when OCR was requested', async () => {
+  const ui = await setup(() => ({}), {}, [{ id: 'old', active: true, capabilities: { scan: true }, scannerNames: ['scanner'] }])
+  await ui.run('startScan()')
+  assert.equal(ui.requests.filter(request => request.url === '/api/scan-jobs').length, 0)
+  assert.equal(ui.notices[0].title, 'Agent aktualisieren')
+})

frontend/tests/selfhost-updates.test.cjs

@@ -99,3 +99,90 @@ test('an older polling response cannot erase a new submission whose response was
   assert.equal(f.ui.pending.value.body.requestId, id)
   assert.equal(f.ui.canApply.value, false)
 })
+
+test('failed jobs automatically show logs and refresh them only while open', async t => {
+  const failed = { status: 'failed', command: 'apply', requestId: 'failed-request', error: { code: 'UPDATE_FAILED', message: 'Auftrag fehlgeschlagen.' } }
+  let content = 'pull access denied'
+  const f = await setup(t, async path => {
+    if (path.endsWith('/logs')) return { requestId: failed.requestId, logs: [{ source: 'commands', content, truncated: false }] }
+    return path.endsWith('/status') ? { ...overview(failed), capabilities: { requestLogs: true } } : failed
+  })
+  assert.equal(f.ui.logsOpen.value, true)
+  assert.equal(f.ui.logEntries.value[0].content, content)
+  content = 'final failure detail'
+  await f.ui.refresh()
+  assert.equal(f.ui.logEntries.value[0].content, content)
+  await f.ui.toggleLogs()
+  const count = f.calls.filter(c => c.path.endsWith('/logs')).length
+  await f.ui.refresh()
+  assert.equal(f.ui.logsOpen.value, false)
+  assert.equal(f.calls.filter(c => c.path.endsWith('/logs')).length, count)
+})
+
+test('older host services explain the required upgrade without requesting unsupported endpoints', async t => {
+  const failed = { status: 'failed', command: 'apply', requestId: 'failed-request' }
+  const f = await setup(t, async path => path.endsWith('/status') ? overview(failed) : failed)
+  assert.equal(f.ui.logsOpen.value, true)
+  assert.match(f.ui.logsMessage.value, /Host-Update-Dateien aktualisieren/)
+  assert.equal(f.calls.some(c => c.path.endsWith('/logs')), false)
+  assert.equal(f.ui.connected.value, true)
+})
+
+test('log failures preserve job state and revoked admin access clears the displayed logs', async t => {
+  let failure
+  const f = await setup(t, async path => {
+    if (path.endsWith('/logs')) {
+      if (failure) throw failure
+      return { requestId: check.requestId, logs: [{ source: 'commands', content: 'diagnostic', truncated: false }] }
+    }
+    return path.endsWith('/status') ? { ...overview(), capabilities: { requestLogs: true } } : check
+  })
+  await f.ui.toggleLogs()
+  failure = { statusCode: 503 }
+  await f.ui.refreshLogs()
+  assert.match(f.ui.logsMessage.value, /konnte nicht geladen/)
+  assert.equal(f.ui.connected.value, true)
+  assert.equal(f.ui.job.value.status, 'checked')
+  failure = { statusCode: 403 }
+  await f.ui.refreshLogs()
+  assert.equal(f.ui.accessDenied.value, true)
+  assert.equal(f.ui.logEntries.value.length, 0)
+})
+
+test('a delayed log response cannot appear under a newly submitted job', async t => {
+  let release
+  const f = await setup(t, async (path, options) => {
+    if (options.method === 'POST') return { status: 'queued', requestId: options.body.requestId, command: 'check' }
+    if (path.endsWith('/logs')) return new Promise(resolve => { release = resolve })
+    return path.endsWith('/status') ? { ...overview(), capabilities: { requestLogs: true } } : check
+  })
+  const loading = f.ui.toggleLogs()
+  await new Promise(setImmediate)
+  await f.ui.check()
+  release({ requestId: check.requestId, logs: [{ source: 'commands', content: 'previous job', truncated: false }] })
+  await loading
+  assert.notEqual(f.ui.job.value.requestId, check.requestId)
+  assert.equal(f.ui.logEntries.value.length, 0)
+})
+
+test('the update card renders log output as escaped text and explains truncation', async t => {
+  const f = await setup(t, async path => path.endsWith('/status') ? overview() : check)
+  f.ui.logsOpen.value = true
+  const { parse, compileScript } = require('@vue/compiler-sfc')
+  const descriptor = parse(fs.readFileSync(__dirname + '/../components/SelfhostUpdateCard.vue', 'utf8')).descriptor
+  const compiled = compileScript(descriptor, { id: 'update-card-test', inlineTemplate: true }).content
+  const exports = {}
+  const context = vm.createContext({ exports, require, ref: vue.ref, computed: vue.computed,
+    useRuntimeConfig: () => ({ public: {} }), useSelfhostUpdates: () => ({ ...f.ui,
+      logEntries: vue.ref([{ source: 'commands', content: '<script>alert("log")</script>', truncated: true }]) }) })
+  vm.runInContext(ts.transpile(compiled, { target: ts.ScriptTarget.ES2022, module: ts.ModuleKind.CommonJS }), context)
+  const app = vue.createSSRApp(exports.default)
+  for (const name of ['UCard', 'UIcon', 'UBadge', 'UAlert', 'UProgress', 'UButton', 'UModal', 'UCheckbox']) {
+    app.component(name, { setup: (_props, { slots }) => () => vue.h('div', slots.default?.()) })
+  }
+  const html = await require('@vue/server-renderer').renderToString(app)
+  assert.match(html, /Werkzeugausgaben/)
+  assert.match(html, /Gekürzte Anzeige/)
+  assert.match(html, /&lt;script&gt;/)
+  assert.doesNotMatch(html, /<script>/)
+})

frontend/utils/agentInstaller.mjs

@@ -26,6 +26,7 @@ export function renderAgentInstaller(bundle, options) {
     'FEDEO_POLL_SECONDS=5', 'FEDEO_SCAN_FORMAT=pdf', 'FEDEO_SCAN_RESOLUTION=300',
     'FEDEO_SCAN_MODE=Color', 'FEDEO_SCAN_TIMEOUT_SECONDS=900',
     'FEDEO_SCAN_POSTPROCESS=false', 'FEDEO_SCAN_POSTPROCESS_PROFILE=document', '',
+    'FEDEO_SCAN_BACKGROUND_CLEANUP=true', 'FEDEO_SCAN_OCR=true', 'FEDEO_SCAN_OCR_LANGUAGE=deu+eng', '',
   ].join('\n')
   const replacements = {
     AGENT_ID: options.agentId, AUTOSTART: String(Boolean(options.autostart)),
@@ -40,6 +41,7 @@ export function renderAgentInstaller(bundle, options) {
 export const ix1600ScanDefaults = {
   format: 'pdf', resolution: 300, mode: 'Color', source: 'ADF Duplex',
   paperSize: 'A4', postprocess: false, postprocessProfile: 'document',
+  backgroundCleanup: true, ocr: true, ocrLanguage: 'deu+eng',
 }
 
 export function agentSetupReady(agent, since = 0, now = Date.now()) {

scripts/selfhost-update-service.py

@@ -14,6 +14,7 @@ import socketserver
 import stat
 import threading
 import traceback
+import urllib.parse
 
 spec = importlib.util.spec_from_file_location("fedeo_updater", Path(__file__).with_name("selfhost-update.py"))
 engine = importlib.util.module_from_spec(spec)
@@ -33,7 +34,8 @@ ERRORS = {
     "COMPOSE_REVIEW_REQUIRED": "Die Compose-Datei muss zuerst auf dem Host geprüft und übernommen werden.",
     "RECOVERY_REQUIRED": "Ein vorheriges Update benötigt Wiederherstellung auf dem Host.",
     "PLAN_CHANGED": "Release oder Konfiguration haben sich geändert. Erneut prüfen.",
-    "UPDATE_FAILED": "Auftrag fehlgeschlagen. Geschützte Protokolle auf dem Host prüfen.",
+    "UPDATE_FAILED": "Auftrag fehlgeschlagen.",
+    "LOGS_UNAVAILABLE": "Das Update-Protokoll kann gerade nicht gelesen werden.",
     "INTERRUPTED": "Der Auftrag wurde unterbrochen. Status auf dem Host prüfen.",
 }
 
@@ -82,6 +84,74 @@ def public_plan(plan):
     return result
 
 
+LOG_BYTES = 6000
+SECRET_KEY = r"[\w.-]*(?:password|passwd|secret|token|credential|authorization|(?:api|private|access|encryption|signing)[_-]?key|database[_-]?url|dsn|auth)[\w.-]*"
+QUOTED_VALUE = r"\"(?:\\.|[^\"\\])*\"|'[^']*'"
+SECRET_VALUE = re.compile(r"(?im)(?<![\w.-])([\"']?" + SECRET_KEY + r"[\"']?\s*[:=]\s*)(" + QUOTED_VALUE + r"|[^\r\n,}\]]+)")
+CONFIG_SECRET = re.compile(r"(?im)^\s*(?:export\s+|-\s*)?[\"']?" + SECRET_KEY + r"[\"']?\s*[:=]\s*(" + QUOTED_VALUE + r"|[^\r\n]+)")
+
+
+def read_private(path, *, tail=False):
+    """Read only regular files, never follow a log/config symlink or block on a FIFO."""
+    try:
+        fd = os.open(path, os.O_RDONLY | os.O_NOFOLLOW | os.O_NONBLOCK)
+    except FileNotFoundError:
+        return "", False
+    with os.fdopen(fd, "rb") as stream:
+        size = os.fstat(stream.fileno()).st_size
+        if not stat.S_ISREG(os.fstat(stream.fileno()).st_mode):
+            raise OSError("Not a regular file")
+        limit = 32768 if tail else 1024 * 1024
+        if size > limit and not tail:
+            raise OSError("Configuration too large to redact safely")
+        truncated = size > limit
+        if truncated:
+            stream.seek(-limit, os.SEEK_END)
+            stream.readline(limit)  # Never expose a suffix of a secret on a cut line.
+        return stream.read(limit).decode("utf-8", errors="replace"), truncated
+
+
+def log_redactor(directory, job):
+    # Include original values after an update changed the active configuration.
+    paths = [directory / name for name in (".env", "docker-compose.yml", "docker-compose.override.yml", "docker-compose.override.yaml")]
+    if job:
+        paths += [job / "target.env", job / "target.compose.yml"]
+        paths += [job / "backup/config" / name for name in (".env", "docker-compose.yml", "docker-compose.override.yml", "docker-compose.override.yaml")]
+    values = {value for key, value in os.environ.items() if re.fullmatch(SECRET_KEY, key, re.I) and value}
+    for path in paths:
+        content, _ = read_private(path)
+        for match in CONFIG_SECRET.finditer(content):
+            value = match[1].strip()
+            if value[:1] in ("'", '"'):
+                if value.startswith('"'):
+                    try:
+                        values.add(json.loads(value))
+                    except ValueError:
+                        pass
+                value = value[1:-1]
+            else:
+                value = value.split(" #", 1)[0].strip()
+            if value:
+                values.add(value)
+        for match in re.finditer(r"[a-z][a-z0-9+.-]*://[^\s\"']+", content, re.I):
+            password = urllib.parse.urlsplit(match[0]).password
+            if password:
+                values.update((password, urllib.parse.unquote(password)))
+    values |= {urllib.parse.quote(value, safe="") for value in values}
+
+    def redact(content):
+        content = re.sub(r"\x1b(?:\[[0-?]*[ -/]*[@-~]|\][^\x07]*(?:\x07|\x1b\\))", "", content)
+        for value in sorted(values, key=len, reverse=True):
+            content = content.replace(value, "[REDACTED]")
+        content = re.sub(r"(?is)-----BEGIN [^-]*PRIVATE KEY-----.*?(?:-----END [^-]*PRIVATE KEY-----|\Z)", "[REDACTED]", content)
+        content = re.sub(r"(?i)([a-z][a-z0-9+.-]*://)[^\s/]*@", r"\1[REDACTED]@", content)
+        content = re.sub(r"(?i)\b(Bearer|Basic|token)\s+[A-Za-z0-9._~+/=-]+", r"\1 [REDACTED]", content)
+        content = SECRET_VALUE.sub(lambda m: m[1] + "[REDACTED]", content)
+        content = re.sub(r"(?i)(--" + SECRET_KEY + r"\s+)(" + QUOTED_VALUE + r"|\S+)", r"\1[REDACTED]", content)
+        return re.sub(r"[\x00-\x08\x0b-\x1f\x7f]", "", content)
+    return redact
+
+
 class Controller:
     def __init__(self, directory, *, updater_factory=engine.Updater):
         self.directory = Path(directory).resolve()
@@ -140,9 +210,47 @@ class Controller:
                 result["progress"] = public_state(state)
         return result
 
+    def request_logs(self, request_id):
+        record = self.read(request_id)
+        try:
+            home = self.home.parent
+            state = engine.read_json(home / "status.json", {})
+            candidates = [Path(record["job"])] if record.get("job") else []
+            if state.get("requestId") == request_id and state.get("job"):
+                candidates.append(Path(state["job"]))
+            # Older service versions did not persist the engine-job association.
+            if not candidates:
+                candidates = [path.parent for path in home.glob("*/status.json")]
+            job = None
+            for candidate in candidates:
+                if candidate.parent != home or candidate.is_symlink() or not candidate.is_dir():
+                    continue
+                content, _ = read_private(candidate / "status.json")
+                if content and json.loads(content).get("requestId") == request_id:
+                    job = candidate
+                    break
+            redact = log_redactor(self.directory, job)
+            files = [("service", self.home / (request_id + ".log"))]
+            if job:
+                files += [("events", job / "events.jsonl"), ("commands", job / "commands.log")]
+            logs = []
+            for source, path in files:
+                content, truncated = read_private(path, tail=True)
+                content = redact(content)
+                encoded = content.encode("utf-8")
+                if len(encoded) > LOG_BYTES:
+                    content = encoded[-LOG_BYTES:].decode("utf-8", errors="ignore")
+                    truncated = True
+                if content or truncated:
+                    logs.append({"source": source, "content": content, "truncated": truncated})
+            return {"requestId": request_id, "logs": logs}
+        except (OSError, ValueError, TypeError):
+            raise ApiError("LOGS_UNAVAILABLE", 503) from None
+
     def status(self):
         updater = self.factory(self.directory)
-        result = {"apiVersion": 1, "engine": public_state(updater.status()), "acceptingRequests": not self.closing}
+        result = {"apiVersion": 1, "engine": public_state(updater.status()), "acceptingRequests": not self.closing,
+                  "capabilities": {"requestLogs": True}}
         installed = engine.read_json(updater.home / "installed.json", {}).get("manifest", {})
         result["installed"] = {k: v for k, v in public_state({"version": installed.get("version"),
                               "revision": installed.get("commit")}).items() if k != "status"}
@@ -208,6 +316,7 @@ class Controller:
 
     def run(self, record, fingerprint):
         request_id = record["requestId"]
+        updater = None
         try:
             record.update(status="running", updatedAt=engine.now())
             engine.write_json(self.path(request_id), record)
@@ -224,6 +333,8 @@ class Controller:
                 os.chmod(log.name, 0o600)
                 traceback.print_exc(file=log)
         finally:
+            if updater and updater.state and updater.state.get("requestId") == request_id:
+                record["job"] = str(updater.job)
             record["updatedAt"] = engine.now()
             engine.write_json(self.path(request_id), record)
 
@@ -267,7 +378,9 @@ class Handler(BaseHTTPRequestHandler):
             if self.command == "GET" and self.path == "/v1/status":
                 self.send_json(200, controller.status())
             elif self.command == "GET" and self.path.startswith("/v1/requests/"):
-                self.send_json(200, controller.request_status(self.path.removeprefix("/v1/requests/")))
+                suffix = self.path.removeprefix("/v1/requests/")
+                self.send_json(200, controller.request_logs(suffix.removesuffix("/logs"))
+                               if suffix.endswith("/logs") else controller.request_status(suffix))
             elif self.command == "POST" and self.path in ("/v1/check", "/v1/apply"):
                 lengths = self.headers.get_all("Content-Length", [])
                 if (len(lengths) != 1 or not lengths[0].isdigit() or not 1 <= int(lengths[0]) <= 4096

scripts/selfhost-update.py

@@ -566,7 +566,13 @@ class Updater:
         if not self.managed and not accept_compose:
             raise UpdateError("Compose-Datei ist noch nicht verwaltet oder wurde bearbeitet. compose.diff prüfen, Anpassungen in die Override-Datei übertragen und --accept-compose-replacement verwenden.")
         self.save("pulling")
-        self.compose(True, "pull", timeout=3600)
+        # Exact digests are immutable: reuse local copies even when their registry is offline.
+        # Mutable infrastructure tags must still be refreshed before entering maintenance.
+        for policy in ("missing", "always"):
+            services = [name for name, service in self.candidate["services"].items()
+                        if bool(re.search(r"@sha256:[a-f0-9]{64}$", service["image"])) == (policy == "missing")]
+            if services:
+                self.compose(True, "pull", "--policy", policy, *services, timeout=3600)
         self.check_space(self.backup_size, self.home)
         if any(sha(Path(path).read_bytes()) != value for path, value in self.before.items()):
             raise UpdateError("Konfiguration wurde während der Vorbereitung verändert; erneut prüfen.")
@@ -590,13 +596,13 @@ class Updater:
         atomic(self.root / ".env.example", (self.job / ".env.example").read_bytes())
         atomic(self.root / "scripts/selfhost-setup.sh", (self.job / "selfhost-setup.sh").read_bytes())
         os.chmod(self.root / "scripts/selfhost-setup.sh", 0o700)
-        self.compose(True, "up", "-d", "--no-deps", "--wait", "--wait-timeout", "180", "db", "minio", "node-exporter")
-        self.compose(True, "run", "--rm", "--no-deps", "createbuckets")
+        self.compose(True, "up", "--pull", "never", "-d", "--no-deps", "--wait", "--wait-timeout", "180", "db", "minio", "node-exporter")
+        self.compose(True, "run", "--pull", "never", "--rm", "--no-deps", "createbuckets")
         self.save("migrating", migrationContainer="fedeo-update-" + self.job.name)
-        self.compose(True, "run", "--rm", "--no-deps", "--name", self.state["migrationContainer"],
+        self.compose(True, "run", "--pull", "never", "--rm", "--no-deps", "--name", self.state["migrationContainer"],
                      "-e", "FEDEO_RUN_MIGRATIONS=false", "--entrypoint", "npm", "backend", "run", "migrate")
         self.save("starting")
-        self.compose(True, "up", "-d", "--wait", "--wait-timeout", "600", timeout=900)
+        self.compose(True, "up", "--pull", "never", "-d", "--wait", "--wait-timeout", "600", timeout=900)
         self.save("verifying")
         self.verify_matrix_stopped(retired=True)
         for name, image in self.images.items():

scripts/update/tests/test_service.py

@@ -8,6 +8,7 @@ import subprocess
 import tempfile
 import threading
 import unittest
+from unittest.mock import patch
 
 ROOT = Path(__file__).resolve().parents[3]
 spec = importlib.util.spec_from_file_location('update_service', ROOT / 'scripts/selfhost-update-service.py')
@@ -232,6 +233,86 @@ class ServiceTests(unittest.TestCase):
         self.assertEqual(self.request('GET', '/v1/logs')[0], 404)
         self.assertEqual((self.controller.home / 'check-request-01.log').stat().st_mode & 0o777, 0o600)
 
+    def test_logs_include_failure_commands_and_events_with_credentials_redacted(self):
+        (self.root / '.env').write_text('POSTGRES_PASSWORD="database-secret"\nAPI_TOKEN=api-secret\n'
+                                      'SMTP_PASSWORD=complex,secret/suffix\n'
+                                      'DATABASE_URL=postgres://user:db%2Fsecret@db/fedeo\n')
+        (self.root / 'docker-compose.override.yml').write_text('  MINIO_ROOT_PASSWORD: "minio-secret"\n')
+        self.factory.failure = s.engine.UpdateError('Registry unauthorized: database-secret')
+        request_id = self.check()
+        job = self.root / '.updates' / ('job-' + request_id)
+        (job / 'target.env').write_text("SMTP_PASSWORD='previous-secret'\n")
+        (job / 'commands.log').write_text(
+            '\x1b[31mError: pull access denied\x1b[0m\n'
+            'database-secret api-secret minio-secret previous-secret release-secret\n'
+            'complex,secret/suffix complex%2Csecret%2Fsuffix db/secret\n'
+            'https://user:url-secret@registry.test/image\n'
+            'Authorization: Bearer header-secret\n'
+            '{"password": "json-secret", "access_token": "query-secret"}\n'
+            'TOKEN=unquoted-secret\n--password cli-secret\n')
+        with patch.dict(s.os.environ, {'FEDEO_RELEASE_TOKEN': 'release-secret'}):
+            status, result = self.request('GET', f'/v1/requests/{request_id}/logs')
+        self.assertEqual(status, 200)
+        self.assertEqual(result['requestId'], request_id)
+        logs = {entry['source']: entry['content'] for entry in result['logs']}
+        self.assertIn('Registry unauthorized', logs['service'])
+        self.assertIn('pull access denied', logs['commands'])
+        self.assertIn('validating', logs['events'])
+        for secret in ('database-secret', 'api-secret', 'minio-secret', 'previous-secret', 'release-secret',
+                       'url-secret', 'header-secret', 'json-secret', 'query-secret', 'unquoted-secret', 'cli-secret',
+                       'complex,secret/suffix', 'complex%2Csecret%2Fsuffix', 'db/secret', '\x1b'):
+            self.assertNotIn(secret, json.dumps(result))
+        self.assertTrue(self.request('GET', '/v1/status')[1]['capabilities']['requestLogs'])
+
+    def test_logs_survive_restart_and_find_legacy_jobs_without_crossing_requests(self):
+        first = self.check()
+        first_job = self.root / '.updates' / ('job-' + first)
+        (first_job / 'commands.log').write_text('first request only\n')
+        # Simulate the private request record from an older host service.
+        record = self.controller.read(first)
+        record.pop('job')
+        s.engine.write_json(self.controller.path(first), record)
+        second = self.check('check-request-02')
+        (self.root / '.updates' / ('job-' + second) / 'commands.log').write_text('second request only\n')
+        self.controller.close()
+        self.controller = s.Controller(self.root, updater_factory=self.factory)
+        self.server.controller = self.controller
+        for request_id, own, other in [(first, 'first', 'second'), (second, 'second', 'first')]:
+            payload = json.dumps(self.request('GET', f'/v1/requests/{request_id}/logs')[1])
+            self.assertIn(own + ' request only', payload)
+            self.assertNotIn(other + ' request only', payload)
+        self.assertEqual(self.request('GET', '/v1/requests/missing-request/logs')[0], 404)
+        self.assertEqual(self.request('GET', '/v1/requests/../../secret/logs')[0], 400)
+
+    def test_running_logs_are_bounded_and_do_not_expose_partial_secret_lines(self):
+        self.factory.release.clear()
+        request_id = 'check-running-logs'
+        self.request('POST', '/v1/check', {'requestId': request_id})
+        self.assertTrue(self.factory.started.wait(3))
+        job = self.root / '.updates' / ('job-' + request_id)
+        (job / 'commands.log').write_text('PASSWORD=' + 'secret-fragment' * 4000 + '\n' + 'old entry\n' * 900 + 'Last diagnostic: ä\n')
+        result = self.request('GET', f'/v1/requests/{request_id}/logs')[1]
+        commands = next(entry for entry in result['logs'] if entry['source'] == 'commands')
+        self.assertTrue(commands['truncated'])
+        self.assertLessEqual(len(commands['content'].encode()), s.LOG_BYTES)
+        self.assertIn('Last diagnostic: ä', commands['content'])
+        self.assertNotIn('secret-fragment', commands['content'])
+        self.assertLess(len(json.dumps(result).encode()), 65536)
+
+    def test_logs_fail_closed_for_symlinks_and_unreadable_redaction_configuration(self):
+        request_id = self.check()
+        job = self.root / '.updates' / ('job-' + request_id)
+        private = self.root / 'private.txt'
+        private.write_text('never publish this')
+        (job / 'commands.log').symlink_to(private)
+        status, result = self.request('GET', f'/v1/requests/{request_id}/logs')
+        self.assertEqual(status, 503)
+        self.assertEqual(result['error']['code'], 'LOGS_UNAVAILABLE')
+        self.assertNotIn('never publish this', json.dumps(result))
+        (job / 'commands.log').unlink()
+        (self.root / '.env').symlink_to(private)
+        self.assertEqual(self.request('GET', f'/v1/requests/{request_id}/logs')[0], 503)
+
     def test_api_rejects_unsupported_commands_options_bodies_and_browser_requests(self):
         for body in [[], {'requestId': 'bad/path'}, {'requestId': 'check-request-01', 'version': ';reboot'},
                      {'requestId': 'check-request-01', 'directory': '/etc'},

scripts/update/tests/test_updater.py

@@ -79,6 +79,32 @@ class Runner:
         self.matrix_containers = {}
         self.on_stop = None
         self.real_tar = False
+        self.compose_configs = {}
+        self.local_images = set()
+        self.unavailable_images = set()
+        self.downloads = []
+
+    @staticmethod
+    def image_reference(image):
+        # Docker resolves tag@digest using the repository and digest, not the tag.
+        if '@' not in image:
+            return image
+        name, digest = image.split('@', 1)
+        repository, slash, basename = name.rpartition('/')
+        return repository + slash + basename.split(':', 1)[0] + '@' + digest
+
+    def pull_image(self, image, policy):
+        reference = self.image_reference(image)
+        if policy == 'never':
+            if reference not in self.local_images:
+                raise u.UpdateError('Image missing locally')
+            return
+        if policy == 'missing' and reference in self.local_images:
+            return
+        self.downloads.append(reference)
+        if reference in self.unavailable_images:
+            raise u.UpdateError('Registry unavailable for ' + reference)
+        self.local_images.add(reference)
 
     def run(self, args, output=None, input_file=None, **kwargs):
         self.calls.append(args)
@@ -108,6 +134,7 @@ class Runner:
                                     capture_output=True, text=True)
             if result.returncode:
                 raise u.UpdateError('Compose invalid: ' + result.stderr)
+            self.compose_configs[args[args.index('-f') + 1]] = json.loads(result.stdout)
             return result.stdout
         if 'ps' in args:
             if args[:2] == ['docker', 'ps']:
@@ -121,6 +148,15 @@ class Runner:
                     container['State']['Running'] = 'start' in args
             if 'stop' in args and self.on_stop:
                 self.on_stop()
+        if args[:2] == ['docker', 'compose'] and any(command in args for command in ('pull', 'up', 'run')):
+            services = self.compose_configs[args[args.index('-f') + 1]]['services']
+            selected = [service for name, service in services.items() if name in args]
+            if not selected:
+                selected = list(services.values())
+            for service in selected:
+                policy = (args[args.index('--policy') + 1] if '--policy' in args else 'always') if 'pull' in args else (
+                    args[args.index('--pull') + 1] if '--pull' in args else service.get('pull_policy', 'missing'))
+                self.pull_image(service['image'], policy)
             if 'pull' in args and self.on_pull:
                 self.on_pull()
         if any('pg_database_size' in part for part in args):
@@ -493,6 +529,49 @@ class UpdateTests(unittest.TestCase):
             self.run_update()
         self.assert_untouched()
 
+    def test_cached_exact_images_work_with_unavailable_registries_and_forced_pull_overrides(self):
+        self.updater.execute('check', '3.1.0')
+        services = self.updater.candidate['services']
+        refs = {name: self.runner.image_reference(service['image']) for name, service in services.items()}
+        self.runner.local_images.update(refs.values())
+        self.runner.unavailable_images.update(ref for ref in refs.values() if '@sha256:' in ref)
+        (self.root / 'docker-compose.override.yml').write_text(
+            'services:\n' + ''.join(f'  {name}:\n    pull_policy: always\n' for name in services))
+        result = self.run_update()
+        self.assertEqual(result['status'], 'succeeded')
+        self.assertEqual(set(self.runner.downloads), {refs[name] for name in ('db', 'traefik', 'node-exporter')})
+        self.assertEqual(len(self.runner.downloads), 3)
+
+    def test_missing_digest_is_downloaded_even_if_tag_or_another_digest_is_cached(self):
+        self.updater.execute('check', '3.1.0')
+        services = self.updater.candidate['services']
+        for service in services.values():
+            image = service['image']
+            if '@' in image:
+                self.runner.local_images.add(image.split('@')[0])
+                self.runner.local_images.add(self.runner.image_reference(image).split('@')[0] + '@sha256:' + '0' * 64)
+        self.assertEqual(self.run_update()['status'], 'succeeded')
+        self.assertEqual(set(self.runner.downloads), {self.runner.image_reference(s['image']) for s in services.values()})
+
+    def test_missing_unavailable_digest_stops_before_maintenance_despite_cached_tag(self):
+        self.updater.execute('check', '3.1.0')
+        image = self.updater.candidate['services']['minio']['image']
+        self.runner.local_images.add(image.split('@')[0])
+        self.runner.unavailable_images.add(self.runner.image_reference(image))
+        with self.assertRaisesRegex(u.UpdateError, 'Registry unavailable'):
+            self.run_update()
+        self.assert_untouched()
+        self.assertFalse(self.updater.status()['recoveryRequired'])
+
+    def test_cached_mutable_tag_still_requires_a_successful_registry_refresh(self):
+        self.updater.execute('check', '3.1.0')
+        services = self.updater.candidate['services']
+        self.runner.local_images.update(self.runner.image_reference(s['image']) for s in services.values())
+        self.runner.unavailable_images.add(services['db']['image'])
+        with self.assertRaisesRegex(u.UpdateError, 'Registry unavailable'):
+            self.run_update()
+        self.assert_untouched()
+
     def test_configuration_changed_during_pull_is_detected(self):
         self.runner.on_pull = lambda: (self.root / '.env').write_bytes(self.original_env + b'\nNEW=value\n')
         with self.assertRaisesRegex(u.UpdateError, 'während'):