@lemoncat7/dsh-partner 2.8.1 → 2.8.2

package/resources/skills/story-creation/upstream/skills/story-long-analyze/scripts/render_relation_chart.py

@@ -0,0 +1,425 @@
+#!/usr/bin/env python3
+"""Render the character-relationship charts from 角色/角色关系.md.
+
+The Markdown file ``人物关系图/人物关系图.md`` (Mermaid graphs plus a plain
+list) is always written: it shows Chinese names in any Markdown viewer. PNG
+pictures are optional (``--png``) and are drawn only when matplotlib and a font
+that really contains every drawn character are available. Emoji and other
+symbols are left out of the pictures; without a font for the remaining text no
+PNG is written, so names are never replaced by pinyin, initials or empty boxes.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import math
+import os
+import re
+import shutil
+import subprocess
+import sys
+import tempfile
+import unicodedata
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple
+
+
+RELATION_FILE = Path("角色") / "角色关系.md"
+CHART_DIR = Path("人物关系图")
+CHART_MD = CHART_DIR / "人物关系图.md"
+CORE_PNG = CHART_DIR / "核心人物关系.png"
+EVOLUTION_PNG = CHART_DIR / "关键关系演变.png"
+# 「甲 → 乙」, 「甲 ↔ 乙」 (both ways) and chains such as 「甲 → 乙 → 丙」.
+ARROW_RE = re.compile(r"\s*(↔|<->|⇔|<=>|→|->|⇒|=>|—>|－>)\s*")
+BOTH_WAYS = {"↔", "<->", "⇔", "<=>"}
+# Common Simplified-Chinese capable fonts on macOS, Windows and Linux.
+CJK_FONT_NAMES = (
+    "PingFang SC", "Hiragino Sans GB", "Heiti SC", "STHeiti", "Songti SC", "STSong",
+    "Microsoft YaHei", "SimHei", "SimSun", "DengXian",
+    "Noto Sans CJK SC", "Noto Sans SC", "Noto Serif CJK SC", "Source Han Sans SC",
+    "Source Han Sans CN", "WenQuanYi Micro Hei", "WenQuanYi Zen Hei", "Sarasa Gothic SC",
+    "Arial Unicode MS",
+)
+# Fonts that answer every character with a placeholder box; they never count.
+FALLBACK_FONT_RE = re.compile(r"last\s*resort|adobe\s*blank", re.IGNORECASE)
+BASE_TEXT = "中文人物关系未命名→…"
+MD_VIEW = "关系图已经写在 人物关系图/人物关系图.md 里，用能预览 Markdown 的编辑器打开就能看到中文。"
+NO_FONT_MESSAGE = (
+    "这台电脑上没有找到能显示中文的字体，所以没有生成关系图图片（硬画会把人名变成拼音或方块）。"
+    + MD_VIEW + "装上任意一款中文字体（如思源黑体、文泉驿）后再生成一次，就能得到图片版。"
+)
+MISSING_GLYPH_MESSAGE = (
+    "这台电脑上的中文字体显示不了关系图里的「%s」，所以没有生成图片（硬画会变成方块）。"
+    + MD_VIEW + "装上字更全的中文字体（如思源黑体）后再生成一次，就能得到图片版。"
+)
+PNG_FAILED_MESSAGE = "关系图图片这次没画成。" + MD_VIEW
+NO_MATPLOTLIB_MESSAGE = "当前环境不能画图片，" + MD_VIEW
+
+
+class ChartError(ValueError):
+    def __init__(self, code: str, author_message: str) -> None:
+        super().__init__(code)
+        self.code = code
+        self.author_message = author_message
+
+
+def atomic_write(path: Path, data: bytes) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    fd, temporary = tempfile.mkstemp(prefix=".%s." % path.name, suffix=".tmp", dir=str(path.parent))
+    try:
+        with os.fdopen(fd, "wb") as handle:
+            handle.write(data)
+        os.replace(temporary, str(path))
+    except BaseException:
+        try:
+            os.unlink(temporary)
+        except OSError:
+            pass
+        raise
+
+
+def cells(line: str) -> List[str]:
+    """Table cells; an escaped 「\\|」 stays inside its cell."""
+    stripped = line.strip()
+    stripped = stripped[1:] if stripped.startswith("|") else stripped
+    stripped = stripped[:-1] if stripped.endswith("|") and not stripped.endswith("\\|") else stripped
+    return [cell.strip().replace("\\|", "|") for cell in re.split(r"(?<!\\)\|", stripped)]
+
+
+def pick(row: Dict[str, str], *names: str) -> str:
+    for name in names:
+        for key, value in row.items():
+            if name in key and value and value not in {"-", "—", "无"}:
+                return value
+    return ""
+
+
+SOURCE_KEYS = ("主体", "角色A", "人物A", "A", "甲")
+TARGET_KEYS = ("客体", "角色B", "人物B", "B", "乙")
+
+
+def exact(row: Dict[str, str], keys: Tuple[str, ...]) -> str:
+    for key in keys:
+        value = row.get(key, "")
+        if value and value not in {"-", "—"}:
+            return value
+    return ""
+
+
+def parse_relations(text: str) -> Tuple[List[Dict[str, str]], int]:
+    """Relations of a relationship table, plus the number of rows left out.
+
+    Current files use one 「主体 → 客体」 column; older ones use separate
+    主体/客体 or A/B columns. Any table with such a pair of columns counts.
+    """
+    relations = []  # type: List[Dict[str, str]]
+    skipped = 0
+    header = None  # type: Optional[List[str]]
+    for line in text.splitlines():
+        stripped = line.strip()
+        if not stripped.startswith("|"):
+            header = None
+            continue
+        row_cells = cells(stripped)
+        if header is None:
+            names = set(row_cells)
+            if (any("主体" in cell and "客体" in cell for cell in row_cells) or "关系方向" in names
+                    or (names & set(SOURCE_KEYS) and names & set(TARGET_KEYS))):
+                header = row_cells
+            continue
+        if not "".join(row_cells).strip("-: "):
+            continue
+        row = dict(zip(header, row_cells))
+        pair = next((value for key, value in row.items() if ("主体" in key and "客体" in key) or key == "关系方向"), "")
+        tokens = [part.strip() for part in ARROW_RE.split(pair)] if pair else []
+        names, arrows = tokens[0::2], tokens[1::2]
+        if len(names) < 2:
+            names = [exact(row, SOURCE_KEYS), exact(row, TARGET_KEYS)]
+            arrows = ["→"]
+        if len(names) < 2 or not all(names):
+            skipped += 1
+            continue
+        fields = {
+            "id": pick(row, "关系ID", "ID"),
+            "label": pick(row, "关系动作", "类型", "表面关系"),
+            "state": pick(row, "变化后状态", "真实关系", "表面关系"),
+            "trigger": pick(row, "触发事件", "触发"),
+        }
+        for index, arrow in enumerate(arrows):
+            ends = [(names[index], names[index + 1])]
+            if arrow in BOTH_WAYS:
+                ends.append((names[index + 1], names[index]))
+            for source, target in ends:
+                relations.append(dict(fields, source=source, target=target))
+    return relations, skipped
+
+
+def mermaid_text(value: str, placeholder: str) -> str:
+    value = re.sub(r"[`*]|%{2,}", "", value)
+    value = value.replace('"', "'").replace("|", "/").replace("\n", " ").strip()
+    return value or placeholder
+
+
+def latest_by_pair(relations: List[Dict[str, str]]) -> List[Dict[str, str]]:
+    latest = {}  # type: Dict[Tuple[str, str], Dict[str, str]]
+    for relation in relations:
+        latest[(relation["source"], relation["target"])] = relation
+    return list(latest.values())
+
+
+def evolutions(relations: List[Dict[str, str]]) -> List[Tuple[Tuple[str, str], List[Dict[str, str]]]]:
+    history = {}  # type: Dict[Tuple[str, str], List[Dict[str, str]]]
+    for relation in relations:
+        history.setdefault((relation["source"], relation["target"]), []).append(relation)
+    return [(pair, rows) for pair, rows in history.items() if len(rows) > 1]
+
+
+def render_markdown(book: str, relations: List[Dict[str, str]], png_note: str) -> str:
+    names = []  # type: List[str]
+    for relation in relations:
+        for name in (relation["source"], relation["target"]):
+            if name not in names:
+                names.append(name)
+    node = {name: "p%s" % index for index, name in enumerate(names, start=1)}
+    lines = ["# 人物关系图：%s" % book, "",
+             "> 由 角色/角色关系.md 生成；关系事实以该文件为准。%s" % png_note, "",
+             "## 核心人物关系", "", "```mermaid", "graph LR"]
+    lines.extend('    %s["%s"]' % (node[name], mermaid_text(name, "未命名")) for name in names)
+    for relation in latest_by_pair(relations):
+        label = relation["state"] or relation["label"] or "关系"
+        lines.append('    %s -->|"%s"| %s' % (node[relation["source"]], mermaid_text(label, "关系"),
+                                             node[relation["target"]]))
+    lines.extend(["```", ""])
+    for relation in latest_by_pair(relations):
+        label = relation["state"] or relation["label"] or "关系"
+        lines.append("- %s → %s：%s" % (relation["source"], relation["target"], label))
+    lines.extend(["", "## 关键关系演变", ""])
+    changes = evolutions(relations)
+    if not changes:
+        lines.append("本书没有记录到同一对人物前后变化的关系。")
+    for (source, target), rows in changes:
+        steps = []
+        for row in rows:
+            state = row["state"] or row["label"] or "关系"
+            steps.append("%s（%s）" % (state, row["trigger"]) if row["trigger"] else state)
+        lines.append("- %s → %s：%s" % (source, target, " → ".join(steps)))
+    return "\n".join(lines) + "\n"
+
+
+def png_text(value: str, placeholder: str = "") -> str:
+    """Text as drawn in the pictures: emoji and other symbols are left out."""
+    kept = "".join(char for char in value
+                   if char in BASE_TEXT or not unicodedata.category(char).startswith(("S", "C", "M")))
+    return re.sub(r"\s+", " ", kept).strip() or placeholder
+
+
+def font_face(path: str):  # noqa: ANN201 - matplotlib type
+    from matplotlib.ft2font import FT2Font
+    return FT2Font(path)
+
+
+def missing_glyphs(path: str, text: str) -> Optional[List[str]]:
+    """Characters of text the font (face 0) lacks; None when the font is unusable."""
+    if FALLBACK_FONT_RE.search(os.path.basename(path)):
+        return None
+    try:
+        font = font_face(path)
+        if FALLBACK_FONT_RE.search(str(getattr(font, "family_name", ""))):
+            return None
+        return sorted({char for char in text if not char.isspace() and not font.get_char_index(ord(char))})
+    except Exception:  # noqa: BLE001 - any font loading failure means "not usable"
+        return None
+
+
+def font_candidates() -> List[str]:
+    """Installed fonts worth trying, well-known Chinese fonts first."""
+    try:
+        from matplotlib import font_manager
+    except ImportError:
+        return []
+    by_name = {}  # type: Dict[str, List[str]]
+    for entry in font_manager.fontManager.ttflist:
+        paths = by_name.setdefault(entry.name, [])
+        if entry.fname not in paths:
+            paths.append(entry.fname)
+    candidates = [path for name in CJK_FONT_NAMES for path in by_name.get(name, [])]
+    if shutil.which("fc-list"):
+        try:
+            listed = subprocess.run(["fc-list", ":lang=zh", "file"], capture_output=True, text=True,
+                                    timeout=10, check=False).stdout
+        except (OSError, subprocess.SubprocessError):
+            listed = ""
+        candidates.extend(line.split(":", 1)[0].strip() for line in listed.splitlines())
+    return candidates
+
+
+def find_cjk_font(text: str) -> Tuple[Optional[str], List[str]]:
+    """(font that draws every character of text, []) or (None, characters no Chinese font draws)."""
+    text = text + BASE_TEXT
+    override = os.environ.get("STORY_ANALYZE_CHART_FONT")
+    if override == "none":
+        return None, []
+    fewest = None  # type: Optional[List[str]]
+    seen = set()
+    for path in [override] if override else font_candidates():
+        if not path or path in seen:
+            continue
+        seen.add(path)
+        missing = missing_glyphs(path, text)
+        if missing is None or any(char in BASE_TEXT for char in missing):
+            continue
+        if not missing:
+            return path, []
+        if fewest is None or len(missing) < len(fewest):
+            fewest = missing
+    return None, fewest or []
+
+
+def matplotlib_available() -> bool:
+    try:
+        import matplotlib  # noqa: F401
+    except ImportError:
+        return False
+    return True
+
+
+def short_label(value: str, limit: int = 10) -> str:
+    value = png_text(re.sub(r"[`*]|\[[^\]]*\]", "", value), "关系")
+    return value if len(value) <= limit else value[:limit - 1] + "…"
+
+
+def draw_png(relations: List[Dict[str, str]], font_path: str, root: Path) -> List[str]:
+    import matplotlib
+    matplotlib.use("Agg")
+    from matplotlib import font_manager
+    import matplotlib.pyplot as plt
+
+    font = font_manager.FontProperties(fname=font_path)
+    (root / CHART_DIR).mkdir(parents=True, exist_ok=True)
+    written = []
+    core = latest_by_pair(relations)
+    degree = {}  # type: Dict[str, int]
+    for relation in core:
+        for name in (relation["source"], relation["target"]):
+            degree[name] = degree.get(name, 0) + 1
+    # The most connected character (usually the protagonist) sits in the middle.
+    names = sorted(degree, key=lambda name: -degree[name])
+    ring = names[1:] if len(names) > 2 else names
+    positions = {name: (math.cos(2 * math.pi * index / len(ring)), math.sin(2 * math.pi * index / len(ring)))
+                 for index, name in enumerate(ring)}
+    if len(names) > 2:
+        positions[names[0]] = (0.0, 0.0)
+    size = min(14, 8 + 0.3 * len(names))
+    figure, axis = plt.subplots(figsize=(size, size))
+    axis.set_axis_off()
+    bend = 0.18  # curve both directions of a pair apart so their labels do not overlap
+    for relation in core:
+        (x1, y1), (x2, y2) = positions[relation["source"]], positions[relation["target"]]
+        axis.annotate("", xy=(x2, y2), xytext=(x1, y1),
+                      arrowprops={"arrowstyle": "->", "color": "#666666", "shrinkA": 22, "shrinkB": 22,
+                                  "connectionstyle": "arc3,rad=%s" % bend})
+        label_x = (x1 + x2) / 2 + bend * (y2 - y1) / 2
+        label_y = (y1 + y2) / 2 - bend * (x2 - x1) / 2
+        axis.text(label_x, label_y, short_label(relation["state"] or relation["label"] or "关系"),
+                  fontproperties=font, fontsize=9, ha="center", va="center", color="#333333",
+                  bbox={"boxstyle": "round", "fc": "white", "ec": "none", "alpha": 0.8})
+    for name, (x, y) in positions.items():
+        axis.text(x, y, png_text(name, "未命名"), fontproperties=font, fontsize=13, ha="center", va="center",
+                  bbox={"boxstyle": "round,pad=0.5", "fc": "#e8eef7", "ec": "#4a6fa5"})
+    axis.set_aspect("equal")
+    axis.set_xlim(-1.4, 1.4)
+    axis.set_ylim(-1.4, 1.4)
+    figure.savefig(str(root / CORE_PNG), dpi=150, bbox_inches="tight")
+    plt.close(figure)
+    written.append(CORE_PNG.as_posix())
+
+    changes = evolutions(relations)
+    if changes:
+        figure, axis = plt.subplots(figsize=(10, 1.2 + 0.8 * len(changes)))
+        axis.set_axis_off()
+        for row_index, ((source, target), rows) in enumerate(changes):
+            y = len(changes) - row_index
+            pair = "%s → %s" % (png_text(source, "未命名"), png_text(target, "未命名"))
+            axis.text(0, y, pair, fontproperties=font, fontsize=11, ha="left", va="center")
+            for step, row in enumerate(rows):
+                x = 3 + step * 2.2
+                if step:
+                    axis.annotate("", xy=(x - 0.6, y), xytext=(x - 1.6, y),
+                                  arrowprops={"arrowstyle": "->", "color": "#a58a4a"})
+                axis.text(x, y, png_text(row["state"] or row["label"], "关系"), fontproperties=font,
+                          fontsize=10, ha="center", va="center",
+                          bbox={"boxstyle": "round", "fc": "#f3efe6", "ec": "#a58a4a"})
+                if row["trigger"]:
+                    axis.text(x, y - 0.32, png_text(row["trigger"]), fontproperties=font, fontsize=8,
+                              ha="center", va="center", color="#666666")
+        axis.set_xlim(-0.2, 3 + 2.2 * max(len(rows) for _, rows in changes))
+        axis.set_ylim(0.3, len(changes) + 0.7)
+        figure.savefig(str(root / EVOLUTION_PNG), dpi=150, bbox_inches="tight")
+        plt.close(figure)
+        written.append(EVOLUTION_PNG.as_posix())
+    return written
+
+
+def render(root: Path, want_png: bool) -> Dict[str, object]:
+    source = root / RELATION_FILE
+    try:
+        text = source.read_text(encoding="utf-8-sig")
+    except OSError:
+        raise ChartError("relation_file_missing", "还没有 角色/角色关系.md，先整理人物关系，再生成关系图。")
+    relations, skipped = parse_relations(text)
+    if not relations:
+        raise ChartError("relation_table_empty", "角色/角色关系.md 里没有认得出的关系表（需要「主体 → 客体」一列），关系图没有生成。")
+    # The Markdown chart is the main result: write it before trying any picture.
+    atomic_write(root / CHART_MD, render_markdown(root.name, relations, "").encode("utf-8"))
+    png_written = []  # type: List[str]
+    messages = []  # type: List[str]
+    if skipped:
+        messages.append("角色/角色关系.md 里有 %s 行没认出是谁和谁的关系，没画进关系图。" % skipped)
+    if want_png:
+        drawn = "".join(png_text(value) for relation in relations
+                        for value in (relation["source"], relation["target"], relation["state"],
+                                      relation["label"], relation["trigger"]))
+        font_path, missing = find_cjk_font(drawn)
+        if not matplotlib_available():
+            messages.append(NO_MATPLOTLIB_MESSAGE)
+        elif not font_path:
+            messages.append(MISSING_GLYPH_MESSAGE % "".join(missing[:5]) if missing else NO_FONT_MESSAGE)
+        else:
+            try:
+                png_written = draw_png(relations, font_path, root)
+            except Exception:  # noqa: BLE001 - the Markdown chart is already written
+                messages.append(PNG_FAILED_MESSAGE)
+            if png_written:
+                atomic_write(root / CHART_MD,
+                             render_markdown(root.name, relations, "图片版见同目录的 PNG。").encode("utf-8"))
+    return {
+        "ok": True, "markdown": CHART_MD.as_posix(), "png": png_written,
+        "relations": len(relations), "skipped_rows": skipped,
+        "author_message": "".join(messages) or None,
+    }
+
+
+def main() -> int:
+    for stream in (sys.stdout, sys.stderr):
+        if hasattr(stream, "reconfigure"):
+            stream.reconfigure(encoding="utf-8")
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--root", required=True, type=Path, help="拆文库/{书名} 目录")
+    parser.add_argument("--png", action="store_true", help="also draw PNG pictures when a Chinese font exists")
+    args = parser.parse_args()
+    try:
+        if not args.root.is_dir():
+            raise ChartError("root_not_found", "没找到这本书的拆文目录，确认书名或路径。")
+        payload = render(args.root.resolve(), args.png)
+    except ChartError as exc:
+        payload = {"ok": False, "error": exc.code, "author_message": exc.author_message}
+    except (OSError, UnicodeError) as exc:
+        payload = {"ok": False, "error": "io_error", "detail": str(exc),
+                   "author_message": "关系图文件写入失败，检查拆文目录是否可写。"}
+    print(json.dumps(payload, ensure_ascii=False, indent=2))
+    return 0 if payload.get("ok") else 2
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

package/resources/skills/story-creation/upstream/skills/story-long-scan/SKILL.md

@@ -0,0 +1,193 @@
+---
+name: story-long-scan
+version: 1.0.0
+description: "长篇网文扫榜。分析起点、番茄、晋江等平台排行榜数据，提炼市场趋势与热门题材。触发方式：/story-long-scan、/长篇扫榜、「长篇什么火」「起点排行」。"
+metadata: {"openclaw":{"source":"https://github.com/zenstory-ai/oh-story-claudecode"}}
+---
+# story-long-scan：长篇网文扫榜
+
+你是网络小说市场分析师。你的任务是基于榜单样本识别长篇网文市场格局，并输出可执行的题材候选、风险阈值和验证动作。
+
+**核心信念：单本排名只提供线索；跨样本重复模式才算信号。** 排行榜只能证明样本存在；必须通过多榜单、多作品和近期数据判断需求强度。
+
+---
+
+## 核心哲学
+
+### 原则 1：扫榜看模式，别只看排名
+
+排名会波动，模式必须用重复样本验证。扫榜要提取：反复出现的题材、设定、套路、书名词和开篇卖点。单本上榜只能记为个例；同类样本达到可比数量后，才能标记为趋势候选。
+
+### 原则 2：流量型平台和付费型平台看的东西不同
+
+番茄看的是流量和完读率，起点看的是订阅和追读，晋江看的是收藏和积分。不同平台的成功标准不同，扫榜方法也不同。
+
+### 原则 3：扫榜的目的是找到你能写的爆款题材
+
+不按热度直接给结论。每个方向都要做项目可行性判断：素材储备、题材边界、篇幅承载、目标平台样本是否足够。
+
+---
+
+## 扫榜流程
+
+### Phase 1：确认平台和方向
+
+问用户：**「你想看哪个平台？（起点/番茄/晋江/七猫/刺猬猫/其他）有没有关注的题材方向？」**
+
+关键判断：
+- 用户已有方向 → 针对该方向做深度扫榜
+- 用户没有方向 → 做全榜概览 + 找趋势
+- 用户想跨平台比较 → 做平台对比分析
+
+---
+
+### Phase 2：确定数据来源
+
+**扫榜需要真实数据支撑。** 根据当前环境选择数据来源：
+
+| 优先级 | 模式 | 说明 | 何时用 |
+|--------|------|------|--------|
+| 1 | **脚本采集** | 直接抓取平台页面/SSR 数据，产出结构化文件 | 优先；起点默认不需要 Chrome |
+| 2 | **用户提供** | 用户粘贴榜单截图/文字/链接 | 用户已有数据时 |
+| 3 | **内置知识** | 基于知识库趋势数据做分析 | 无法联网、用户无数据时 |
+
+#### 脚本采集模式
+
+1. 只读所选平台那一份：[起点](references/platform-qidian.md) / [番茄](references/platform-fanqie.md) / [七猫](references/platform-qimao.md) / [晋江](references/platform-jjwxc.md) / [刺猬猫](references/platform-ciweimao.md)。里面有榜单网址、命令、字段、默认组合、故障排查和该平台的分析维度；需要浏览器态的平台先 `/browser-cdp` 启动 Chrome。
+2. 每次扫榜新建输出目录 `扫榜/{YYYYMMDD}/`（同日再扫加 `-2`），存本次榜单，不往旧目录追加（聚合会混进旧榜）；文件名 `{平台}{榜单}_{YYYYMMDD}.md`。脚本已做清洗和质量标注。
+3. 聚合：`node scripts/aggregate-rank.js {输出目录} --out {输出目录}/扫榜聚合.md`（小平台加 `--sparse 10`）。**主会话只读这份聚合**，不整份读原始榜单：题材分布、本数占比、热度与字数中位、标签热词、书名常见词、多榜重合、每题材前 3 本代表作。
+4. 看聚合的「采集情况」：标了问题（书名没解出来、没采到、只有书单）的榜单按平台参考排查重采；重采不了就在报告里说明结论不含它。同方向不到 15 本（小平台 10 本）标「少」，等同 `[数据稀疏]`。
+5. 要看原始条目（开篇卖点、新人设、新桥段）时按需抽样：`node scripts/aggregate-rank.js {输出目录} --sample {题材/标签/书名词} --n 5`，一次一个方向。
+
+#### 其他数据来源
+
+**用户提供：** 已有扫描结果文件 → 直接聚合；链接 → 用 WebFetch 抓取；粘贴/截图 → 整理成 `# {平台} · {榜单名}` 标题、每本一段 `### #{排名} {书名}` 加一行 `*作者 · 题材 · 连载中 · 120万字 · 3.5万收藏*`（热度写「数字+口径」；有标签再加 `**标签：** a、b`）存进输出目录，再同样聚合。
+
+**内置知识：** 加载 `references/genre-trends.md`，明确标注「以下分析基于历史趋势数据；未完成实时榜单校验前只能作为候选假设。」并列出需要复扫的榜单。
+
+---
+
+### Phase 3：数据分析
+
+以聚合结果为主，按所选平台参考里的「分析维度」看，再提取通用维度：
+
+1. **题材分布**：题材表的本数、占比与热度中位
+2. **新题材信号**：新书榜列、多榜重合；`扫榜/` 下上一次的日期目录里有 `扫榜聚合.md` 就对比
+3. **经典题材变化**：老牌题材的走势（上升/稳定/下降），没有上期数据就只写现状
+4. **字数**：字数分布与各题材字数中位
+5. **书名模式**：书名常见词 + 代表作书名
+6. **开头卖点**：标签热词 + 代表作简介
+7. **新元素**：对候选方向抽样原始条目，标出新的人物设定、开篇切入点、桥段套路
+
+---
+
+### Phase 4：输出扫榜报告
+
+报告写给作者：讲市场结论和能写的方向。脚本名、命令和采集状态码不进报告；某个榜没采到，就说一句「XX 榜这次没拿到（原因），结论不含它」。
+
+报告展示给作者，同时写进输出目录的 `选题决策.md`：文件不存在就先按 [references/topic-decision.md](references/topic-decision.md) 的模板写文件头，再把报告作为「扫榜结论」一节写入（已存在同名一节就整节替换）。作者只扫榜不选题，结论也已落盘。
+
+<!-- author-report -->
+```md
+## 扫榜结论：{平台名称}
+
+### 市场概况
+- 扫榜时间：{日期}
+- 核心发现：{一句话总结}
+
+### 题材热度排行
+| 排名 | 题材 | 榜上数量 | 趋势 | 代表作 |
+|------|------|----------|------|--------|
+| 1 | {题材} | {N本} | ↑/→/↓ | {书名} |
+
+### 新题材信号
+- {新出现或正在上升的题材，附依据}
+
+### 经典题材动态
+- {老牌题材的现状，附依据}
+
+### 新元素提取
+- 新人物设定：{新模式描述 + 代表作}
+- 新开篇切入点：{新切入点描述 + 代表作}
+- 新桥段/套路：{新桥段描述 + 代表作}
+
+### 关键数据洞察
+- 字数区间：上榜作品集中在 {X}-{Y} 万字
+- 书名特征：{命名模式总结}
+- 标签热词：{高频标签词}
+
+### 值得关注的方向
+1. {方向 + 为什么值得关注 + 可行性评估}
+2. {方向 + 为什么值得关注 + 可行性评估}
+3. {方向 + 为什么值得关注 + 可行性评估}
+
+### 一句话
+{犀利的总结}
+```
+
+---
+
+### Phase 5：选题决策
+
+把扫榜结果变成能直接用的选题建议，产出 `选题决策.md`。完整方法（选题四步 + 可行性判断 + 输出模板）见 [references/topic-decision.md](references/topic-decision.md)。
+
+**从文件接上：** 先读 `{outdir}/选题决策.md` 的「扫榜结论」和 `扫榜聚合.md`，不靠对话记忆；两者都没有就回 Phase 2。
+
+**如信息不足，向用户补齐项目条件：**「目标平台、已有素材、擅长题材/写作约束、计划篇幅是什么？」
+
+按 `topic-decision.md` 的选题四步产出 2-3 个推荐选题（能爆的原因 → 市场验证 → 差异化定位 → 可行性+失败风险+验证动作），写入**本次扫榜输出目录** `{outdir}/选题决策.md` 的「推荐选题」一节，保留「扫榜结论」。
+
+**硬规则：**
+- 可行性上限：背靠榜单标了 `[数据稀疏]`、聚合表标「少」或同方向样本 <15（小平台<10）⇒ 不许给"高"，强制降到"中" + 写明先验证；内置知识模式一律给"中"。
+- 不输出项目素材无法支撑的题材；不只看热度，必须给可行性和失败风险；不忽略平台调性差异（起点男频和晋江女频审美完全不同）。
+
+---
+
+## 平台特性速查
+
+| 平台 | 调性 | 核心指标 | 主力读者 | 适合类型 |
+|------|------|----------|----------|----------|
+| 起点中文网 | 男频为主，硬核爽文 | 追读率、月票 | 18-35 男性 | 玄幻、都市、科幻、游戏 |
+| 番茄小说 | 下沉市场，免费阅读 | 在读数、阅读榜排名 | 大众读者 | 脑洞、快节奏、强爽感 |
+| 晋江文学城 | 女频为主，精品路线 | 收藏、营养液、积分 | 16-30 女性 | 言情、纯爱、衍生 |
+| 七猫小说 | 下沉市场，免费阅读 | 热度、大热榜排名 | 大众读者 | 快节奏爽文 |
+| 刺猬猫 | 二次元、轻小说 | 追读 | 15-25 ACG | 同人、二次元、轻小说 |
+
+---
+
+## 流程衔接
+
+**流水线：** 长篇
+**位置：** 扫榜（第 1/3 步）
+
+| 时机 | 跳转到 | 命令 |
+|---|---|---|
+| 找到方向 | story-long-analyze | `/story-long-analyze` |
+| 直接开写 | story-long-write | `/story-long-write` |
+| 更适合短篇 | story-short-scan | `/story-short-scan` |
+
+## 参考资料
+
+按需加载以下文件：
+
+| 文件 | 何时加载 |
+|------|----------|
+| [references/topic-decision.md](references/topic-decision.md) | 「选题决策」：选题四步 + 可行性判断 + 选题决策.md 模板 |
+| [references/reader-profiling.md](references/reader-profiling.md) | 需要分析目标读者画像时 |
+| [references/genre-trends.md](references/genre-trends.md) | 查看题材趋势候选、切入约束和样本校验规则时 |
+| [references/publishing-guide.md](references/publishing-guide.md) | 平台适配+推荐机制校验+数据指标+简介设计 |
+| 五份平台参考（链接见「脚本采集模式」第 1 步） | 「确定数据来源」：只读所选平台那份——榜单网址、命令、字段、故障排查、平台分析维度 |
+| [scripts/aggregate-rank.js](scripts/aggregate-rank.js) | 把输出目录里的原始榜单聚合成短表（`--out` 落盘，`--sample` 抽原始条目，`--json` 机读），主会话只读它 |
+| [scripts/cdp-utils.js](scripts/cdp-utils.js) | CDP 公共工具函数（ab/sleep/evalJSON/safeStr/scrollLoad/getArg），各采集脚本共用 |
+| [scripts/fanqie-rank-scraper.js](scripts/fanqie-rank-scraper.js) | 番茄榜单采集，用法见番茄参考 |
+| [scripts/qidian-rank-scraper.js](scripts/qidian-rank-scraper.js) | 起点榜单采集，用法见起点参考 |
+| [scripts/qimao-rank-scraper.js](scripts/qimao-rank-scraper.js) | 七猫榜单采集，用法见七猫参考 |
+| [scripts/jjwxc-rank-scraper.js](scripts/jjwxc-rank-scraper.js) | 晋江榜单采集，用法见晋江参考 |
+| [scripts/ciweimao-rank-scraper.js](scripts/ciweimao-rank-scraper.js) | 刺猬猫榜单采集，用法见刺猬猫参考 |
+
+---
+
+## 语言
+
+- 跟随用户的语言回复，用户用什么语言就用什么语言回复
+- 中文回复遵循《中文文案排版指北》