文件内容
scripts/generate_volume2_chapters.py
#!/usr/bin/env python3
"""
Generate volume_02 chapters 2-11 with quality checks.
Uses DeepSeek API directly.
"""
import json, os, re, sys, time, urllib.request
from pathlib import Path
# ── Config ──────────────────────────────────────────────────
DEEPSEEK_KEY = os.environ.get("DEEPSEEK_API_KEY", "")
API_URL = "https://api.deepseek.com/v1/chat/completions"
MODEL = "deepseek-chat" # DeepSeek V4 Flash
NOVEL_DIR = Path(os.environ.get("NOVEL_DIR", "."))
CHAPTER_DIR = NOVEL_DIR / "chapters" / "volume_02"
PLAN_PATH = NOVEL_DIR / "plans" / "volume_02_plan.json"
BIBLE_PATH = NOVEL_DIR / "bible.json"
QC_SCRIPT = NOVEL_DIR / "review" / "quality_check.py"
# ── Load reference files ────────────────────────────────────
plan_data = json.loads(PLAN_PATH.read_text("utf-8"))
bible_data = json.loads(BIBLE_PATH.read_text("utf-8"))
CHAPTERS_INFO = {ch["number"]: ch for ch in plan_data["chapters"]}
# Read existing chapters
EXISTING_TEXTS = {}
for f in sorted(CHAPTER_DIR.glob("*.md")):
EXISTING_TEXTS[f.stem] = f.read_text("utf-8")
# ── System Prompt (蓝晶翻译体风格) ───────────────────────────
SYSTEM_PROMPT = """你是一个优秀的长篇中文网文作者，擅长「蓝晶翻译体」风格的西幻写作。
## 核心风格要求：
1. **翻译体语感**：大量使用破折号（——）做句中停顿和场景过渡，营造翻译体特有的叙事韵律
2. **纯中文写作**：不使用任何英文词汇
3. **对话格式**：对话不使用引号，用破折号+直角引号——「对话内容」——这样写
4. **第三人称有限视角**：主要从主角「理查德」的视角出发，偶尔可切换到「梅丽安」
5. **段落控制**：段落不宜太长，多用短段和空行分隔不同场景/动作/情绪
6. **描写风格**：暖丽细腻的场景描写 + 幽默诙谐的对话 + 人物鲜活的动作细节
7. **章末钩子**：每章结尾必须留悬念（问句/意外事件/紧急状况），让读者想看下一章
8. **章节字数**：每章4000-5000汉字
## 角色性格：
- 理查德（嘴硬乐观/天生蛮力/有正义感也有私心/会弹鲁特琴但弹得很烂）
- 梅丽安（外冷内热/博学冷静/背负家族仇恨/帝国魔法学院叛逃学者）
- 托德（朴实忠诚/爱喝酒/理查德发小）
## 剧情要求：
1. 推进角色弧：理查德从不熟练使用龙血到逐渐掌控
2. 每章至少包含：场景描写 + 对话互动 + 动作/战斗场面 + 情感觉知
3. 注意前后照应和伏笔埋设
4. 保持DND 5e战斗风格感"""
def call_deepseek(messages, temp=0.75, max_tokens=8192, timeout=300):
payload = json.dumps({
"model": MODEL,
"messages": messages,
"temperature": temp,
"max_tokens": max_tokens,
}).encode("utf-8")
req = urllib.request.Request(
API_URL, data=payload,
headers={
"Content-Type": "application/json",
"Authorization": f"Bearer {DEEPSEEK_KEY}",
})
with urllib.request.urlopen(req, timeout=timeout) as resp:
data = json.loads(resp.read().decode("utf-8"))
return data["choices"][0]["message"]["content"]
def count_chinese(text):
return len([c for c in text if '\u4e00' <= c <= '\u9fff'])
def get_continuity(chapter_num):
"""Get end of previous chapter or existing files"""
prev_num = chapter_num - 1
# Check if previous chapter was already written
for f in sorted(CHAPTER_DIR.glob(f"Ch{prev_num:02d}_*.md")):
text = f.read_text("utf-8")
return text[-1000:]
return ""
def get_chapter_text(chapter_num):
"""Get full text of an existing chapter"""
for f in sorted(CHAPTER_DIR.glob(f"Ch{chapter_num:02d}_*.md")):
return f.read_text("utf-8")
return ""
def run_quality_check(chapter_path):
import subprocess
result = subprocess.run(
["python3", str(QC_SCRIPT), str(chapter_path), "--json"],
capture_output=True, text=True, timeout=30
)
if result.returncode == 0 and result.stdout.strip():
return json.loads(result.stdout)
return None
def fix_p0_issues(text, chapter_num, issues):
p0 = [i for i in issues if i.get("severity") == "P0"]
if not p0:
return text
instructions = []
for i in p0[:5]:
desc = i.get("description", "")
suggestion = i.get("suggestion", "")
instructions.append(f"- 问题: {desc}\n 建议: {suggestion}")
fix_prompt = (
f"以下是第{chapter_num}章质量检查发现的P0问题，请修复。\n"
+ "\n".join(instructions) +
"\n\n输出完整的修复后章节文本，保留原标题、风格和叙事节奏。"
)
try:
fixed = call_deepseek([
{"role": "system", "content": "你是一个专业网文编辑。直接输出修复后的完整章节文本，不要额外说明。"},
{"role": "user", "content": f"{fix_prompt}\n\n原文：\n{text[:7000]}"},
], temp=0.3, max_tokens=8192, timeout=180)
if fixed and len(fixed) > 1000 and "修复" not in fixed[:50]:
return fixed
except Exception as e:
print(f" Fix failed: {e}")
return text
def generate_chapter(chapter_num):
info = CHAPTERS_INFO.get(chapter_num, {})
title = info.get("title", "")
summary = info.get("summary", "")
location = info.get("location", "")
char_focus = info.get("character_focus", [])
scenes = info.get("scenes", 3)
# Read continuity from previous chapter
prev_text = get_continuity(chapter_num)
# Bible characters
chars = bible_data.get("characters", {})
char_summary = "\n".join([
f"- {k}: {'、'.join(v.get('traits',[]))}；背景：{v.get('background','')[:80]}"
for k, v in chars.items()
])
# Build the generation prompt
user_prompt = (
f"# 第{chapter_num}章「{title}」\n\n"
f"## 本章概要\n{summary}\n\n"
f"## 场景数\n{scenes}个场景\n\n"
f"## 地点\n{location}\n\n"
f"## 重点角色\n{', '.join(char_focus)}\n\n"
f"## 上章结尾（连续性参考）\n{prev_text}\n\n"
f"## 角色设定\n{char_summary}\n\n"
f"---\n"
f"请写出完整的第{chapter_num}章「{title}」文本。标题格式：# 第X章 [标题]，然后空行、分隔线、正文。章末留悬念钩子。"
)
print(f"\n{'='*60}")
print(f"Generating Ch{chapter_num:02d}: {title}...")
sys.stdout.flush()
text = call_deepseek([
{"role": "system", "content": SYSTEM_PROMPT},
{"role": "user", "content": user_prompt},
], temp=0.75, max_tokens=8192, timeout=300)
# Ensure proper format
if not text.startswith("# 第"):
text = f"# 第{chapter_num}章 {title}\n\n---\n\n{text}"
cn_count = count_chinese(text)
text += f"\n\n【字数：约{cn_count}字】"
# Save
safe_title = re.sub(r'[\\/:*?"<>|]', '', title)
filename = f"Ch{chapter_num:02d}_{safe_title}.md"
filepath = CHAPTER_DIR / filename
filepath.write_text(text, "utf-8")
print(f" ✓ Saved: {filename} ({cn_count} Chinese chars)")
sys.stdout.flush()
# Quality check
time.sleep(1)
try:
report = run_quality_check(filepath)
if report:
issues = report.get("issues", [])
scores = report.get("scores", {})
p0 = [i for i in issues if i.get("severity") == "P0"]
p1 = [i for i in issues if i.get("severity") == "P1"]
p2 = [i for i in issues if i.get("severity") == "P2"]
print(f" QC: P0={len(p0)}, P1={len(p1)}, P2={len(p2)} | Scores: {scores.get('综合','?')}/100")
if p0:
print(f" 🔧 Fixing {len(p0)} P0 issues...")
sys.stdout.flush()
fixed = fix_p0_issues(text, chapter_num, issues)
if fixed and fixed != text and count_chinese(fixed) > 1000:
filepath.write_text(fixed, "utf-8")
cn2 = count_chinese(fixed)
print(f" ✓ Re-saved ({cn2} chars)")
time.sleep(1)
r2 = run_quality_check(filepath)
if r2:
p0a = len([i for i in r2.get("issues", []) if i.get("severity") == "P0"])
print(f" QC after fix: P0={p0a}")
else:
print(f" ⚠ Fix produced invalid result, keeping original")
else:
print(f" QC: No report")
except Exception as e:
print(f" QC error: {e}")
return filepath
def main():
chapters = list(range(2, 12))
results = {}
for cn in chapters:
try:
path = generate_chapter(cn)
results[cn] = {"status": "ok", "path": str(path)}
except Exception as e:
print(f" ✗ Ch{cn:02d} FAILED: {e}")
import traceback
traceback.print_exc()
results[cn] = {"status": "error", "error": str(e)}
time.sleep(3)
print(f"\n{'='*60}")
print("COMPLETION REPORT:")
print(f"{'='*60}")
total_chars = 0
for cn in chapters:
r = results.get(cn, {})
if r.get("status") == "ok":
p = Path(r["path"])
if p.exists():
t = p.read_text("utf-8")
c = count_chinese(t)
total_chars += c
print(f" Ch{cn:02d}: {p.name} — {c}字 ✅")
else:
print(f" Ch{cn:02d}: File not found ❌")
else:
print(f" Ch{cn:02d}: {r.get('error')} ❌")
print(f"\n Total: {total_chars} Chinese characters across {len(chapters)} chapters")
# Save manifest
manifest = {
"volume": 2,
"chapters_written": [cn for cn in chapters if results.get(cn, {}).get("status") == "ok"],
"total_chinese_chars": total_chars,
"results": {str(k): v["status"] for k, v in results.items()},
}
manifest_path = CHAPTER_DIR / ".generation_manifest.json"
manifest_path.write_text(json.dumps(manifest, ensure_ascii=False, indent=2), "utf-8")
print(f"\nManifest saved: {manifest_path}")
print(f"\nDone.")
if __name__ == "__main__":
main()