Newspaper Brief

Create mobile-friendly newspaper-style long images from raw text or summaries by extracting key points and rendering structured content in a clean HTML/PNG f...

Security Scan

The skill's files and runtime instructions are coherent with its stated purpose: it extracts/structures text and renders a local HTML/PNG newspaper-style long image, and it does not request unrelated credentials or network installs.

Purpose & Capability

Name/description (newspaper-style mobile long images) match the included SKILL.md and the render_newspaper.py script. The script and example JSON are exactly what a formatter/renderer would need; no unrelated credentials, binaries, or cloud access are requested.

Instruction Scope

SKILL.md instructs the agent to extract key points from user-provided text (or accept pre-made summaries) and then run the provided script to produce HTML/PNG. The instructions reference only local files and a local browser for headless screenshots. There are no directives to read unrelated system files, send data to external endpoints, or collect additional secrets.

Install Mechanism

No install spec is present (instruction-only skill with a helper script). The script is pure Python and uses standard libs plus optional Pillow; it does not download arbitrary archives or run network installs. This is a low-risk install posture.

Credentials

The skill requires no environment variables, credentials, or config paths. The script attempts to locate local Edge/Chrome executables (common for converting HTML to PNG) and can use Pillow if available; this is proportional to rendering functionality and does not imply access to unrelated secrets.

Persistence & Privilege

Flags show always:false and normal user-invocable/autonomous settings. The skill does not request persistent or elevated system privileges, nor does it modify other skills or system-wide configs.

Assessment

This skill appears to be what it says: a local renderer that turns structured or extracted text into a newspaper-style HTML/PNG. Before using: (1) be aware the agent (or you) will process any text you supply to produce highlights/sections — do not supply secrets you don't want rendered into files; (2) the script will try to invoke a local Chrome/Edge for headless screenshots and will write output files (e.g., output/newspaper-brief/demo.*) to disk, so ensure the runtime environment has a browser or accept only HTML output; (3) no network exfiltration or external endpoints are present in the code, but always inspect generated HTML/PNG before sharing publicly. If you need higher assurance, run the script in a sandboxed environment or review the full script locally (it is included).

把整理结果从“聊天流”切成“版面流”，先产出适合手机阅读的报纸风长图。

默认工作方式

优先使用这套流程：

1. 读用户给的原始材料（文章、聊天记录、链接摘要、会议纪要）
2. 先整理成结构化要点
3. 再渲染成报纸风长图

如果用户已经给了现成摘要/要点，也可以直接跳到第 3 步。

触发词

以下中文说法都应直接进入本技能：

• 报纸模式
• 做成报纸图
• 按报纸样式输出

也兼容这些自然说法：

• 整理成简报图
• 做成报纸样式
• 用报纸风输出
• 给我一张报纸风长图

v1 范围

• 仅做 手机版长图
• 先输出 HTML，再截图为 PNG
• 默认风格：黑白报纸 + 轻米色纸张背景 + 少量强调线
• 适合 Telegram 直接查看，不追求打印排版

数据准备

优先整理成下面这种结构，再交给脚本：

{
  "paper_name": "紫音简报",
  "issue": "2026-03-07",
  "title": "AI 输出界面值得重做",
  "subtitle": "把聊天流改造成版面流，比单纯堆更长回答更有效",
  "summary": "一句导语，放在标题下方。",
  "highlights": [
    "先给 3-5 条 TL;DR",
    "正文使用 2-4 个版块分栏",
    "手机端版式必须优先保证可读性"
  ],
  "quote": "AI 时代，真正该重做的可能不是模型，而是输出界面。",
  "sections": [
    {
      "heading": "核心判断",
      "body": "这里放正文段落。支持多段文本。"
    },
    {
      "heading": "产品建议",
      "bullets": ["建议一", "建议二", "建议三"]
    }
  ],
  "footer_note": "来源：群聊讨论｜整理：紫音"
}

原始材料模式

当用户只给原文/聊天记录时，先由代理自己完成提炼：

• 先写标题
• 再写一句导语
• 提炼 3-5 条 highlights
• 拆成 2-4 个 sections
• 最后补一句 footer

标题规则（v1.1）

• 标题最多两行
• 不允许用 ... / …… 省略截断
• 默认优先 重写成更短、更像报纸标题的表达
• 只有在仍然接近边界时，才轻微缩小字号

不要把未经整理的大段原文直接硬塞进模板，除非用户明确要求“原汁原味”。

脚本

1) 生成 HTML

python skills/newspaper-brief/scripts/render_newspaper.py \
  --input skills/newspaper-brief/examples/ai-interface-sample.json \
  --html output/newspaper-brief/demo.html \
  --normalized output/newspaper-brief/demo.normalized.json

2) 可选：顺手导出 PNG（本地浏览器兜底）

python skills/newspaper-brief/scripts/render_newspaper.py \
  --input skills/newspaper-brief/examples/ai-interface-sample.json \
  --html output/newspaper-brief/demo.html \
  --png output/newspaper-brief/demo.png

脚本会尝试寻找本机 Edge/Chrome 进行 headless screenshot，并在导出后自动裁掉底部多余空白。若 PNG 失败，HTML 仍然可用。

推荐截图方式

首选：OpenClaw browser 工具

当 browser 工具可用时：

1. 打开生成后的本地 HTML
2. 使用全页截图
3. 导出 PNG

优点：比纯命令行截图更稳，尤其是长图。

兜底：脚本内置浏览器截图

适合快速 smoke test，但超长页面可能截断。遇到特别长的内容时：

• 缩短单张图内容
• 或后续扩展为分页输出

版面建议

v1 保持固定结构：

• 报头（刊名 / 期号 / 日期）
• 大标题
• 导语
• 左侧主内容区
• 右侧速览栏（highlights + quote）
• 页脚（来源 / 整理说明）

质量检查

完成前至少检查：

• 标题是否够像“报纸标题”，不要像聊天句子
• highlights 是否是短句，不要过长
• section 数量是否控制在 2-4 个
• 正文是否有明显层次
• 手机上是否一眼能看到重点

样例

直接参考：

• skills/newspaper-brief/examples/ai-interface-sample.json

先用样例跑通，再替换成真实内容。