sophhub 0.4.19 → 0.4.21

package/skills/sophnet-id-photo/src/scripts/id_photo_face_crop.py

@@ -1,313 +1,313 @@
-"""
-人脸检测与证件照导向裁切：调用 Sophnet detect_and_embed，将输入落地为本地文件后取置信度达标的人脸中**检测框面积最大**者，
-按目标输出尺寸的宽高比裁切头肩区域，再供图改图使用。多张达标人脸时不拒审，仅按最大脸裁切。
-"""
-
-from __future__ import annotations
-
-import base64
-import os
-import re
-import sys
-import tempfile
-from typing import Any
-
-import cv2
-import requests
-
-SOPH_BASE_URL = os.environ.get("SOPH_BASE_URL", "https://www.sophnet.com").rstrip("/")
-FACE_API_URL = f"{SOPH_BASE_URL}/api/open-apis/projects/detect_and_embed"
-
-
-def parse_output_size(size_str: str) -> tuple[int, int]:
-    m = re.match(r"^(\d+)\s*\*\s*(\d+)$", (size_str or "").strip())
-    if not m:
-        raise ValueError(f"invalid size: {size_str!r}")
-    return int(m.group(1)), int(m.group(2))
-
-
-def _decode_data_uri(data_uri: str) -> bytes:
-    if not data_uri.startswith("data:") or ";base64," not in data_uri:
-        raise ValueError("not a data URI")
-    _meta, b64 = data_uri.split(";base64,", 1)
-    return base64.b64decode(b64)
-
-
-def materialize_image(ref: str, api_key: str, download_timeout: int = 120) -> tuple[str, list[str]]:
-    """
-    将输入落地为本地路径。返回 (local_path, cleanup_paths)；cleanup_paths 为需事后删除的临时文件
-    （不含用户本机原始路径）。
-    """
-    raw = ref.lstrip("@").strip()
-    cleanup: list[str] = []
-
-    if raw.startswith(("http://", "https://")):
-        headers = {"Authorization": f"Bearer {api_key}"}
-        resp = requests.get(raw, headers=headers, timeout=download_timeout, stream=True)
-        resp.raise_for_status()
-        ext = os.path.splitext(raw.split("?")[0])[-1].lower() or ".jpg"
-        if ext not in (".jpg", ".jpeg", ".png", ".webp", ".bmp", ".gif"):
-            ext = ".jpg"
-        fd, tmp = tempfile.mkstemp(suffix=ext, prefix="id_photo_src_")
-        os.close(fd)
-        with open(tmp, "wb") as f:
-            for chunk in resp.iter_content(8192):
-                if chunk:
-                    f.write(chunk)
-        cleanup.append(tmp)
-        return tmp, cleanup
-
-    if raw.startswith("data:image/"):
-        blob = _decode_data_uri(raw)
-        fd, tmp = tempfile.mkstemp(suffix=".jpg", prefix="id_photo_src_")
-        os.close(fd)
-        with open(tmp, "wb") as f:
-            f.write(blob)
-        cleanup.append(tmp)
-        return tmp, cleanup
-
-    if os.path.isfile(raw):
-        return os.path.realpath(raw), cleanup
-
-    raise FileNotFoundError(raw)
-
-
-def detect_faces(image_path: str, api_key: str, timeout: int = 30) -> dict[str, Any]:
-    """调用 Sophnet detect_and_embed，成功时返回含 output 列表的字典。"""
-    if not os.path.isfile(image_path):
-        raise FileNotFoundError(image_path)
-
-    ext = os.path.splitext(image_path)[1].lower()
-    mime_types = {
-        ".jpg": "image/jpeg",
-        ".jpeg": "image/jpeg",
-        ".png": "image/png",
-        ".webp": "image/webp",
-        ".bmp": "image/bmp",
-        ".gif": "image/gif",
-    }
-    mime_type = mime_types.get(ext, "image/jpeg")
-
-    with open(image_path, "rb") as f:
-        files = {"file": (os.path.basename(image_path), f, mime_type)}
-        headers = {"Authorization": f"Bearer {api_key}"}
-        resp = requests.post(FACE_API_URL, files=files, headers=headers, timeout=timeout)
-
-    if resp.status_code != 200:
-        raise RuntimeError(f"detect_and_embed HTTP {resp.status_code}: {resp.text[:300]}")
-
-    body = resp.json()
-    return body.get("result") or {}
-
-
-def get_largest_face(faces: list[dict[str, Any]]) -> dict[str, Any] | None:
-    """在列表中取检测框面积 (x2-x1)*(y2-y1) 最大的人脸。"""
-    if not faces:
-        return None
-    best: dict[str, Any] | None = None
-    best_area = 0.0
-    for face in faces:
-        box = face.get("box") or []
-        if len(box) < 4:
-            continue
-        x1, y1, x2, y2 = (float(box[0]), float(box[1]), float(box[2]), float(box[3]))
-        area = max(0.0, x2 - x1) * max(0.0, y2 - y1)
-        if area > best_area:
-            best_area = area
-            best = face
-    return best
-
-
-def compute_id_photo_crop(
-    img_w: int,
-    img_h: int,
-    x1: float,
-    y1: float,
-    x2: float,
-    y2: float,
-    out_w: int,
-    out_h: int,
-) -> tuple[int, int, int, int] | None:
-    """按目标输出宽高比，围绕人脸得到 (left, top, right, bottom)，用于 numpy 切片 [top:bottom, left:right]。"""
-    x1 = max(0.0, min(float(img_w), float(x1)))
-    y1 = max(0.0, min(float(img_h), float(y1)))
-    x2 = max(0.0, min(float(img_w), float(x2)))
-    y2 = max(0.0, min(float(img_h), float(y2)))
-    if x2 <= x1 or y2 <= y1:
-        return None
-
-    fh = y2 - y1
-    cx = (x1 + x2) * 0.5
-    aspect = out_w / float(out_h)
-
-    margin_top = 0.58 * fh
-    margin_bot = 1.08 * fh
-    top_ref = y1 - margin_top
-    bot_ref = y2 + margin_bot
-    ref_h = bot_ref - top_ref
-    ref_h = max(ref_h, fh * 2.15)
-    ref_h = min(float(img_h), ref_h)
-
-    crop_h = ref_h
-    crop_w = crop_h * aspect
-    if crop_w > img_w:
-        crop_w = float(img_w)
-        crop_h = crop_w / aspect
-    if crop_h > img_h:
-        crop_h = float(img_h)
-        crop_w = crop_h * aspect
-
-    left = cx - crop_w * 0.5
-    left = max(0.0, min(left, float(img_w) - crop_w))
-
-    # 头位略靠上：脸顶约在裁切框高度 30% 处
-    top = y1 - 0.30 * crop_h
-    top = max(0.0, min(top, float(img_h) - crop_h))
-
-    crop_w_i = int(round(crop_w))
-    crop_h_i = int(round(crop_h))
-    left_i = int(round(left))
-    top_i = int(round(top))
-    right_i = left_i + crop_w_i
-    bottom_i = top_i + crop_h_i
-
-    if right_i > img_w:
-        right_i = img_w
-        left_i = max(0, right_i - crop_w_i)
-    if bottom_i > img_h:
-        bottom_i = img_h
-        top_i = max(0, bottom_i - crop_h_i)
-
-    if right_i - left_i < 32 or bottom_i - top_i < 32:
-        return None
-    return left_i, top_i, right_i, bottom_i
-
-
-def prepare_for_id_photo_edit(
-    image_ref: str,
-    api_key: str,
-    size_str: str,
-    *,
-    skip_face_crop: bool = False,
-    strict_face_gate: bool = False,
-    det_thr: float = 0.5,
-    detect_timeout: int = 30,
-    download_timeout: int = 120,
-) -> dict[str, Any]:
-    """
-    返回:
-      edit_ref — 送入图改图解析器的路径或 URL/data URI
-      cleanup — 事后应删除的临时文件列表
-      face_crop — applied | skipped_disabled | skipped_no_face | failed_* | rejected_no_face
-      error — 可选：no_face（strict_face_gate 且零张达标人脸）
-      face_count — 可选：仅 rejected_no_face 时为 0；成功 applied 时可有 face_detected_count
-      face_detected_count — 可选：置信度达标的人脸数（applied 时≥1）
-    """
-    ref = image_ref.lstrip("@")
-    if skip_face_crop:
-        return {"edit_ref": ref, "cleanup": [], "face_crop": "skipped_disabled"}
-
-    try:
-        tw, th = parse_output_size(size_str)
-    except ValueError as e:
-        print(f"Warning: {e}，将不进行人脸裁切。", file=sys.stderr)
-        return {"edit_ref": ref, "cleanup": [], "face_crop": "failed_bad_size"}
-
-    try:
-        local_path, material_cleanup = materialize_image(ref, api_key, download_timeout=download_timeout)
-    except Exception as e:
-        print(f"Warning: 无法落地输入以便人脸检测与裁切，改用原始输入: {e}", file=sys.stderr)
-        return {"edit_ref": ref, "cleanup": [], "face_crop": "failed_materialize"}
-
-    try:
-        api_result = detect_faces(local_path, api_key, timeout=detect_timeout)
-    except Exception as e:
-        print(f"Warning: 人脸检测接口失败，改用未裁切落地图: {e}", file=sys.stderr)
-        if material_cleanup:
-            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "failed_detect"}
-        return {"edit_ref": local_path, "cleanup": [], "face_crop": "failed_detect"}
-
-    faces = api_result.get("output") or []
-    faces = [f for f in faces if float(f.get("det_score", 0) or 0) >= det_thr]
-
-    def _cleanup_material_only():
-        for p in material_cleanup:
-            try:
-                if p and os.path.isfile(p):
-                    os.unlink(p)
-            except OSError:
-                pass
-
-    if strict_face_gate:
-        if len(faces) == 0:
-            _cleanup_material_only()
-            return {
-                "error": "no_face",
-                "edit_ref": ref,
-                "cleanup": [],
-                "face_crop": "rejected_no_face",
-                "face_count": 0,
-            }
-
-    if len(faces) > 1:
-        print(
-            f"Warning: 检出 {len(faces)} 张达标人脸，按检测框面积最大者裁切。",
-            file=sys.stderr,
-        )
-
-    face = get_largest_face(faces)
-
-    image = cv2.imread(local_path)
-    if image is None:
-        print("Warning: OpenCV 无法读取落地图，改用原始输入引用", file=sys.stderr)
-        for p in material_cleanup:
-            try:
-                if os.path.isfile(p):
-                    os.unlink(p)
-            except OSError:
-                pass
-        return {"edit_ref": ref, "cleanup": [], "face_crop": "failed_read"}
-
-    ih, iw = image.shape[:2]
-
-    if not face:
-        print("Warning: 未检出足够置信度的人脸，整图送图改图", file=sys.stderr)
-        if material_cleanup:
-            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "skipped_no_face"}
-        return {"edit_ref": local_path, "cleanup": [], "face_crop": "skipped_no_face"}
-
-    box = face.get("box") or []
-    if len(box) < 4:
-        if material_cleanup:
-            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "skipped_no_face"}
-        return {"edit_ref": local_path, "cleanup": [], "face_crop": "skipped_no_face"}
-
-    rect = compute_id_photo_crop(iw, ih, float(box[0]), float(box[1]), float(box[2]), float(box[3]), tw, th)
-    if not rect:
-        print("Warning: 无法计算裁切框，使用整图送图改图", file=sys.stderr)
-        if material_cleanup:
-            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "failed_crop_math"}
-        return {"edit_ref": local_path, "cleanup": [], "face_crop": "failed_crop_math"}
-
-    l, t, r, b = rect
-    cropped = image[t:b, l:r]
-    fd, out_path = tempfile.mkstemp(suffix=".jpg", prefix="id_photo_crop_")
-    os.close(fd)
-    ok = cv2.imwrite(out_path, cropped, [int(cv2.IMWRITE_JPEG_QUALITY), 95])
-    if not ok:
-        print("Warning: 裁切图写入失败，使用未裁切落地图", file=sys.stderr)
-        try:
-            os.unlink(out_path)
-        except OSError:
-            pass
-        if material_cleanup:
-            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "failed_write"}
-        return {"edit_ref": local_path, "cleanup": [], "face_crop": "failed_write"}
-
-    cleanup_all = list(material_cleanup) + [out_path]
-    return {
-        "edit_ref": out_path,
-        "cleanup": cleanup_all,
-        "face_crop": "applied",
-        "face_detected_count": len(faces),
-    }
+"""
+人脸检测与证件照导向裁切：调用 Sophnet detect_and_embed，将输入落地为本地文件后取置信度达标的人脸中**检测框面积最大**者，
+按目标输出尺寸的宽高比裁切头肩区域，再供图改图使用。多张达标人脸时不拒审，仅按最大脸裁切。
+"""
+
+from __future__ import annotations
+
+import base64
+import os
+import re
+import sys
+import tempfile
+from typing import Any
+
+import cv2
+import requests
+
+SOPH_BASE_URL = os.environ.get("SOPH_BASE_URL", "https://www.sophnet.com").rstrip("/")
+FACE_API_URL = f"{SOPH_BASE_URL}/api/open-apis/projects/detect_and_embed"
+
+
+def parse_output_size(size_str: str) -> tuple[int, int]:
+    m = re.match(r"^(\d+)\s*\*\s*(\d+)$", (size_str or "").strip())
+    if not m:
+        raise ValueError(f"invalid size: {size_str!r}")
+    return int(m.group(1)), int(m.group(2))
+
+
+def _decode_data_uri(data_uri: str) -> bytes:
+    if not data_uri.startswith("data:") or ";base64," not in data_uri:
+        raise ValueError("not a data URI")
+    _meta, b64 = data_uri.split(";base64,", 1)
+    return base64.b64decode(b64)
+
+
+def materialize_image(ref: str, api_key: str, download_timeout: int = 120) -> tuple[str, list[str]]:
+    """
+    将输入落地为本地路径。返回 (local_path, cleanup_paths)；cleanup_paths 为需事后删除的临时文件
+    （不含用户本机原始路径）。
+    """
+    raw = ref.lstrip("@").strip()
+    cleanup: list[str] = []
+
+    if raw.startswith(("http://", "https://")):
+        headers = {"Authorization": f"Bearer {api_key}"}
+        resp = requests.get(raw, headers=headers, timeout=download_timeout, stream=True)
+        resp.raise_for_status()
+        ext = os.path.splitext(raw.split("?")[0])[-1].lower() or ".jpg"
+        if ext not in (".jpg", ".jpeg", ".png", ".webp", ".bmp", ".gif"):
+            ext = ".jpg"
+        fd, tmp = tempfile.mkstemp(suffix=ext, prefix="id_photo_src_")
+        os.close(fd)
+        with open(tmp, "wb") as f:
+            for chunk in resp.iter_content(8192):
+                if chunk:
+                    f.write(chunk)
+        cleanup.append(tmp)
+        return tmp, cleanup
+
+    if raw.startswith("data:image/"):
+        blob = _decode_data_uri(raw)
+        fd, tmp = tempfile.mkstemp(suffix=".jpg", prefix="id_photo_src_")
+        os.close(fd)
+        with open(tmp, "wb") as f:
+            f.write(blob)
+        cleanup.append(tmp)
+        return tmp, cleanup
+
+    if os.path.isfile(raw):
+        return os.path.realpath(raw), cleanup
+
+    raise FileNotFoundError(raw)
+
+
+def detect_faces(image_path: str, api_key: str, timeout: int = 30) -> dict[str, Any]:
+    """调用 Sophnet detect_and_embed，成功时返回含 output 列表的字典。"""
+    if not os.path.isfile(image_path):
+        raise FileNotFoundError(image_path)
+
+    ext = os.path.splitext(image_path)[1].lower()
+    mime_types = {
+        ".jpg": "image/jpeg",
+        ".jpeg": "image/jpeg",
+        ".png": "image/png",
+        ".webp": "image/webp",
+        ".bmp": "image/bmp",
+        ".gif": "image/gif",
+    }
+    mime_type = mime_types.get(ext, "image/jpeg")
+
+    with open(image_path, "rb") as f:
+        files = {"file": (os.path.basename(image_path), f, mime_type)}
+        headers = {"Authorization": f"Bearer {api_key}"}
+        resp = requests.post(FACE_API_URL, files=files, headers=headers, timeout=timeout)
+
+    if resp.status_code != 200:
+        raise RuntimeError(f"detect_and_embed HTTP {resp.status_code}: {resp.text[:300]}")
+
+    body = resp.json()
+    return body.get("result") or {}
+
+
+def get_largest_face(faces: list[dict[str, Any]]) -> dict[str, Any] | None:
+    """在列表中取检测框面积 (x2-x1)*(y2-y1) 最大的人脸。"""
+    if not faces:
+        return None
+    best: dict[str, Any] | None = None
+    best_area = 0.0
+    for face in faces:
+        box = face.get("box") or []
+        if len(box) < 4:
+            continue
+        x1, y1, x2, y2 = (float(box[0]), float(box[1]), float(box[2]), float(box[3]))
+        area = max(0.0, x2 - x1) * max(0.0, y2 - y1)
+        if area > best_area:
+            best_area = area
+            best = face
+    return best
+
+
+def compute_id_photo_crop(
+    img_w: int,
+    img_h: int,
+    x1: float,
+    y1: float,
+    x2: float,
+    y2: float,
+    out_w: int,
+    out_h: int,
+) -> tuple[int, int, int, int] | None:
+    """按目标输出宽高比，围绕人脸得到 (left, top, right, bottom)，用于 numpy 切片 [top:bottom, left:right]。"""
+    x1 = max(0.0, min(float(img_w), float(x1)))
+    y1 = max(0.0, min(float(img_h), float(y1)))
+    x2 = max(0.0, min(float(img_w), float(x2)))
+    y2 = max(0.0, min(float(img_h), float(y2)))
+    if x2 <= x1 or y2 <= y1:
+        return None
+
+    fh = y2 - y1
+    cx = (x1 + x2) * 0.5
+    aspect = out_w / float(out_h)
+
+    margin_top = 0.58 * fh
+    margin_bot = 1.08 * fh
+    top_ref = y1 - margin_top
+    bot_ref = y2 + margin_bot
+    ref_h = bot_ref - top_ref
+    ref_h = max(ref_h, fh * 2.15)
+    ref_h = min(float(img_h), ref_h)
+
+    crop_h = ref_h
+    crop_w = crop_h * aspect
+    if crop_w > img_w:
+        crop_w = float(img_w)
+        crop_h = crop_w / aspect
+    if crop_h > img_h:
+        crop_h = float(img_h)
+        crop_w = crop_h * aspect
+
+    left = cx - crop_w * 0.5
+    left = max(0.0, min(left, float(img_w) - crop_w))
+
+    # 头位略靠上：脸顶约在裁切框高度 30% 处
+    top = y1 - 0.30 * crop_h
+    top = max(0.0, min(top, float(img_h) - crop_h))
+
+    crop_w_i = int(round(crop_w))
+    crop_h_i = int(round(crop_h))
+    left_i = int(round(left))
+    top_i = int(round(top))
+    right_i = left_i + crop_w_i
+    bottom_i = top_i + crop_h_i
+
+    if right_i > img_w:
+        right_i = img_w
+        left_i = max(0, right_i - crop_w_i)
+    if bottom_i > img_h:
+        bottom_i = img_h
+        top_i = max(0, bottom_i - crop_h_i)
+
+    if right_i - left_i < 32 or bottom_i - top_i < 32:
+        return None
+    return left_i, top_i, right_i, bottom_i
+
+
+def prepare_for_id_photo_edit(
+    image_ref: str,
+    api_key: str,
+    size_str: str,
+    *,
+    skip_face_crop: bool = False,
+    strict_face_gate: bool = False,
+    det_thr: float = 0.5,
+    detect_timeout: int = 30,
+    download_timeout: int = 120,
+) -> dict[str, Any]:
+    """
+    返回:
+      edit_ref — 送入图改图解析器的路径或 URL/data URI
+      cleanup — 事后应删除的临时文件列表
+      face_crop — applied | skipped_disabled | skipped_no_face | failed_* | rejected_no_face
+      error — 可选：no_face（strict_face_gate 且零张达标人脸）
+      face_count — 可选：仅 rejected_no_face 时为 0；成功 applied 时可有 face_detected_count
+      face_detected_count — 可选：置信度达标的人脸数（applied 时≥1）
+    """
+    ref = image_ref.lstrip("@")
+    if skip_face_crop:
+        return {"edit_ref": ref, "cleanup": [], "face_crop": "skipped_disabled"}
+
+    try:
+        tw, th = parse_output_size(size_str)
+    except ValueError as e:
+        print(f"Warning: {e}，将不进行人脸裁切。", file=sys.stderr)
+        return {"edit_ref": ref, "cleanup": [], "face_crop": "failed_bad_size"}
+
+    try:
+        local_path, material_cleanup = materialize_image(ref, api_key, download_timeout=download_timeout)
+    except Exception as e:
+        print(f"Warning: 无法落地输入以便人脸检测与裁切，改用原始输入: {e}", file=sys.stderr)
+        return {"edit_ref": ref, "cleanup": [], "face_crop": "failed_materialize"}
+
+    try:
+        api_result = detect_faces(local_path, api_key, timeout=detect_timeout)
+    except Exception as e:
+        print(f"Warning: 人脸检测接口失败，改用未裁切落地图: {e}", file=sys.stderr)
+        if material_cleanup:
+            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "failed_detect"}
+        return {"edit_ref": local_path, "cleanup": [], "face_crop": "failed_detect"}
+
+    faces = api_result.get("output") or []
+    faces = [f for f in faces if float(f.get("det_score", 0) or 0) >= det_thr]
+
+    def _cleanup_material_only():
+        for p in material_cleanup:
+            try:
+                if p and os.path.isfile(p):
+                    os.unlink(p)
+            except OSError:
+                pass
+
+    if strict_face_gate:
+        if len(faces) == 0:
+            _cleanup_material_only()
+            return {
+                "error": "no_face",
+                "edit_ref": ref,
+                "cleanup": [],
+                "face_crop": "rejected_no_face",
+                "face_count": 0,
+            }
+
+    if len(faces) > 1:
+        print(
+            f"Warning: 检出 {len(faces)} 张达标人脸，按检测框面积最大者裁切。",
+            file=sys.stderr,
+        )
+
+    face = get_largest_face(faces)
+
+    image = cv2.imread(local_path)
+    if image is None:
+        print("Warning: OpenCV 无法读取落地图，改用原始输入引用", file=sys.stderr)
+        for p in material_cleanup:
+            try:
+                if os.path.isfile(p):
+                    os.unlink(p)
+            except OSError:
+                pass
+        return {"edit_ref": ref, "cleanup": [], "face_crop": "failed_read"}
+
+    ih, iw = image.shape[:2]
+
+    if not face:
+        print("Warning: 未检出足够置信度的人脸，整图送图改图", file=sys.stderr)
+        if material_cleanup:
+            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "skipped_no_face"}
+        return {"edit_ref": local_path, "cleanup": [], "face_crop": "skipped_no_face"}
+
+    box = face.get("box") or []
+    if len(box) < 4:
+        if material_cleanup:
+            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "skipped_no_face"}
+        return {"edit_ref": local_path, "cleanup": [], "face_crop": "skipped_no_face"}
+
+    rect = compute_id_photo_crop(iw, ih, float(box[0]), float(box[1]), float(box[2]), float(box[3]), tw, th)
+    if not rect:
+        print("Warning: 无法计算裁切框，使用整图送图改图", file=sys.stderr)
+        if material_cleanup:
+            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "failed_crop_math"}
+        return {"edit_ref": local_path, "cleanup": [], "face_crop": "failed_crop_math"}
+
+    l, t, r, b = rect
+    cropped = image[t:b, l:r]
+    fd, out_path = tempfile.mkstemp(suffix=".jpg", prefix="id_photo_crop_")
+    os.close(fd)
+    ok = cv2.imwrite(out_path, cropped, [int(cv2.IMWRITE_JPEG_QUALITY), 95])
+    if not ok:
+        print("Warning: 裁切图写入失败，使用未裁切落地图", file=sys.stderr)
+        try:
+            os.unlink(out_path)
+        except OSError:
+            pass
+        if material_cleanup:
+            return {"edit_ref": local_path, "cleanup": list(material_cleanup), "face_crop": "failed_write"}
+        return {"edit_ref": local_path, "cleanup": [], "face_crop": "failed_write"}
+
+    cleanup_all = list(material_cleanup) + [out_path]
+    return {
+        "edit_ref": out_path,
+        "cleanup": cleanup_all,
+        "face_crop": "applied",
+        "face_detected_count": len(faces),
+    }

package/skills/sophnet-image-edit/skill.json

@@ -1,20 +1,20 @@
-{
-  "name": "sophnet-image-edit",
-  "version": "1.0.0",
-  "types": [
-    "builtin"
-  ],
-  "displayName": "",
-  "description": "",
-  "changelog": [
-    {
-      "version": "1.0.0",
-      "date": "2026-04-09",
-      "changes": [
-        "初次提交"
-      ]
-    }
-  ],
-  "createdAt": "2026-04-09",
-  "updatedAt": "2026-04-09"
-}
+{
+  "name": "sophnet-image-edit",
+  "version": "1.0.0",
+  "types": [
+    "builtin"
+  ],
+  "displayName": "",
+  "description": "",
+  "changelog": [
+    {
+      "version": "1.0.0",
+      "date": "2026-04-09",
+      "changes": [
+        "初次提交"
+      ]
+    }
+  ],
+  "createdAt": "2026-04-09",
+  "updatedAt": "2026-04-09"
+}