Files
agent/skills/image-process/run.py
T
kicer a52b227d8b refactor(skills): 本地安装改成「README 式说明」,install.sh/requirements.txt 归服务器
按三条原则收紧:
1) 线上零影响:office/install.sh 回退成原来的 linux-only 版本(本地本来就不该跑它);
   image-process 的字体候选表撤回(那是在代码里写环境兼容,方向错了)。
2) install.sh / requirements.txt 只做服务器工件:镜像构建与 agent-executor-init 读它们,
   行为一字未变。
3) 本机缺东西时由 .md 指路,模型自己去装:四个技能各加 INSTALL.md(具体命令 + 平台
   资产/字体表 + 验证方式),并在 SKILL.md 里各加一行指向它(skill get 就能看到)。
   报错文案带上 INSTALL.md 的绝对路径,所以「撞到缺依赖」的那一刻就能拿到说明书。

保留的两处代码钩子都不是环境探测,而是「说明书里写的安装方式得真能生效」:
- office: start.sh 把 skills/office/scripts 塞进 PATH(本地没 root 写 /usr/local/bin);
  封装层 ENOENT 指向 INSTALL.md。
- x2pdf: find_soffice() 探 macOS app bundle(brew cask 装的 soffice 不在 PATH);
  WeasyPrint 不可用时回落 LibreOffice,让本地只需一个后端而不是 pango+WeasyPrint 两条链。

INSTALL.md 里 officecli 指向 releases/latest(它会自更新,实测 1.0.157 跑一次变 1.0.160),
不要求与服务器 pin 对齐。

实测:office --version / office create / office view outline 正常;x2pdf 缺 markdown 与
image-process 缺字体两条报错都给出 INSTALL.md 绝对路径,按 font 参数即通过。
2026-11-11 17:34:30 +08:00

422 lines
15 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
image-process 技能主入口:用 Pillow 做确定性的图像处理(不含 AI 生成/语义修图)。
支持动作 action:
- info 查看图片信息(尺寸/模式/格式/体积)
- convert 格式转换(按 output 扩展名)
- resize 缩放(keep_ratio 默认保持宽高比)
- crop 裁剪(left/top/width/height)
- compress 压缩(quality;可选 max_dim 先缩长边)
- watermark 文字水印(text;中文需系统字体或 font 参数)
- slice 等分切片(rows/cols,默认 3×3,输出到独立目录)
- grid 多图拼图 contact sheet(input 为图片所在目录)
约定:输出绝不覆盖输入原图;相对路径由 Skill Manager 预先转成绝对路径。
"""
import json
import sys
from pathlib import Path
from PIL import Image, ImageDraw, ImageFont, ImageOps
IMAGE_EXTS = {'.jpg', '.jpeg', '.png', '.webp', '.bmp', '.gif', '.tif', '.tiff'}
POSITIONS = {
'tl': ('l', 't'), 'tc': ('c', 't'), 'tr': ('r', 't'),
'ml': ('l', 'm'), 'mc': ('c', 'm'), 'mr': ('r', 'm'),
'bl': ('l', 'b'), 'bc': ('c', 'b'), 'br': ('r', 'b'),
}
NAMED_COLORS = {
'white': (255, 255, 255), 'black': (0, 0, 0), 'red': (255, 0, 0),
'green': (0, 128, 0), 'blue': (0, 0, 255), 'yellow': (255, 255, 0),
'gray': (128, 128, 128), 'grey': (128, 128, 128), 'orange': (255, 165, 0),
'purple': (128, 0, 128), 'cyan': (0, 255, 255),
}
CJK_FONT_CANDIDATES = [
'/usr/share/fonts/truetype/wqy/wqy-zenhei.ttc',
'/usr/share/fonts/truetype/wqy/wqy-microhei.ttc',
]
# Noto 只作最后兑底:Pillow(FreeType) 能读它的 CFF 轮廓,但 ttc 默认字型(索引 0)是 JP 字形,
# 中文水印会出日式写法;因此系统文泉驿(CJK_FONT_CANDIDATES)排在它前面。
NOTO_CJK_FALLBACK = '/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc'
def to_int(v, default=0):
if v is None or v == '':
return default
try:
return int(v)
except (TypeError, ValueError):
raise ValueError('参数应为整数:%r' % (v,))
def to_bool(v, default=True):
if v is None or v == '':
return default
if isinstance(v, bool):
return v
if isinstance(v, str):
return v.strip().lower() in ('1', 'true', 'yes', 'y', 'on')
return bool(v)
def parse_color(s, default=(255, 255, 255)):
s = (s or '').strip().lower()
if not s:
return default
if s in NAMED_COLORS:
return NAMED_COLORS[s]
if s.startswith('#'):
h = s[1:]
if len(h) == 3:
h = ''.join(c * 2 for c in h)
if len(h) == 6:
try:
return tuple(int(h[i:i + 2], 16) for i in (0, 2, 4))
except ValueError:
pass
raise ValueError('无法识别的颜色:%s(支持 white/black/red/... 或 #RRGGBB)' % s)
def open_image(path):
"""打开图片并按 EXIF 方向摆正;返回 (img, 原始格式名)。"""
img = Image.open(str(path))
fmt = img.format or ''
img.load()
img = ImageOps.exif_transpose(img)
return img, fmt
def ensure_rgb(img):
if img.mode in ('RGBA', 'LA', 'P'):
return img.convert('RGB')
return img
def save_image(img, out_path, quality=None):
"""按输出扩展名保存;JPEG/WebP 用 quality,PNG 走 optimize。"""
p = Path(out_path)
p.parent.mkdir(parents=True, exist_ok=True)
fmt = (p.suffix or '').lstrip('.').upper()
q = to_int(quality, 90)
q = max(1, min(95, q))
if fmt in ('JPG', 'JPEG'):
ensure_rgb(img).save(str(p), format='JPEG', quality=q, optimize=True)
elif fmt == 'WEBP':
img.save(str(p), format='WEBP', quality=q)
elif fmt == 'PNG':
img.save(str(p), format='PNG', optimize=True)
elif fmt in ('BMP', 'GIF', 'TIFF', 'TIF', 'ICO'):
img.save(str(p))
else:
raise ValueError('不支持的输出扩展名:%s(支持 jpg/jpeg/png/webp/bmp/gif/tiff)'
% (p.suffix or '(无)'))
def default_out(input_path, tag):
"""缺省输出:同目录 <原名>_<tag><原扩展名>,不覆盖原图。"""
p = Path(input_path)
return str(p.parent / ('%s_%s%s' % (p.stem, tag, p.suffix if p.suffix else '.png')))
def ensure_different(src, dst):
if Path(src).resolve() == Path(dst).resolve():
raise ValueError('输出路径与输入相同:%s,不允许覆盖原图,请指定新的 output' % src)
def resample():
return getattr(Image, 'Resampling', Image).LANCZOS
# ---------- 各动作实现 ----------
def action_info(img, fmt, args):
w, h = img.size
lines = ['图片信息:']
lines.append('- 文件:%s' % args['_input'])
lines.append('- 尺寸:%dx%d 像素' % (w, h))
lines.append('- 色彩模式:%s' % img.mode)
lines.append('- 格式:%s' % (fmt or '未知'))
try:
kb = Path(args['_input']).stat().st_size / 1024.0
lines.append('- 文件大小:%.1f KB' % kb)
except OSError:
pass
return '\n'.join(lines)
def action_convert(img, fmt, args):
out = args.get('output')
if not out:
raise ValueError('convert 需要 output 参数(带目标扩展名,如 out.png)')
ensure_different(args['_input'], out)
save_image(img, out, quality=args.get('quality'))
return '格式转换完成:%s -> %s' % (args['_input'], out)
def action_resize(img, fmt, args):
w = to_int(args.get('width'))
h = to_int(args.get('height'))
keep = to_bool(args.get('keep_ratio'), True)
if not w and not h:
raise ValueError('resize 需要 width 或 height(至少一个)')
iw, ih = img.size
if keep:
if w and h:
scale = min(w / iw, h / ih)
nw, nh = max(1, round(iw * scale)), max(1, round(ih * scale))
elif w:
nw, nh = w, max(1, round(ih * (w / iw)))
else:
nw, nh = max(1, round(iw * (h / ih))), h
else:
if not (w and h):
raise ValueError('keep_ratio=false 时 width 与 height 都要提供')
nw, nh = w, h
img2 = img.resize((max(1, nw), max(1, nh)), resample())
out = args.get('output') or default_out(args['_input'], 'resized')
ensure_different(args['_input'], out)
save_image(img2, out, quality=args.get('quality'))
return '缩放完成:%dx%d -> %dx%d,输出 %s' % (iw, ih, img2.size[0], img2.size[1], out)
def action_crop(img, fmt, args):
iw, ih = img.size
left = to_int(args.get('left'))
top = to_int(args.get('top'))
width = to_int(args.get('width'), max(1, iw - left))
height = to_int(args.get('height'), max(1, ih - top))
if width < 1 or height < 1:
raise ValueError('crop 的 width/height 必须 >= 1')
left = max(0, left)
top = max(0, top)
right = min(iw, left + width)
bottom = min(ih, top + height)
if right <= left or bottom <= top:
raise ValueError('裁剪区域超出图片范围(图片 %dx%d)' % (iw, ih))
out = args.get('output') or default_out(args['_input'], 'cropped')
ensure_different(args['_input'], out)
save_image(img.crop((left, top, right, bottom)), out, quality=args.get('quality'))
return '裁剪完成:区域 (%d,%d)-(%d,%d),输出 %s' % (left, top, right, bottom, out)
def action_compress(img, fmt, args):
img2 = img
md = to_int(args.get('max_dim'))
if md > 0 and max(img2.size) > md:
img2 = ImageOps.contain(img2, (md, md), resample())
out = args.get('output') or default_out(args['_input'], 'compressed')
ensure_different(args['_input'], out)
q = to_int(args.get('quality'), 85)
try:
before = Path(args['_input']).stat().st_size
except OSError:
before = 0
save_image(img2, out, quality=q)
try:
after = Path(out).stat().st_size
except OSError:
after = 0
pct = (100.0 * after / before) if before else 0
return '压缩完成:%.1f KB -> %.1f KB(%.0f%%),输出 %s' % (before / 1024.0, after / 1024.0, pct, out)
def find_font(explicit):
"""字体查找:显式 font > 系统 wqy > Noto 兑底。"""
if explicit:
p = Path(explicit)
if not p.exists():
raise ValueError('指定的字体不存在:%s' % explicit)
return str(p)
candidates = list(CJK_FONT_CANDIDATES)
candidates.append(NOTO_CJK_FALLBACK)
for c in candidates:
if Path(c).exists():
return c
return None
def action_watermark(img, fmt, args):
text = (args.get('text') or '').strip()
if not text:
raise ValueError('watermark 需要 text 参数')
w, h = img.size
font_size = to_int(args.get('font_size'), max(14, w // 20))
font_size = max(8, min(400, font_size))
need_cjk = any(ord(ch) > 0x2E80 for ch in text)
font_path = find_font((args.get('font') or '').strip())
if font_path:
font = ImageFont.truetype(font_path, font_size)
elif need_cjk:
raise ValueError('中文水印缺少字体:传 font 参数(TTF/TTC 路径);各平台常见字体路径见 %s'
% (Path(__file__).resolve().parent / 'INSTALL.md'))
else:
try:
font = ImageFont.load_default(size=font_size)
except TypeError:
font = ImageFont.load_default()
pos = (args.get('position') or 'br').strip().lower()
if pos not in POSITIONS:
raise ValueError('position 必须是 tl/tc/tr/ml/mc/mr/bl/bc/br 之一')
align_x, align_y = POSITIONS[pos]
margin = max(0, to_int(args.get('margin'), 20))
color = parse_color(args.get('color'), (255, 255, 255))
opacity = max(0, min(255, to_int(args.get('opacity'), 200)))
# 在透明层上画字再合成,保证半透明水印且边缘平滑
layer = Image.new('RGBA', (w, h), (0, 0, 0, 0))
draw = ImageDraw.Draw(layer)
bbox = draw.textbbox((0, 0), text, font=font)
tw, th = bbox[2] - bbox[0], bbox[3] - bbox[1]
if align_x == 'l':
ox = margin - bbox[0]
elif align_x == 'c':
ox = (w - tw) / 2.0 - bbox[0]
else:
ox = w - margin - bbox[2]
if align_y == 't':
oy = margin - bbox[1]
elif align_y == 'm':
oy = (h - th) / 2.0 - bbox[1]
else:
oy = h - margin - bbox[3]
draw.text((ox, oy), text, font=font, fill=color + (opacity,))
base = img.convert('RGBA')
composed = Image.alpha_composite(base, layer)
out = args.get('output') or default_out(args['_input'], 'watermarked')
ensure_different(args['_input'], out)
save_image(composed, out, quality=args.get('quality'))
return '水印完成(位置 %s,字号 %d,不透明度 %d):%s' % (pos, font_size, opacity, out)
def action_slice(img, fmt, args):
w, h = img.size
rows = to_int(args.get('rows'))
cols = to_int(args.get('cols'))
if rows and not cols:
cols = rows
if cols and not rows:
rows = cols
rows = max(1, min(20, rows or 3))
cols = max(1, min(20, cols or 3))
src = Path(args['_input'])
out_dir = Path(args.get('output')) if args.get('output') else src.parent / (src.stem + '_slice')
out_dir.mkdir(parents=True, exist_ok=True)
ext = src.suffix.lower()
if ext not in IMAGE_EXTS:
ext = '.png'
cw = max(1, w // cols)
chh = max(1, h // rows)
files = []
for r in range(rows):
for c in range(cols):
box = (c * cw, r * chh, min(w, (c + 1) * cw), min(h, (r + 1) * chh))
tile = img.crop(box)
name = '%s_%02d_%02d%s' % (src.stem, r + 1, c + 1, ext)
save_image(tile, str(out_dir / name))
files.append(str(out_dir / name))
return '切片完成:%d×%d 共 %d 块,输出目录 %s\n%s' % (cols, rows, len(files), out_dir, '\n'.join(files))
def action_grid(args):
in_dir = Path(args.get('input'))
if not in_dir.is_dir():
raise ValueError('grid 的 input 应为图片所在目录:%s' % in_dir)
out = args.get('output')
if not out:
out = str(in_dir / ('%s_grid.jpg' % (in_dir.name or 'images')))
imgs = sorted(
p for p in in_dir.iterdir()
if p.is_file() and p.suffix.lower() in IMAGE_EXTS
and not p.name.startswith('.')
and Path(p).resolve() != Path(out).resolve()
)
if not imgs:
raise ValueError('目录中没有可拼图的图片:%s' % in_dir)
n = len(imgs)
cols = to_int(args.get('cols'))
if not cols:
cols = max(1, int(n ** 0.5 + 0.999)) # 缺省近似正方形网格
cols = max(1, min(20, cols))
rows = (n + cols - 1) // cols
cell = max(50, min(2048, to_int(args.get('cell'), 800)))
pad = max(0, to_int(args.get('padding'), 10))
bg = parse_color(args.get('background'), (255, 255, 255))
sheet = Image.new('RGB', (cols * cell + (cols + 1) * pad, rows * cell + (rows + 1) * pad), bg)
placed = []
for i, p in enumerate(imgs):
try:
im, _ = open_image(p)
except Exception as e:
print('跳过无法读取的图片:%s(%s)' % (p, e))
continue
thumb = im.copy()
thumb.thumbnail((cell, cell), resample())
thumb = ensure_rgb(thumb)
x = pad + (i % cols) * cell + (cell - thumb.size[0]) // 2
y = pad + (i // cols) * cell + (cell - thumb.size[1]) // 2
sheet.paste(thumb, (x, y))
im.close()
placed.append(p.name)
if not placed:
raise ValueError('目录中的图片都无法读取:%s' % in_dir)
save_image(sheet, out)
return '拼图完成:%d/%d 张 -> %s(网格 %d×%d,每格长边 %dpx)' % (len(placed), n, out, cols, rows, cell)
ACTIONS = {
'info': action_info,
'convert': action_convert,
'resize': action_resize,
'crop': action_crop,
'compress': action_compress,
'watermark': action_watermark,
'slice': action_slice,
'grid': action_grid,
}
def main():
if len(sys.argv) != 2:
print("Usage: python3 run.py '<json_string>'", file=sys.stderr)
sys.exit(1)
try:
args = json.loads(sys.argv[1])
except json.JSONDecodeError as e:
print('参数不是合法 JSON:%s' % e, file=sys.stderr)
sys.exit(1)
action = (args.get('action') or '').strip().lower()
if not action:
print('Error: 缺少 action(可选:%s)' % '/'.join(sorted(ACTIONS)), file=sys.stderr)
sys.exit(1)
if action not in ACTIONS:
print('Error: 不支持的 action:%s(可选:%s)' % (action, '/'.join(sorted(ACTIONS))), file=sys.stderr)
sys.exit(1)
try:
if action == 'grid':
text = action_grid(args)
else:
input_path = (args.get('input') or '').strip()
if not input_path:
raise ValueError('缺少 input 参数')
if not Path(input_path).exists():
raise ValueError('输入文件不存在:%s' % input_path)
img, fmt = open_image(input_path)
try:
args['_input'] = input_path
text = ACTIONS[action](img, fmt, args)
finally:
img.close()
print(text)
except Exception as e:
print('Error: %s' % e, file=sys.stderr)
sys.exit(1)
if __name__ == '__main__':
main()