fix(vision): file non immagine e grafo malformato non escono più come 500 (#15)

Worker: Image.open().convert() in try/except (422 "image is not
decodable"); il grafo deve essere un dict (422 altrimenti); run_graph
distingue costruzione (GraphBuildError -> 422) da esecuzione (500 con
motivo). Router: verifica dei magic bytes (PNG/JPEG/BMP/TIFF/WebP/GIF)
prima di chiamare il worker.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

src/backend/api/routers/vision.py

@@ -28,6 +28,29 @@ _IMAGE_EXTENSIONS = {
     "image/webp": "webp",
 }
 
+# Sniffed from the first bytes, independent of the declared Content-Type: a
+# text file uploaded as "image/png" passes the header check above and must
+# still be caught before it reaches the worker as a bare 500 (#15). BMP and
+# TIFF are included even though `_IMAGE_EXTENSIONS` does not accept them,
+# because PIL - and so the worker - can decode both; this check only refuses
+# what is not a real image at all, it does not narrow the allow-list above.
+_IMAGE_SIGNATURES: tuple[tuple[bytes, ...], ...] = (
+    (b"\x89PNG\r\n\x1a\n",),         # PNG
+    (b"\xff\xd8\xff",),              # JPEG
+    (b"BM",),                        # BMP
+    (b"II*\x00", b"MM\x00*"),        # TIFF, little- and big-endian
+    (b"GIF87a", b"GIF89a"),          # GIF
+)
+
+
+def _looks_like_an_image(content: bytes) -> bool:
+    """True if `content` starts with a known image format's magic bytes."""
+    for signatures in _IMAGE_SIGNATURES:
+        if any(content.startswith(sig) for sig in signatures):
+            return True
+    # WebP: "RIFF" + 4-byte size + "WEBP", the size in between is not fixed.
+    return len(content) >= 12 and content[:4] == b"RIFF" and content[8:12] == b"WEBP"
+
 
 async def _task_or_404(db: AsyncSession, task_id: int) -> RecipeTask:
     """Shared by the endpoints of this router."""
@@ -60,6 +83,11 @@ async def _read_image(image: UploadFile) -> bytes:
                 f"{settings.max_upload_size_mb}MB"
             ),
         )
+    if not _looks_like_an_image(content):
+        raise HTTPException(
+            status_code=status.HTTP_422_UNPROCESSABLE_CONTENT,
+            detail="the file's content does not match a supported image format",
+        )
     return content
 
 

src/backend/tests/test_vision_upload_limits.py

@@ -102,6 +102,32 @@ async def test_execute_rejects_a_non_image_upload(
     assert (await db_session.execute(select(Measurement))).scalars().all() == []
 
 
+@pytest.mark.asyncio
+async def test_execute_rejects_a_file_whose_bytes_do_not_match_its_declared_type(
+    client, db_session, measurement_tec_user, worker_says,
+):
+    """#15: a text file declared image/png passes the Content-Type check but
+    must still not reach the worker - it is refused on its magic bytes."""
+    recipe = await create_test_recipe(db_session, measurement_tec_user.id)
+    version, task, subtask = await recipe_parts(db_session, recipe.id)
+    task.task_type = "camera_measure"
+    task.vision_json = {"schema_version": 1, "name": "p", "tools": []}
+    subtask.vision_output = "t1.diameter"
+    await db_session.commit()
+    worker_says({"t1.diameter": float(subtask.nominal)})
+
+    response = await client.post(
+        "/api/vision/execute",
+        headers=auth_headers(measurement_tec_user),
+        files={"image": ("p.png", b"i am not an image", "image/png")},
+        data={"task_id": str(task.id)},
+    )
+
+    assert response.status_code == 422
+    assert (await db_session.execute(select(VisionResult))).scalars().all() == []
+    assert (await db_session.execute(select(Measurement))).scalars().all() == []
+
+
 @pytest.mark.asyncio
 async def test_execute_rejects_an_oversized_upload(
     client, db_session, measurement_tec_user, worker_says, monkeypatch,