Skip to content

Yuan1z0825/nature-skills / skills/nature-paper-to-patent

nature-paper-to-patent

Flagged

Turn research papers or inventor materials into evidence-grounded Chinese invention patent drafts and technical disclosures. Use for 技术交底书、专利撰写、现有技术对比 and Chinese DOCX patent packages; not general manuscript writing.

FileKindSize
README.mdmarkdown2.9 KB
README_EN.mdmarkdown3.3 KB
SKILL.mdskill md4.5 KB
agents/openai.yamlconfig248 B
evals/evals.jsonconfig1.5 KB
manifest.yamlconfig3.1 KB
references/cn-patent-drafting-guide.mdmarkdown8.0 KB
references/corpus-derived-patterns.mdmarkdown3.5 KB
references/corpus-pair-audit.mdmarkdown2.9 KB
references/disclosure/correction_handler.mdmarkdown2.8 KB
references/disclosure/disclosure_builder.mdmarkdown12.1 KB
references/disclosure/disclosure_preview.mdmarkdown917 B
references/disclosure/disclosure_self_check.mdmarkdown5.9 KB
references/disclosure/intake.mdmarkdown943 B
references/disclosure/iteration_context.mdmarkdown4.3 KB
references/disclosure/merger.mdmarkdown3.4 KB
references/disclosure/patent-disclosure-skill-MIT-LICENSE.txtmarkdown1.0 KB
references/disclosure/patent_points_analyzer.mdmarkdown951 B
references/disclosure/prior_art_search.mdmarkdown10.4 KB
references/disclosure/project_scan.mdmarkdown3.6 KB
references/disclosure/template_reference.mdmarkdown9.2 KB
references/disclosure/tooling.mdmarkdown16.2 KB
references/draft-schema.mdmarkdown9.0 KB
references/patent-figure-guide.mdmarkdown2.9 KB
requirements.txtmanifest121 B
scripts/audit_claims.pyscript6.4 KB
scripts/build_patent_package.pyscript6.5 KB
scripts/disclosure/cnipa_epub_crawler.pyscript10.7 KB
scripts/disclosure/cnipa_epub_parse.pyscript8.8 KB
scripts/disclosure/cnipa_epub_search.pyscript5.9 KB
scripts/disclosure/docx_to_md.pyscript3.7 KB
scripts/disclosure/iteration_dialog_log.pyscript3.2 KB
scripts/disclosure/math_render.pyscript12.7 KB
scripts/disclosure/md_to_docx.pyscript32.2 KB
scripts/disclosure/mermaid_render.pyscript15.2 KB
scripts/disclosure/package-lock.jsonmanifest84.4 KB
scripts/disclosure/package.jsonmanifest279 B
scripts/disclosure/pptx_to_md.pyscript4.9 KB
scripts/disclosure/requirements-cnipa.txtmarkdown172 B
scripts/extract_pdf_text.pyscript2.3 KB
scripts/init_patent_project.pyscript3.1 KB
scripts/math_to_omml.pyscript4.8 KB
scripts/render_flowchart_svg.pyscript14.3 KB
scripts/render_patent_docx.pyscript10.3 KB
scripts/validate_patent_draft.pyscript14.6 KB
static/core/output-contract.mdmarkdown2.0 KB
static/core/principles.mdmarkdown1016 B
static/core/workflow.mdmarkdown3.8 KB
static/fragments/invention/algorithm-software.mdmarkdown366 B
static/fragments/invention/apparatus-system.mdmarkdown288 B
static/fragments/invention/mixed.mdmarkdown255 B
static/fragments/invention/process-material.mdmarkdown278 B
static/fragments/source/mixed-project.mdmarkdown369 B
static/fragments/source/pasted-text.mdmarkdown316 B
static/fragments/source/pdf-text.mdmarkdown290 B
static/fragments/source/scanned-pdf.mdmarkdown272 B
static/fragments/task/claim-set.mdmarkdown249 B
static/fragments/task/disclosure-analysis.mdmarkdown236 B
static/fragments/task/disclosure-iteration.mdmarkdown1005 B
static/fragments/task/full-draft.mdmarkdown247 B
static/fragments/task/paper-patent-audit.mdmarkdown280 B
static/fragments/task/technical-disclosure.mdmarkdown1.5 KB
tests/test_audit_claims.pyscript1.6 KB
tests/test_inline_math_paren.pyscript1.7 KB
tests/test_math_to_omml_superscript.pyscript2.2 KB
tests/test_md_to_docx_source_overwrite.pyscript1.4 KB
tests/test_mermaid_render.pyscript2.3 KB
tests/test_validation.pyscript6.0 KB
tests/test_validation_figure_node_ids.pyscript614 B

scripts/disclosure/md_to_docx.py

View on GitHub
#!/usr/bin/env python3
"""
将 Markdown 转为 Word(.docx),按标题层级映射为 Word 内置「标题 1–9」样式,
便于交底书交付代理人或所内流程。

支持:ATX 标题 (#–######)、段落、**粗体**、行内 `代码`、无序/有序列表、
围栏代码块、简单 GFM 表格、引用块(>)、水平线(---)、行内图片 ``![](path.png)``
(在最大宽、最大高约束下**等比缩放**,竖图自动缩小宽度以整图落入版面)。

**连续多行正文**(中间无空行、且非列表/标题等)时,**每一行**输出为 Word 中**独立一段**,
以便「(1)…(2)…」等分条换行;若须在同一段内接排,请写**同一行**内或用 Markdown 空行分隔逻辑段。

定稿宜先用同目录 **`mermaid_render.py`** 将 **mermaid** 转为 PNG;**LaTeX 公式**(``$...$`` / ``$$...$$``)由 **`math_render.py`**(或 ``md_to_docx`` 自动调用)转为 PNG;失败时保留原文写入 Word。

用法:
  python md_to_docx.py --input disclosure.md --output disclosure.docx
  python md_to_docx.py -i a.md -o b.docx --base-dir .   # 解析图片相对路径

依赖:python-docx
"""

from __future__ import annotations

import argparse
import re
import sys
from pathlib import Path

from docx import Document
from docx.enum.text import WD_ALIGN_PARAGRAPH
from docx.oxml.ns import qn
from docx.shared import Inches, Pt, RGBColor

# 插图最大尺寸(英寸):在常见 A4、默认边距下保证整图可见、按比例缩放(不过宽也不过高)。
_DEFAULT_IMAGE_MAX_W_IN = 5.5
_DEFAULT_IMAGE_MAX_H_IN = 8.2
# 公式图在 Word 中统一按固定高度嵌入(英寸),避免块级式随 PNG 像素被放大、行内式过小
_FORMULA_DISPLAY_MAX_H_IN = 0.17
# 兼容旧名
_FORMULA_INLINE_MAX_H_IN = _FORMULA_DISPLAY_MAX_H_IN
_FORMULA_BLOCK_MAX_W_IN = 4.0  # 仅作块级超宽时的宽度上限(通常由固定高度约束)
_FORMULA_BLOCK_MAX_H_IN = _FORMULA_DISPLAY_MAX_H_IN

_MD_IMAGE_RE = re.compile(r"!\[([^\]]*)\]\(([^)]+)\)")
_HIDDEN_MD_IMAGE_COMMENT_RE = re.compile(
    r"<!--\s*!\[([^\]]*)\]\(([^)]+)\)\s*-->"
)
_INLINE_MATH_WITH_HIDDEN_IMG_RE = re.compile(
    r"(?<!\$)\$(?!\$)((?:\\.|[^$\n])+?)\$(?!\$)\s*"
    r"<!--\s*!\[([^\]]*)\]\(([^)]+)\)\s*-->"
)
_INLINE_MATH_PAREN_WITH_HIDDEN_IMG_RE = re.compile(
    r"\\\(((?:(?!\\\)).)+?)\\\)\s*"
    r"<!--\s*!\[([^\]]*)\]\(([^)]+)\)\s*-->"
)


def _parse_hidden_image_comment(line: str) -> tuple[str, str] | None:
    m = _HIDDEN_MD_IMAGE_COMMENT_RE.match(line.strip())
    if not m:
        return None
    return m.group(1), m.group(2).strip()


def _try_embed_hidden_comment_line(
    doc: Document,
    line: str,
    base_dir: Path | None,
    *,
    image_max_w_in: float,
    image_max_h_in: float,
) -> bool:
    hidden = _parse_hidden_image_comment(line)
    if not hidden or not base_dir:
        return False
    alt, src = hidden
    if not _resolve_image_path(src, base_dir):
        return False
    _embed_from_image_ref(
        alt,
        src,
        base_dir,
        doc=doc,
        image_max_w_in=image_max_w_in,
        image_max_h_in=image_max_h_in,
    )
    return True


def _image_pixel_size(path: Path) -> tuple[int, int] | None:
    """读取常见位图宽高(像素),失败返回 None。不依赖 Pillow。"""
    try:
        raw = path.read_bytes()
    except OSError:
        return None
    if len(raw) >= 24 and raw.startswith(b"\x89PNG\r\n\x1a\n") and raw[12:16] == b"IHDR":
        w = int.from_bytes(raw[16:20], "big")
        h = int.from_bytes(raw[20:24], "big")
        if w > 0 and h > 0:
            return w, h
    if len(raw) >= 10 and raw[:3] == b"GIF" and raw[3:6] in (b"87a", b"89a"):
        w = int.from_bytes(raw[6:8], "little")
        h = int.from_bytes(raw[8:10], "little")
        if w > 0 and h > 0:
            return w, h
    if len(raw) >= 4 and raw.startswith(b"\xff\xd8"):
        i = 2
        n = len(raw)
        while i < n:
            if raw[i] != 0xFF:
                i += 1
                continue
            i += 1
            while i < n and raw[i] == 0xFF:
                i += 1
            if i >= n:
                break
            marker = raw[i]
            i += 1
            if marker in (0xD8, 0xD9):
                continue
            if marker == 0xDA:
                break
            if 0xD0 <= marker <= 0xD7:
                continue
            if i + 2 > n:
                break
            seg_len = int.from_bytes(raw[i : i + 2], "big")
            if seg_len < 2:
                break
            i += 2
            if marker in (0xC0, 0xC1, 0xC2) and i + 5 <= n:
                h = int.from_bytes(raw[i + 1 : i + 3], "big")
                w = int.from_bytes(raw[i + 3 : i + 5], "big")
                if w > 0 and h > 0:
                    return w, h
            i += seg_len - 2
    return None


def _fit_image_display_inches(
    px_w: int,
    px_h: int,
    *,
    max_w_in: float,
    max_h_in: float,
) -> tuple[Inches, Inches]:
    """在不超过 max_w / max_h 的前提下等比缩放,使整图落入版面。"""
    if px_w <= 0 or px_h <= 0:
        return Inches(max_w_in), Inches(max_h_in * 0.5)
    aw = max_w_in
    ah = aw * px_h / px_w
    if ah > max_h_in:
        ah = max_h_in
        aw = ah * px_w / px_h
    return Inches(aw), Inches(ah)


def _formula_image_kind(alt: str, src: str) -> str | None:
    """返回 ``block`` / ``inline`` 表示公式图,否则 None(含注释内引用)。"""
    a = alt or ""
    s = src.replace("\\", "/")
    if "math_figures" not in s and "公式" not in a:
        return None
    if "行内" in a:
        return "inline"
    return "block"


def _is_diagram_image(alt: str, src: str) -> bool:
    """mermaid 系统框图 / 流程图等(非公式,用全幅插图尺寸)。"""
    a = alt or ""
    s = src.replace("\\", "/")
    if "mermaid_figures" in s:
        return True
    if a.startswith("图示") or a.startswith("图 "):
        return True
    return False


def _span_overlaps(spans: list[tuple[int, int]], start: int, end: int) -> bool:
    return any(not (end <= s or start >= e) for s, e in spans)


def _embed_from_image_ref(
    alt: str,
    src: str,
    base_dir: Path | None,
    *,
    doc: Document | None = None,
    paragraph=None,
    image_max_w_in: float = _DEFAULT_IMAGE_MAX_W_IN,
    image_max_h_in: float = _DEFAULT_IMAGE_MAX_H_IN,
) -> None:
    """按公式 / 框图 / 普通图规则嵌入 PNG(仅公式用小尺寸)。"""
    ipath = _resolve_image_path(src, base_dir) if base_dir else None
    missing = f"[图片缺失: {alt or src}]"
    if not ipath:
        if paragraph is not None:
            paragraph.add_run(missing)
        elif doc is not None:
            doc.add_paragraph().add_run(missing)
        return

    kind = _formula_image_kind(alt, src)
    if kind == "inline":
        p = paragraph
        if p is None and doc is not None:
            p = doc.add_paragraph()
            p.paragraph_format.space_after = Pt(6)
            p.paragraph_format.line_spacing = 1.15
        if p is not None:
            _embed_picture_inline(p, ipath, max_h_in=_FORMULA_DISPLAY_MAX_H_IN)
        return

    if doc is None:
        if paragraph is not None:
            paragraph.add_run(missing)
        return

    if kind == "block":
        p = doc.add_paragraph()
        p.paragraph_format.space_after = Pt(6)
        p.paragraph_format.space_before = Pt(3)
        _embed_picture_inline(
            p,
            ipath,
            max_h_in=_FORMULA_DISPLAY_MAX_H_IN,
            max_w_in=_FORMULA_BLOCK_MAX_W_IN,
        )
    else:
        _embed_picture(
            doc,
            ipath,
            alt=alt,
            src=src,
            max_w_in=image_max_w_in,
            max_h_in=image_max_h_in,
            center=False,
        )


def _maybe_render_math_md(md_text: str, base_dir: Path) -> str:
    """若含 LaTeX 公式则尝试调用 ``math_render``(已注释的 PNG 引用会跳过)。"""
    if not re.search(r"\$\$|\\\[|\\\(|(?<!\$)\$(?!\$)", md_text):
        return md_text
    try:
        from math_render import render_markdown_math
    except ImportError:
        print(
            "[md_to_docx] 未安装 matplotlib,公式将按原文写入 Word",
            file=sys.stderr,
        )
        return md_text
    stub = base_dir / "_md_to_docx_math_stub.md"
    new_md, ok, failed = render_markdown_math(
        md_text,
        out_md_path=stub,
        assets_rel="math_figures",
    )
    if ok or failed:
        print(
            f"[md_to_docx] 公式渲染:{ok} 成功,{failed} 保留原文",
            file=sys.stderr,
        )
    return new_md


def _add_math_fallback_block(doc: Document, lines: list[str]) -> None:
    """未渲染成功的 ``$$ ... $$`` 以等宽原文写入 Word。"""
    body = [ln.rstrip("\n") for ln in lines]
    _add_code_block(doc, ["$$", *body, "$$"])


def _embed_picture(
    doc: Document,
    path: Path,
    *,
    alt: str,
    src: str,
    max_w_in: float,
    max_h_in: float,
    center: bool,
) -> None:
    p = doc.add_paragraph()
    if center:
        p.alignment = WD_ALIGN_PARAGRAPH.CENTER
    p.paragraph_format.space_after = Pt(6)
    p.paragraph_format.space_before = Pt(3)
    try:
        dims = _image_pixel_size(path)
        if dims:
            w_in, h_in = _fit_image_display_inches(
                *dims, max_w_in=max_w_in, max_h_in=max_h_in
            )
            run = p.add_run()
            run.font.bold = False
            run.add_picture(str(path.resolve()), width=w_in, height=h_in)
        else:
            run = p.add_run()
            run.font.bold = False
            run.add_picture(str(path.resolve()), width=Inches(max_w_in))
    except Exception:
        p.add_run(f"[图片无法嵌入: {alt or src} — {path}]")


def _embed_picture_inline(
    paragraph,
    path: Path,
    *,
    max_h_in: float,
    max_w_in: float | None = None,
) -> None:
    try:
        dims = _image_pixel_size(path)
        run = paragraph.add_run()
        run.font.bold = False
        if dims:
            px_w, px_h = dims
            h_in = max_h_in
            w_in = h_in * px_w / px_h if px_h else max_h_in
            if max_w_in is not None and w_in > max_w_in:
                w_in = max_w_in
                h_in = w_in * px_h / px_w if px_w else max_h_in
            run.add_picture(str(path.resolve()), width=Inches(w_in), height=Inches(h_in))
        else:
            run.add_picture(str(path.resolve()), height=Inches(max_h_in))
    except Exception:
        paragraph.add_run(f"[行内公式图缺失: {path}]")


def _add_rich_content_to_paragraph(
    paragraph,
    text: str,
    base_dir: Path | None,
    *,
    formula_inline_max_h_in: float = _FORMULA_DISPLAY_MAX_H_IN,
    image_max_w_in: float = _DEFAULT_IMAGE_MAX_W_IN,
    image_max_h_in: float = _DEFAULT_IMAGE_MAX_H_IN,
    mono: bool = False,
) -> None:
    """同一段内混排文字(**粗体**/`代码`)与公式/插图(含 HTML 注释隐藏引用)。"""
    taken: list[tuple[int, int]] = []
    tokens: list[tuple[int, int, str, tuple]] = []

    for m in _INLINE_MATH_WITH_HIDDEN_IMG_RE.finditer(text):
        tokens.append((m.start(), m.end(), "math_img", (m.group(2), m.group(3).strip())))
        taken.append((m.start(), m.end()))

    for m in _INLINE_MATH_PAREN_WITH_HIDDEN_IMG_RE.finditer(text):
        if _span_overlaps(taken, m.start(), m.end()):
            continue
        tokens.append((m.start(), m.end(), "math_img", (m.group(2), m.group(3).strip())))
        taken.append((m.start(), m.end()))

    for m in _HIDDEN_MD_IMAGE_COMMENT_RE.finditer(text):
        if _span_overlaps(taken, m.start(), m.end()):
            continue
        tokens.append((m.start(), m.end(), "hidden_img", (m.group(1), m.group(2).strip())))
        taken.append((m.start(), m.end()))

    for m in _MD_IMAGE_RE.finditer(text):
        if _span_overlaps(taken, m.start(), m.end()):
            continue
        tokens.append((m.start(), m.end(), "visible_img", (m.group(1), m.group(2).strip())))
        taken.append((m.start(), m.end()))

    inline_pat = re.compile(r"(\*\*[^*]+?\*\*|`[^`]+?`)")
    for m in inline_pat.finditer(text):
        if _span_overlaps(taken, m.start(), m.end()):
            continue
        tokens.append((m.start(), m.end(), "inline", (m.group(1),)))
        taken.append((m.start(), m.end()))

    tokens.sort(key=lambda t: t[0])
    pos = 0
    for start, end, kind, payload in tokens:
        if start > pos:
            _add_inline_to_paragraph(paragraph, text[pos:start], mono=mono)
        if kind == "inline":
            token = payload[0]
            if token.startswith("**"):
                run = paragraph.add_run(token[2:-2])
                _set_run_font(run, "宋体", 10.5, bold=True)
            else:
                run = paragraph.add_run(token[1:-1])
                _set_run_font(run, "Consolas", 9)
                run.font.color.rgb = RGBColor(0x33, 0x33, 0x33)
        else:
            alt, src = payload[0], payload[1]
            _embed_from_image_ref(
                alt,
                src,
                base_dir,
                paragraph=paragraph,
                image_max_w_in=image_max_w_in,
                image_max_h_in=image_max_h_in,
            )
        pos = end
    if pos < len(text):
        _add_inline_to_paragraph(paragraph, text[pos:], mono=mono)


def _set_run_font(run, name: str = "宋体", size_pt: float | None = None, bold: bool | None = None):
    run.font.name = name
    run._element.rPr.rFonts.set(qn("w:eastAsia"), name)
    if size_pt is not None:
        run.font.size = Pt(size_pt)
    if bold is not None:
        run.font.bold = bold


def _add_inline_to_paragraph(paragraph, text: str, *, mono: bool = False):
    """解析 **粗体**、`行内代码` 与普通文本,写入同一段落。"""
    if not text:
        return
    # 拆分为:粗体、行内代码、普通
    pattern = re.compile(r"(\*\*[^*]+?\*\*|`[^`]+?`)")
    pos = 0
    for m in pattern.finditer(text):
        if m.start() > pos:
            run = paragraph.add_run(text[pos : m.start()])
            _set_run_font(run, "Consolas" if mono else "宋体", 10.5 if not mono else 9)
        token = m.group(1)
        if token.startswith("**"):
            run = paragraph.add_run(token[2:-2])
            _set_run_font(run, "宋体", 10.5, bold=True)
        else:  # `code`
            run = paragraph.add_run(token[1:-1])
            _set_run_font(run, "Consolas", 9)
            run.font.color.rgb = RGBColor(0x33, 0x33, 0x33)
        pos = m.end()
    if pos < len(text):
        run = paragraph.add_run(text[pos:])
        _set_run_font(run, "Consolas" if mono else "宋体", 10.5 if not mono else 9)


def _add_heading(doc: Document, level: int, text: str):
    """level 1–9 对应 Word 标题 1–标题 9;去除行内标记时保留可读文本。"""
    plain = re.sub(r"\*\*([^*]+)\*\*", r"\1", text)
    plain = re.sub(r"`([^`]+)`", r"\1", plain)
    h = doc.add_heading(plain, level=min(max(level, 1), 9))
    for run in h.runs:
        _set_run_font(run, "黑体" if level <= 2 else "宋体")


def _add_body_paragraph(
    doc: Document,
    text: str,
    base_dir: Path | None = None,
    *,
    image_max_h_in: float = _DEFAULT_IMAGE_MAX_H_IN,
):
    p = doc.add_paragraph()
    p.paragraph_format.space_after = Pt(6)
    p.paragraph_format.line_spacing = 1.15
    if (
        _MD_IMAGE_RE.search(text)
        or _HIDDEN_MD_IMAGE_COMMENT_RE.search(text)
        or _INLINE_MATH_WITH_HIDDEN_IMG_RE.search(text)
        or _INLINE_MATH_PAREN_WITH_HIDDEN_IMG_RE.search(text)
    ):
        _add_rich_content_to_paragraph(
            p,
            text,
            base_dir,
            image_max_w_in=_DEFAULT_IMAGE_MAX_W_IN,
            image_max_h_in=image_max_h_in,
        )
    else:
        _add_inline_to_paragraph(p, text)
    for run in p.runs:
        if run.font.name in (None, ""):
            _set_run_font(run, "宋体", 10.5)


def _add_code_block(doc: Document, lines: list[str]):
    p = doc.add_paragraph()
    p.paragraph_format.left_indent = Inches(0.2)
    p.paragraph_format.space_after = Pt(6)
    p.paragraph_format.keep_together = True
    body = "\n".join(lines)
    run = p.add_run(body)
    _set_run_font(run, "Consolas", 9)
    run.font.color.rgb = RGBColor(0x20, 0x20, 0x20)


def _add_list_item(
    doc: Document,
    text: str,
    ordered: bool,
    base_dir: Path | None,
    *,
    image_max_h_in: float = _DEFAULT_IMAGE_MAX_H_IN,
):
    style = "List Number" if ordered else "List Bullet"
    try:
        p = doc.add_paragraph(style=style)
    except (KeyError, ValueError):
        p = doc.add_paragraph()
        p.paragraph_format.left_indent = Inches(0.35)
    p.paragraph_format.space_after = Pt(3)
    if (
        _MD_IMAGE_RE.search(text)
        or _HIDDEN_MD_IMAGE_COMMENT_RE.search(text)
        or _INLINE_MATH_WITH_HIDDEN_IMG_RE.search(text)
        or _INLINE_MATH_PAREN_WITH_HIDDEN_IMG_RE.search(text)
    ):
        _add_rich_content_to_paragraph(
            p,
            text,
            base_dir,
            image_max_w_in=_DEFAULT_IMAGE_MAX_W_IN,
            image_max_h_in=image_max_h_in,
        )
    else:
        _add_inline_to_paragraph(p, text)
    for run in p.runs:
        _set_run_font(run, "宋体", 10.5)


def _is_table_row(line: str) -> bool:
    s = line.strip()
    return s.startswith("|") and s.endswith("|") and "|" in s[1:-1]


def _split_table_cells(line: str) -> list[str]:
    """按列分隔符 ``|`` 拆分表格行,忽略 ``\\(...\\)``、``$...$``、``<!-- -->`` 与 ``\\|`` 内的竖线。"""
    s = line.strip()
    if s.startswith("|"):
        s = s[1:]
    if s.endswith("|"):
        s = s[:-1]

    cells: list[str] = []
    buf: list[str] = []
    i = 0
    n = len(s)

    while i < n:
        if s.startswith("<!--", i):
            end = s.find("-->", i)
            if end == -1:
                buf.append(s[i:])
                break
            buf.append(s[i : end + 3])
            i = end + 3
            continue

        if s.startswith("\\(", i):
            end = s.find("\\)", i + 2)
            if end == -1:
                buf.append(s[i:])
                break
            buf.append(s[i : end + 2])
            i = end + 2
            continue

        if s[i] == "$":
            if i + 1 < n and s[i + 1] == "$":
                end = s.find("$$", i + 2)
                if end == -1:
                    buf.append(s[i:])
                    break
                buf.append(s[i : end + 2])
                i = end + 2
                continue
            j = i + 1
            while j < n:
                if s[j] == "$" and (j == 0 or s[j - 1] != "\\"):
                    buf.append(s[i : j + 1])
                    i = j + 1
                    break
                j += 1
            else:
                buf.append(s[i:])
                break
            continue

        if s[i] == "\\" and i + 1 < n and s[i + 1] == "|":
            buf.append("\\|")
            i += 2
            continue

        if s[i] == "|":
            cells.append("".join(buf).strip())
            buf = []
            i += 1
            continue

        buf.append(s[i])
        i += 1

    cells.append("".join(buf).strip())
    return cells


def _parse_table_row(line: str) -> list[str]:
    return _split_table_cells(line)


def _is_table_sep(row: list[str]) -> bool:
    if not row:
        return False
    return all(re.match(r"^:?-{3,}:?$", c.strip()) for c in row if c.strip())


def _add_table(doc: Document, rows: list[list[str]], base_dir: Path | None = None):
    if not rows:
        return
    ncols = max(len(r) for r in rows)
    table = doc.add_table(rows=len(rows), cols=ncols)
    table.style = "Table Grid"
    for i, row in enumerate(rows):
        for j in range(ncols):
            cell_text = row[j] if j < len(row) else ""
            cell = table.rows[i].cells[j]
            cell.text = ""
            p = cell.paragraphs[0]
            if _line_has_embeddable_images(cell_text):
                _add_rich_content_to_paragraph(p, cell_text, base_dir)
            else:
                _add_inline_to_paragraph(p, cell_text)
            for run in p.runs:
                _set_run_font(run, "宋体", 10)


def _add_horizontal_rule(doc: Document):
    p = doc.add_paragraph()
    p.paragraph_format.space_after = Pt(8)
    p.paragraph_format.space_before = Pt(8)
    run = p.add_run("─" * 32)
    _set_run_font(run, "宋体", 8)
    run.font.color.rgb = RGBColor(0xAA, 0xAA, 0xAA)


def _resolve_image_path(src: str, base_dir: Path | None) -> Path | None:
    if not base_dir:
        return None
    path = (base_dir / src).resolve() if not Path(src).is_absolute() else Path(src)
    return path if path.is_file() else None


def _try_add_image(
    doc: Document,
    line: str,
    base_dir: Path | None,
    *,
    max_w_in: float = _DEFAULT_IMAGE_MAX_W_IN,
    max_h_in: float = _DEFAULT_IMAGE_MAX_H_IN,
) -> bool:
    m = _MD_IMAGE_RE.match(line.strip())
    if not m or not base_dir:
        return False
    alt, src = m.group(1), m.group(2).strip()
    _embed_from_image_ref(
        alt,
        src,
        base_dir,
        doc=doc,
        image_max_w_in=max_w_in,
        image_max_h_in=max_h_in,
    )
    return True


def _line_has_embeddable_images(line: str) -> bool:
    return bool(
        _MD_IMAGE_RE.search(line)
        or _HIDDEN_MD_IMAGE_COMMENT_RE.search(line)
        or _INLINE_MATH_WITH_HIDDEN_IMG_RE.search(line)
        or _INLINE_MATH_PAREN_WITH_HIDDEN_IMG_RE.search(line)
    )


def _add_paragraph_with_inline_images(
    doc: Document,
    line: str,
    base_dir: Path | None,
    *,
    max_w_in: float = _DEFAULT_IMAGE_MAX_W_IN,
    max_h_in: float = _DEFAULT_IMAGE_MAX_H_IN,
) -> None:
    """段落内混排文字与公式/插图(含 HTML 注释隐藏引用)。"""
    p = doc.add_paragraph()
    p.paragraph_format.space_after = Pt(6)
    p.paragraph_format.line_spacing = 1.15
    _add_rich_content_to_paragraph(
        p,
        line,
        base_dir,
        image_max_w_in=max_w_in,
        image_max_h_in=max_h_in,
    )
    for run in p.runs:
        if run.font.name in (None, ""):
            _set_run_font(run, "宋体", 10.5)


def convert_md_to_docx(
    md_text: str,
    base_dir: Path | None,
    *,
    image_max_w_in: float = _DEFAULT_IMAGE_MAX_W_IN,
    image_max_h_in: float = _DEFAULT_IMAGE_MAX_H_IN,
) -> Document:
    doc = Document()
    # 默认正文样式
    try:
        style = doc.styles["Normal"]
        style.font.name = "宋体"
        if style._element.rPr is not None:
            style._element.rPr.rFonts.set(qn("w:eastAsia"), "宋体")
        style.font.size = Pt(10.5)
    except (AttributeError, KeyError):
        pass

    lines = md_text.splitlines()
    i = 0
    para_buf: list[str] = []

    def flush_paragraph():
        nonlocal para_buf
        if not para_buf:
            return
        # 每行独立成段,避免「(1)…\n(2)…」被空格拼成一段(Word 内不换行)
        for p in para_buf:
            t = p.strip()
            if t:
                _add_body_paragraph(
                    doc,
                    t,
                    base_dir,
                    image_max_h_in=image_max_h_in,
                )
        para_buf = []

    while i < len(lines):
        raw = lines[i]
        line = raw.rstrip("\n")

        if line.strip() == "":
            flush_paragraph()
            i += 1
            continue

        # 围栏代码块
        if line.strip().startswith("```"):
            flush_paragraph()
            fence_lang = line.strip()[3:].strip()
            i += 1
            code_lines: list[str] = []
            while i < len(lines) and not lines[i].strip().startswith("```"):
                code_lines.append(lines[i])
                i += 1
            if i < len(lines):
                i += 1
            # 定稿 MD 保留 mermaid 源码 + 图示注释:Word 只嵌 PNG,不写源码块
            if fence_lang.lower() == "mermaid":
                j = i
                while j < len(lines) and lines[j].strip() == "":
                    j += 1
                if j < len(lines):
                    cm = _HIDDEN_MD_IMAGE_COMMENT_RE.match(lines[j].strip())
                    if cm and _is_diagram_image(cm.group(1), cm.group(2).strip()):
                        continue
            _add_code_block(doc, code_lines)
            continue

        # 块级公式:\[ ... \] + 可选 HTML 注释
        if line.strip() == "\\[":
            flush_paragraph()
            i += 1
            math_lines: list[str] = []
            while i < len(lines) and lines[i].strip() != "\\]":
                math_lines.append(lines[i])
                i += 1
            if i < len(lines):
                i += 1
            hidden: tuple[str, str] | None = None
            if i < len(lines):
                cm = _HIDDEN_MD_IMAGE_COMMENT_RE.match(lines[i].strip())
                if cm:
                    hidden = (cm.group(1), cm.group(2).strip())
                    i += 1
            if hidden and _formula_image_kind(*hidden):
                ipath = _resolve_image_path(hidden[1], base_dir)
                if ipath:
                    _embed_from_image_ref(
                        hidden[0],
                        hidden[1],
                        base_dir,
                        doc=doc,
                        image_max_w_in=image_max_w_in,
                        image_max_h_in=image_max_h_in,
                    )
                    continue
            _add_math_fallback_block(doc, ["\\[", *math_lines, "\\]"])
            continue

        # 块级公式:$$ ... $$ + 可选 HTML 注释(Word 嵌 PNG;预览见 LaTeX 原文)
        if line.strip() == "$$":
            flush_paragraph()
            i += 1
            math_lines: list[str] = []
            while i < len(lines) and lines[i].strip() != "$$":
                math_lines.append(lines[i])
                i += 1
            if i < len(lines):
                i += 1
            hidden: tuple[str, str] | None = None
            if i < len(lines):
                cm = _HIDDEN_MD_IMAGE_COMMENT_RE.match(lines[i].strip())
                if cm:
                    hidden = (cm.group(1), cm.group(2).strip())
                    i += 1
            if hidden and _formula_image_kind(*hidden):
                ipath = _resolve_image_path(hidden[1], base_dir)
                if ipath:
                    _embed_from_image_ref(
                        hidden[0],
                        hidden[1],
                        base_dir,
                        doc=doc,
                        image_max_w_in=image_max_w_in,
                        image_max_h_in=image_max_h_in,
                    )
                    continue
            _add_math_fallback_block(doc, math_lines)
            continue

        # 独立 HTML 注释行(公式图 / mermaid 框图引用)
        if _HIDDEN_MD_IMAGE_COMMENT_RE.fullmatch(line.strip()):
            flush_paragraph()
            _try_embed_hidden_comment_line(
                doc,
                line,
                base_dir,
                image_max_w_in=image_max_w_in,
                image_max_h_in=image_max_h_in,
            )
            i += 1
            continue

        # 图片行或含行内公式/注释的段落
        if _line_has_embeddable_images(line):
            flush_paragraph()
            stripped = line.strip()
            if _MD_IMAGE_RE.fullmatch(stripped) or (
                stripped.startswith("![") and stripped.count("![") == 1
            ):
                _try_add_image(
                    doc,
                    line,
                    base_dir,
                    max_w_in=image_max_w_in,
                    max_h_in=image_max_h_in,
                )
            else:
                _add_paragraph_with_inline_images(
                    doc,
                    line,
                    base_dir,
                    max_w_in=image_max_w_in,
                    max_h_in=image_max_h_in,
                )
            i += 1
            continue

        # 水平线
        if re.match(r"^[\s\-*_]{3,}\s*$", line) and set(line.strip()) <= {"-", "*", "_", " "}:
            flush_paragraph()
            _add_horizontal_rule(doc)
            i += 1
            continue

        # 标题
        m = re.match(r"^(#{1,6})\s+(.+)$", line)
        if m:
            flush_paragraph()
            level = len(m.group(1))
            title = m.group(2).strip()
            title = re.sub(r"\s+#+\s*$", "", title)
            _add_heading(doc, level, title)
            i += 1
            continue

        # 引用
        if line.lstrip().startswith("> "):
            flush_paragraph()
            quote = line.lstrip()[2:].strip()
            p = doc.add_paragraph()
            p.paragraph_format.left_indent = Inches(0.25)
            p.paragraph_format.space_after = Pt(4)
            _add_inline_to_paragraph(p, quote)
            for run in p.runs:
                _set_run_font(run, "宋体", 10.5)
            i += 1
            continue

        # 表格块
        if _is_table_row(line):
            flush_paragraph()
            table_rows: list[list[str]] = []
            while i < len(lines) and _is_table_row(lines[i]):
                row = _parse_table_row(lines[i])
                if not _is_table_sep(row):
                    table_rows.append(row)
                i += 1
            _add_table(doc, table_rows, base_dir)
            continue

        # 无序列表
        um = re.match(r"^(\s*)[-*+]\s+(.+)$", line)
        if um:
            flush_paragraph()
            _add_list_item(
                doc,
                um.group(2).strip(),
                ordered=False,
                base_dir=base_dir,
                image_max_h_in=image_max_h_in,
            )
            i += 1
            continue

        # 有序列表
        om = re.match(r"^(\s*)\d+\.\s+(.+)$", line)
        if om:
            flush_paragraph()
            _add_list_item(
                doc,
                om.group(2).strip(),
                ordered=True,
                base_dir=base_dir,
                image_max_h_in=image_max_h_in,
            )
            i += 1
            continue

        para_buf.append(line)
        i += 1

    flush_paragraph()
    return doc


def main(argv: list[str] | None = None) -> int:
    p = argparse.ArgumentParser(description="Markdown → Word(标题样式映射)")
    p.add_argument("-i", "--input", required=True, help="输入 .md 路径")
    p.add_argument("-o", "--output", required=True, help="输出 .docx 路径")
    p.add_argument(
        "--base-dir",
        default=None,
        help="解析 ![](/相对路径) 图片时的根目录(默认使用 .md 所在目录)",
    )
    p.add_argument(
        "--image-max-width-inches",
        type=float,
        default=_DEFAULT_IMAGE_MAX_W_IN,
        metavar="IN",
        help=f"插图最大宽度(英寸,默认 {_DEFAULT_IMAGE_MAX_W_IN}),与高度共同约束等比缩放",
    )
    p.add_argument(
        "--image-max-height-inches",
        type=float,
        default=_DEFAULT_IMAGE_MAX_H_IN,
        metavar="IN",
        help=f"插图最大高度(英寸,默认 {_DEFAULT_IMAGE_MAX_H_IN}),避免竖图仅按宽度缩放后超出单页可视区域",
    )
    p.add_argument(
        "--no-math-render",
        action="store_true",
        help="不自动调用 math_render(默认会先渲染 $ / $$ 公式为 PNG)",
    )
    args = p.parse_args(argv)

    in_path = Path(args.input).resolve()
    if not in_path.is_file():
        print(f"错误:找不到输入文件 {in_path}", file=sys.stderr)
        return 1
    out_path = Path(args.output).resolve()
    if in_path == out_path:
        print("输出 Word 文件不能覆盖输入 Markdown 原稿。", file=sys.stderr)
        return 1

    base = Path(args.base_dir).resolve() if args.base_dir else in_path.parent
    try:
        md_text = in_path.read_text(encoding="utf-8")
    except UnicodeDecodeError:
        md_text = in_path.read_text(encoding="utf-8", errors="replace")
        print("警告:输入文件含非 UTF-8 字节,已使用替换字符解码后继续转换。", file=sys.stderr)

    if not args.no_math_render:
        md_text = _maybe_render_math_md(md_text, base)

    doc = convert_md_to_docx(
        md_text,
        base_dir=base,
        image_max_w_in=args.image_max_width_inches,
        image_max_h_in=args.image_max_height_inches,
    )
    out_path.parent.mkdir(parents=True, exist_ok=True)
    doc.save(str(out_path))
    print(f"已写入: {out_path}")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall