Update docclass.py

docclass.py

@@ -1623,8 +1623,13 @@ def _extract(actual_type: str, data: bytes) -> tuple[object, bool]:
     return ex, False
 
 
-def analyze_file(data: bytes, filename: str, model: Layer3DocType) -> dict:
-    """All three layers for one file — the handover's classify_file, bounded."""
+def analyze_file(data: bytes, filename: str, model: Layer3DocType, return_text: bool = False) -> dict:
+    """All three layers for one file — the handover's classify_file, bounded.
+
+    return_text: also return the extracted / OCR'd text (at most MAX_TEXT_CHARS),
+    so CLOAK can run its PII scan on images and scanned PDFs without a second OCR
+    pass. Off by default: a caller that only wants the document type is not sent
+    the contents back."""
     analysis = analyze_bytes(data, filename=filename, limits=SCAN_LIMITS, keep_bytes=True)
     identity = analysis.identity
     ex, ocr_used = _extract(identity.actual_type, data)
@@ -1654,6 +1659,7 @@ def analyze_file(data: bytes, filename: str, model: Layer3DocType) -> dict:
         "truncated":         bool(getattr(ex, "truncated", False)),
         "nested_entries":    nested,
         "nested_truncated":  bool(analysis.container.truncated) or len(analysis.container.entries) > MAX_NESTED_CLASSIFIED,
+        **({"text": (getattr(ex, "text", None) or "")[:MAX_TEXT_CHARS]} if return_text else {}),
     }
 
 
@@ -1692,7 +1698,8 @@ def build_router(verify: Callable[[Request], None]) -> APIRouter:
         t0 = time.time()
         from starlette.concurrency import run_in_threadpool
         try:
-            result = await run_in_threadpool(analyze_file, data, filename, model)
+            return_text = request.headers.get("x-return-text", "") == "1"
+            result = await run_in_threadpool(analyze_file, data, filename, model, return_text)
         except HTTPException:
             raise
         except Exception as exc:  # noqa: BLE001