ZJU-REAL/Easel / skills/builtin/skill-wechat-publisher
skill-wechat-publisher
Verified微信公众号文章自动创作与发布工具。给定参考文章、文字或文档,自动搜索整理全网相关信息、生成图文并茂的公众号文章,并发布到微信公众号草稿箱。特别强调反 AI 检测写作。 触发场景(沾边就用):用户提到"公众号 / 微信文章 / 推文 / 公号 / 发文 / mp / 草稿箱 / 群发",或要求把话题/文档/笔记写成公众号文章并发布到微信。
scripts/newspic_build.py
View on GitHub#!/usr/bin/env python3
"""
贴图(图片消息 / newspic)拆卡 + 出图计划生成器
输入: 一个 brief.md,含 YAML frontmatter + 要点列表 + (可选)短文本
输出: 一份 card_plan.json,列出每张卡的主副文字 + 生图 prompt + 目标文件名
本脚本**不直接生成图片**。生图由主流程 Claude 调项目内置 `scripts/generate_image.py`
按 card_plan.json 的 prompts 批量执行,保存到 `<slug>/images/`。
典型流程:
# 1. 人类写 brief.md(话题+要点+短文本)
# 2. 拆卡 + 生成 prompts + 短文本
python3 newspic_build.py brief.md
# 3. Claude 按 card_plan.json 逐张生图,保存到 images/01.png 起
# 4. Claude 跑 AI 味检测(newspic 模式)
python3 ai_score.py article.md --mode newspic
# 5. 发布
python3 publish.py --account main --type newspic --brief brief.md
brief.md 格式示例:
---
topic: "Claude Code /rewind 命令"
image_style: marker-lime # 可选,不写就用账号默认或 hand-drawn-blue
card_count: 6 # 可选,不写就跟要点数走
title: "Claude Code 里,最有用的命令之一"
account: main
---
# 要点
1. /rewind 厉害的地方不是"撤销一下",而是给你一个更对的工作流
2. 你可以输入 /rewind,也可以连续按两次 Esc,快速回滚代码
3. AI 解决不好问题,常常不是因为它不够会写,而是你不敢让它放手试
4. /rewind 的价值,就是把"试错"这件事真正变得可控
# 短文本
(可选,不写由 Claude 根据"要点"生成 100-300 字的短描述)
-------
prompt 模板中可用的变量占位符(由 build_card_plan 替换):
{topic} — brief.md 的 topic 字段
{card_main} — 当前卡片主文字(要点首段)
{card_sub} — 当前卡片副文字(要点后段,可能为空)
{point_full} — 完整原始要点(未切分),信息图类模板用它拿到更多上下文
{card_index} — 当前卡片序号,两位数字如 "01"
{card_total} — 卡片总数,两位数字如 "06"
{image_subject} — "topic - card_main" 拼接,兼容老风格
marker-* / data-chart 等风格只用 card_main/card_sub,
infographic-blue / -warm / -dark / -mint 新风格用 point_full + card_index 做高密度信息图。
"""
from __future__ import annotations
import argparse
import json
import re
import sys
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple
sys.path.insert(0, str(Path(__file__).parent))
try:
import yaml
except ImportError as exc:
raise ImportError(
"缺少 pyyaml 依赖。请执行: pip install pyyaml --break-system-packages"
) from exc
from config import (
ConfigError,
DEFAULT_IMAGE_STYLE,
get_image_style,
resolve_image_style,
)
# ============================================================
# brief.md 解析
# ============================================================
FRONTMATTER_RE = re.compile(r"^\s*---\s*\n(.*?)\n---\s*\n", re.DOTALL)
def parse_brief(md_text: str) -> Dict[str, Any]:
"""
把 brief.md 拆成:
{
"frontmatter": dict,
"sections": {"要点": [str, ...], "短文本": str, ...}
}
"""
m = FRONTMATTER_RE.match(md_text)
if not m:
raise ValueError(
"brief.md 必须以 YAML frontmatter 开头(--- ... --- 包围),请补上"
)
try:
frontmatter = yaml.safe_load(m.group(1)) or {}
except yaml.YAMLError as e:
raise ValueError(f"brief.md frontmatter 解析失败: {e}") from e
body = md_text[m.end():]
# 按一级标题切分 body,收集每个 section 的文字
sections: Dict[str, str] = {}
current_name: Optional[str] = None
buf: List[str] = []
for line in body.splitlines():
h1 = re.match(r"^#\s+(.+?)\s*$", line)
if h1:
if current_name is not None:
sections[current_name] = "\n".join(buf).strip()
current_name = h1.group(1).strip()
buf = []
else:
if current_name is not None:
buf.append(line)
if current_name is not None:
sections[current_name] = "\n".join(buf).strip()
return {"frontmatter": frontmatter, "sections": sections}
def extract_bullet_points(section_text: str) -> List[str]:
"""从一个 section 里提取列表项。支持 `1.` / `-` / `*` / `+` 三种。"""
points: List[str] = []
for line in section_text.splitlines():
line = line.strip()
if not line:
continue
m = re.match(r"^(?:\d+[.、)]|[-*+])\s+(.+)$", line)
if m:
points.append(m.group(1).strip())
return points
# ============================================================
# 拆卡
# ============================================================
_VERSION_TOKEN_RE = re.compile(
r"[A-Za-z][A-Za-z0-9]*(?:[-\.][A-Za-z0-9]+)+"
)
def _protect_version_tokens(text: str) -> Tuple[str, Dict[str, str]]:
"""把 GPT-5.5 / V4-Pro / 2026-04-25 / API价$5/M 这样的 token 用占位符替换,
避免被英文句点/连字符误切。返回 (替换后文本, 占位符到原文 map)。"""
mapping: Dict[str, str] = {}
def repl(m: re.Match) -> str:
key = f"\x01TOK{len(mapping):03d}\x01"
mapping[key] = m.group(0)
return key
return _VERSION_TOKEN_RE.sub(repl, text), mapping
def _restore_tokens(text: str, mapping: Dict[str, str]) -> str:
for k, v in mapping.items():
text = text.replace(k, v)
return text
def _split_card_text(point: str, max_main: int = 18) -> Tuple[str, str]:
"""
把一条"要点"拆成卡片主文字 + 副文字。
策略:
- 短于 max_main 直接当主文字,副文字空
- **先保护版本号 token** ("GPT-5.5" / "V4-Pro" / "2026-04-25") 不被切开
- 找中文句号/逗号/分号/冒号/破折号,在合理位置切
- 实在没自然断点才硬切
"""
point = point.strip()
if len(point) <= max_main:
return point, ""
protected, tokens = _protect_version_tokens(point)
# 占位符长度不同,简单用字符级 index 工作即可
# pattern 1: 中文/英文断句标点
safe_splits = re.compile(r"——|[,。;::,;!?!?]")
for m in safe_splits.finditer(protected):
if 3 <= m.start() <= max_main + 8:
main = _restore_tokens(protected[: m.start()], tokens).strip()
sub = _restore_tokens(protected[m.end():], tokens).strip()
if main and sub and len(main) >= 3:
return main, sub
# pattern 2: 单独的 `-`(保护过版本号后的安全连字符),两侧至少一侧是中文才切
for m in re.finditer(r"-", protected):
i = m.start()
if not (3 <= i <= max_main + 8):
continue
left = protected[i - 1] if i > 0 else ""
right = protected[i + 1] if i + 1 < len(protected) else ""
if re.match(r"[一-鿿]", left) or re.match(r"[一-鿿]", right):
main = _restore_tokens(protected[:i], tokens).strip()
sub = _restore_tokens(protected[i + 1:], tokens).strip()
if main and sub:
return main, sub
# 硬切(在原始 point 上,但避开占位符)
restored = _restore_tokens(protected, tokens)
return restored[:max_main].strip(), restored[max_main:].strip()
def build_card_plan(
parsed: Dict[str, Any],
image_style: Optional[str] = None,
card_count_override: Optional[int] = None,
output_dir: Optional[Path] = None,
) -> Dict[str, Any]:
"""
把解析好的 brief 拍平成可执行的 card_plan。
每张卡输出:
{
"index": 1,
"card_main": str,
"card_sub": str,
"prompt": str, # 填好 placeholder 的完整 Gemini prompt
"target_file": str, # 相对 output_dir 的目标文件名,如 images/01.png
"aspect": str,
"size": str,
}
"""
fm = parsed["frontmatter"] or {}
sections = parsed["sections"]
# 风格解析: CLI > frontmatter > 账号默认 > newspic 兜底(infographic-warm)
# 贴图模式走 mode="newspic",让兜底用高密度手绘水彩信息图,而不是文章模式
# 的线条手绘(hand-drawn-blue)
style_name = image_style or fm.get("image_style")
account = fm.get("account")
style = resolve_image_style(
cli_value=image_style,
frontmatter_value=fm.get("image_style"),
account_name=account,
mode="newspic",
)
# 提取要点
points_text = sections.get("要点") or sections.get("要点列表") or sections.get("卡片要点") or ""
points = extract_bullet_points(points_text)
if not points:
raise ValueError(
"brief.md 里找不到『# 要点』小节,或要点列表是空的。\n"
"格式: # 要点 下面 1. xxx / - xxx / * xxx 每行一个"
)
target_count = card_count_override or fm.get("card_count") or len(points)
if not 1 <= target_count <= 20:
raise ValueError(f"card_count 必须在 1-20 之间,当前 {target_count}")
if target_count > len(points):
raise ValueError(
f"card_count={target_count} > 要点数({len(points)})。"
f"请补要点或调小 card_count。"
)
points = points[:target_count]
# 构造每张卡
topic = fm.get("topic", "")
tmpl = style.get("prompt_template", {})
newspic_prompt = tmpl.get("newspic_card") or tmpl.get("article_inline", "")
if not newspic_prompt:
raise ValueError(
f"风格 '{style['style_name']}' 缺少 prompt_template.newspic_card,"
"无法用于贴图。改用 newspic_ready=true 的风格。"
)
canvas = style.get("canvas", {})
aspect = canvas.get("newspic_aspect", "1:1")
size = canvas.get("newspic_size", "1080x1080")
total = len(points)
cards: List[Dict[str, Any]] = []
for i, point in enumerate(points, start=1):
main, sub = _split_card_text(point)
prompt = (
newspic_prompt
.replace("{topic}", topic)
.replace("{card_main}", main)
.replace("{card_sub}", sub or main) # 副文字空就复用主文字
.replace("{point_full}", point) # 原始完整要点,给 AI 更多上下文渲染信息图
.replace("{card_index}", f"{i:02d}")
.replace("{card_total}", f"{total:02d}")
.replace("{image_subject}", f"{topic} - {main}")
)
cards.append({
"index": i,
"card_main": main,
"card_sub": sub,
"point_full": point,
"prompt": prompt,
"target_file": f"images/{i:02d}.png",
"aspect": aspect,
"size": size,
})
short_text = sections.get("短文本") or sections.get("短描述") or ""
plan = {
"topic": topic,
"title": fm.get("title", ""),
"account": account,
"image_style": style["style_name"],
"image_style_display": style.get("display_name", style["style_name"]),
"card_count": len(cards),
"cards": cards,
"short_text": short_text,
"short_text_target_chars": [100, 300],
"output_dir": str(output_dir) if output_dir else None,
}
return plan
# ============================================================
# CLI
# ============================================================
def main():
parser = argparse.ArgumentParser(
description="贴图拆卡 + 出图计划生成器",
formatter_class=argparse.RawDescriptionHelpFormatter,
)
parser.add_argument("brief", help="brief.md 路径")
parser.add_argument("-o", "--output", help="card_plan.json 输出路径(默认 <brief_dir>/card_plan.json)")
parser.add_argument("--image-style", help="覆盖 frontmatter 的 image_style")
parser.add_argument("--card-count", type=int, help="覆盖 frontmatter 的 card_count")
parser.add_argument("--dry-run", action="store_true",
help="只打印,不写文件")
args = parser.parse_args()
brief_path = Path(args.brief)
if not brief_path.exists():
print(f"[错误] brief.md 不存在: {brief_path}", file=sys.stderr)
sys.exit(1)
md_text = brief_path.read_text(encoding="utf-8")
try:
parsed = parse_brief(md_text)
plan = build_card_plan(
parsed,
image_style=args.image_style,
card_count_override=args.card_count,
output_dir=brief_path.parent,
)
except (ValueError, ConfigError) as e:
print(f"[错误] {e}", file=sys.stderr)
sys.exit(1)
# 控制台人类友好摘要
print(f"话题: {plan['topic']}")
print(f"风格: {plan['image_style']} ({plan['image_style_display']})")
print(f"卡数: {plan['card_count']}")
print(f"标题: {plan['title'] or '(未填,发布时可用 --title 指定)'}")
print(f"短文本: {'(已填)' if plan['short_text'] else '(未填,需 Claude 根据要点生成)'}")
print("-" * 60)
for c in plan["cards"]:
print(f" 卡 {c['index']:2d}: {c['card_main']}")
if c["card_sub"]:
print(f" └─ {c['card_sub']}")
print(f" → {c['target_file']} ({c['aspect']} {c['size']})")
print("-" * 60)
if args.dry_run:
print("[dry-run] 不写文件")
return
out_path = Path(args.output) if args.output else brief_path.parent / "card_plan.json"
out_path.write_text(json.dumps(plan, ensure_ascii=False, indent=2), encoding="utf-8")
print(f"card_plan.json 已写入: {out_path}")
print("\n下一步:")
print(" 1. Claude 按 card_plan.json 里每张卡的 prompt 调 scripts/generate_image.py 生图")
print(f" 目标路径: {brief_path.parent}/images/01.png, 02.png, ...")
print(" 2. 如短文本未填,Claude 根据要点撰写 100-300 字短描述写入 brief.md '# 短文本' 小节")
print(" 3. python3 ai_score.py <brief.md 里的短文本> --mode newspic --threshold 45")
print(f" 4. python3 publish.py --account <main|tech> --type newspic --brief {brief_path}")
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |