paper-acquire

name	paper-acquire
description	Build or update the canonical raw bundle for a resolved paper under `~/docs/papers/{folder_slug}`. Use this skill whenever the user wants to download a paper PDF and normalize it into `paper.md`. For repository discovery, use `paper-repo` skill after this one. Prefer this skill for raw acquisition work; do not use it to write reading notes.
argument-hint	<folder_slug> | <metadata_path>
allowed-tools	Bash, Read, Write, Edit

Paper Acquire

Hydrate the canonical raw bundle for a resolved paper.

Ownership

This skill owns:

• ~/docs/papers/{folder_slug}/metadata.yaml (extends with assets, normalization)
• ~/docs/papers/{folder_slug}/paper/paper.pdf
• ~/docs/papers/{folder_slug}/paper/paper.md
• ~/docs/papers/{folder_slug}/paper/paper_images/

This skill does not:

• Write the final reading note
• Discover or clone repositories (use paper-repo)
• Modify identity or bibliography sections in metadata.yaml

Workflow

1. Start from a resolved metadata bundle

必须先有 metadata.yaml（由 paper-resolve 产出）。

如果用户只给了一个模糊引用，先用 paper-resolve。

读取 metadata.yaml，提取对 acquire 有用的信息：

# 从 metadata.yaml 提取
identifiers:
  arxiv: "1704.01212"        # → arXiv PDF
  pmcid: "PMC11399094"       # → PMC 全文 PDF
  semantic_scholar: "..."    # → S2 openAccessPdf
  doi: "10.xxx"              # → 出版商 PDF

urls:
  pdf: "https://..."         # → 直接下载 URL
  pmc: "https://pmc..."      # → PMC 页面，提取 PDF

2. PDF Acquisition (Priority Order)

按以下优先级下载 PDF：

优先级	来源	条件	URL 模式
1	arXiv PDF	identity.aliases.arxiv 存在	https://arxiv.org/pdf/{arxiv_id}.pdf
2	PMC PDF	identity.aliases.pmcid 存在	从 urls.pmc 页面提取
3	Semantic Scholar	urls.pdf 来自 S2	直接用 urls.pdf
4	出版商 PDF	identity.aliases.doi 存在	https://doi.org/{doi} 重定向

下载流程：

# 优先级 1: arXiv (免费)
if identity.aliases.arxiv exists:
  wget "https://arxiv.org/pdf/${arxiv_id}.pdf" -O paper/paper.pdf

# 优先级 2: PMC (免费全文)
if identity.aliases.pmcid exists:
  # PMC 页面通常有 PDF 链接
  crwlr crawl -o md "${urls.pmc}"  # 提取 PDF URL
  wget "${pmc_pdf_url}" -O paper/paper.pdf

# 优先级 3: Semantic Scholar openAccessPdf
if urls.pdf exists and looks like S2/pdfs.semanticscholar.org:
  wget "${urls.pdf}" -O paper/paper.pdf

# 优先级 4: 出版商 PDF (可能需要认证)
if identity.aliases.doi exists and no PDF yet:
  # 尝试 wget，可能需要 cdp-download 处理认证
  wget "https://doi.org/${doi}" -O paper/paper.pdf || \
  ${SKILL_DIR}/../web-kit/scripts/cdp-download "https://doi.org/${doi}" paper/paper.pdf

为什么这个优先级：

1. arXiv 最优先：免费 + 无认证
2. PMC 其次：免费全文
3. S2 openAccessPdf：免费开放获取
4. 出版商最后：可能需要订阅/认证

3. Normalization (PDF → Markdown)

使用 pdf-to-md skill 的 MinerU 进行转换：

# PDF → MinerU
python3 "${SKILL_DIR}/../pdf-to-md/scripts/mineru-api.py" paper/paper.pdf -l en
# 输出: paper/paper.md + paper/paper_images/

4. Update metadata.yaml

完成 acquisition 后，更新 metadata.yaml 的 assets 和 normalization 部分：

assets:
  paper_pdf:
    path: "paper/paper.pdf"
    size_bytes: 2950183
    source: "arxiv"  # 或 "pmc", "semantic_scholar", "publisher"
  paper_md:
    path: "paper/paper.md"
    backend: "mineru_vlm"
    generated_at: "2026-04-13"
  paper_images:
    path: "paper/images/"
    count: 12

normalization:
  backend: "mineru_vlm"
  completed_at: "2026-04-13"

Storage Rules

Read references/raw-layout.md for directory structure.

Dependencies

• web-kit skill — wget, cdp-download, crwlr
• pdf-to-md skill — MinerU API PDF conversion

References

文件	用途
raw-layout.md	目录结构规范