ZJU-REAL/Easel / skills/builtin/real-lab-paper-promotion
real-lab-paper-promotion
Verified当用户需要论文小红书宣传稿与忠实截图时,为 REAL Lab AI 论文生成小红书中文宣传文案及与正文对应的原始 PDF 截图。输入本地 PDF、PDF 下载地址、arXiv 链接或编号,输出严格限长的 MD 与截图文件夹;支持修改已有稿件,不用于自动发布或普通长篇论文博客。
| File | Kind | Size |
|---|---|---|
| EASEL-META.md | markdown | 1.5 KB |
| LICENSE | other | 11.1 KB |
| NOTICE | other | 194 B |
| README.md | markdown | 1.8 KB |
| SKILL.md | skill md | 5.5 KB |
| agents/openai.yaml | config | 329 B |
| examples/INDEX.md | markdown | 1.1 KB |
| examples/agent-g2.md | markdown | 1.5 KB |
| examples/computersd.md | markdown | 1.9 KB |
| examples/hybridcua.md | markdown | 1.4 KB |
| examples/spacecast-bench.md | markdown | 2.2 KB |
| examples/ttpo.md | markdown | 1.7 KB |
| examples/worldmembench.md | markdown | 2.9 KB |
| references/style-guidelines.md | markdown | 3.0 KB |
| references/workflow.md | markdown | 6.3 KB |
| requirements.txt | manifest | 16 B |
| scripts/promotion.py | script | 13.9 KB |
| scripts/test_promotion.py | script | 9.1 KB |
references/workflow.md
View on GitHub# 执行与数据契约
用户只需提供 arXiv 链接/编号、可下载 PDF 地址或本地 PDF。Agent 执行脚本并完成阅读、写作、选图和审核;这不是一个脱离模型就能自动写作的命令行服务。
## 命令
`SKILL` 为 Skill 安装目录,`PY` 为装有 requirements.txt 依赖的 Python。`OUT` 为 `outputs/<论文简称>`,`WORK` 为`outputs/<论文简称>/assets/paper-promotion` 子目录;也兼容位于 outputs 内的独立工作目录,但禁止把 OUT 放进 WORK、两者相同或把 WORK 放在 OUT 的其他位置。每篇论文使用新 WORK;同一篇改稿复用工作资料。
```bash
"$PY" "$SKILL/scripts/promotion.py" prepare 'https://arxiv.org/abs/2610.12402' --work "$WORK"
# 同一命令也接受 https://某站/论文.pdf、本地路径、arXiv: abs/pdf 链接或裸编号
"$PY" "$SKILL/scripts/promotion.py" check-text "$OUT/post.md"
"$PY" "$SKILL/scripts/promotion.py" render --work "$WORK" --out "$OUT"
"$PY" "$SKILL/scripts/promotion.py" hashes --work "$WORK" --out "$OUT"
"$PY" "$SKILL/scripts/promotion.py" validate --work "$WORK" --out "$OUT"
```
先 prepare,再由 Agent 写 evidence、post、manifest,之后 render、视觉检查、hashes、填写 review、validate。scripts 不会从空 manifest 猜测图框或生成文章。实际执行时用真实路径替换变量,示例 arXiv 不是要求处理的论文。
下载支持 HTTP(S)、重定向及 arXiv 新旧编号/版本号,校验 PDF 内容并记录来源和 SHA-256;遇到登录页、访问失败或扫描版,不伪装成功。可使用可用浏览工具获取论文并以本地 PDF 继续,不绕过访问控制。自动下载上限 200 MiB。prepare 输出 source.json、paper.pdf、text.txt、pages.json 和 pages/ 预览。扫描版的 caption/事实通过 OCR 辅助提取、视觉核对,原始截图不改变。
## 证据文件
`evidence.json` 是非空数组。为所有定量结果、核心方法、主要结论建立记录;资源、接收状态的证据也记录。正文定稿后 post_quote 必须仍是原文子串。
```json
[
{
"post_quote": "正文中实际出现的完整主张",
"page": 4,
"source_excerpt": "论文原文摘录,包含必要条件",
"conditions": "模型、数据集、评测设置、比较对象与指标单位",
"source_url": "仅工作资料可保存真实链接"
}
]
```
这是 schema 示意,字段值必须用当前论文证据替换,不可直接交付。
## 截图计划
`manifest.json`:
```json
{
"pdf_sha256": "source.json 中的 pdf_sha256",
"images": [
{"file": "01-first-page.png", "page": 1, "kind": "first-page", "bbox": null},
{
"file": "02-method.png",
"page": 3,
"kind": "method",
"bbox": [40, 70, 555, 330],
"caption_bbox": [40, 290, 555, 330],
"source_label": "Figure 2",
"caption": "PDF 中对应原始 caption 的准确文本",
"text_quote": "正文中唯一出现的原句或连续片段",
"reason": "具体解释截图怎样支持正文,以及包含哪些子图",
"dpi": 180
}
]
}
```
页码从1开始;坐标单位是 PDF point,使用 PyMuPDF `page.rect` 对应的显示页面坐标,原点在左上方。`pages.json` 的文本块坐标在旋转页上是未旋转坐标,需乘 `page.rotation_matrix` 转换;预览坐标则按其与 page.rect 尺寸比例转换。不要把图片像素直接当 PDF point。
bbox 是包含图表与原 caption 的**单一矩形**;caption_bbox 是其中 caption 的完整区域。默认180 dpi,后续图允许144–300 dpi。首页必须使用整页可见页面(PDF 的原始 CropBox),不改页边距、旋转、内容或添加标题。校验会从同一 PDF 重新渲染并逐像素比对。
组合图仅讨论部分时,保留对应子图及必要标签/图例和与之对应的原 caption。若单个矩形必然带入未讨论子图或大段邻栏,换图或略去;不能把不相邻区域拼到一起。原始 caption 中用于定义的上下文可以保留,但不能借 caption 携带额外无关分析。caption 跨页而不能原样完整裁剪时选择其他图,或明确说明该图未纳入;不要交付半截 caption。
多图对应同一段时允许同位置,依该段叙述顺序;跨段时按首次引用位置递增。图片数量取决于正文,无固定下限(除首页)或凑数目标。正文确实讨论且图表可安全提取时,应提取相应主图/表,不能为省事只输出首页。
## 语义与视觉审核
Agent 必须打开每张实际截图,并对照 PDF 原页与证据审核;程序检查 caption 的记录/区域,不能判断 caption 内容正确、图表完整、语义支持或事实真实性。核对标题不过度承诺、百分比与百分点、数据集/模型条件、结论范围、资源来源、所有非首页图的相关性。全图中的子图、图例、脚注、数值和 caption 均应清楚可读。
`hashes` 命令输出当前 PDF、正文、manifest、evidence 的摘要。审查后将其放入 review.json 的 hashes;其余结构如下。notes 记录当前论文的具体核对依据,不可机械填写“已检查”。
```json
{
"hashes": {"pdf": "摘要", "post": "摘要", "manifest": "摘要", "evidence": "摘要"},
"facts": {"passed": true, "notes": "逐项核对的页码、结果与实验条件"},
"grounding": {"passed": true, "notes": "各图和正文对应关系、未选图的原因"},
"visual": {"passed": true, "notes": "逐图检查标签、caption、边界与可读性结果"},
"style": {"passed": true, "notes": "选读案例与本篇借鉴之处,避免生硬套用"}
}
```
validate 验证限长、禁链接、图像顺序、有效坐标、caption 区域、图像像素忠实度、陈旧审核和多余交付文件。它不把人/模型的语义审核记录当成数学证明。任何改稿须重新核对证据、映射和语义审核;任何截图改动须重新打开检查。OUT 只允许 `post.md`、`images/`、工作资料 `assets/` 和 Easel 的 `.easel.json`;仅前两项是宣传成品。
计数口径:移除标题行开头的 `# ` 和标题后的一个空白分隔行,正文其余字符全部计入(包括文件末尾换行)。Unicode 码点与 UTF-16 单元都不超限,后者对 emoji 更保守;不宣称等同小红书未公开的实时计数算法。无需压到上限。
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |