#!/usr/bin/env python3
"""
合并各路调研底稿,生成采集阶段结束时给用户看的摘要表格。
扫描 references/research/ 下的 01-06 文件,统计每一路的来源数量、一手/二手占比、
关键发现,以及「和先验不一样的地方」各有几条(推翻 / 加深 / 新增)。
用法:
python3 merge_research.py <skill目录路径>
示例:
python3 merge_research.py .claude/skills/elon-musk-perspective
输出: 打印markdown格式的摘要表格到stdout
"""
import sys
import re
from pathlib import Path
# Windows 的 GBK 控制台打不出表格线和中文,统一成 UTF-8
for _s in (sys.stdout, sys.stderr):
try:
_s.reconfigure(encoding="utf-8", errors="replace")
except Exception:
pass
AGENTS = {
'01-writings': '著作',
'02-conversations': '对话',
'03-expression-dna': '表达',
'04-external-views': '他者',
'05-decisions': '决策',
'06-timeline': '时间线',
}
def count_sources(content: str) -> dict:
"""统计来源数量和一手/二手占比"""
# 计算URL数量作为来源数
urls = re.findall(r'https?://[^\s\)]+', content)
# 检测一手/二手标记
primary_markers = len(re.findall(r'一手|primary|本人|原文|原始|直接引用', content, re.IGNORECASE))
secondary_markers = len(re.findall(r'二手|secondary|转述|总结|评论|分析', content, re.IGNORECASE))
return {
'url_count': len(urls),
'unique_urls': len(set(urls)),
'primary_markers': primary_markers,
'secondary_markers': secondary_markers,
}
def extract_key_findings(content: str, max_items: int = 3) -> list[str]:
"""提取关键发现(取前几个二级标题或加粗项)"""
# 尝试提取##标题
headings = re.findall(r'^##\s+(.+)$', content, re.MULTILINE)
if headings:
return headings[:max_items]
# fallback: 提取加粗项
bolds = re.findall(r'\*\*(.+?)\*\*', content)
if bolds:
return bolds[:max_items]
# fallback: 取前3个非空行
lines = [l.strip() for l in content.split('\n') if l.strip() and not l.startswith('#')]
return [l[:50] + '...' if len(l) > 50 else l for l in lines[:max_items]]
def count_surprises(content: str) -> dict:
"""数「和先验不一样的地方」一段里推翻/加深/新增各几条。"""
m = re.search(r'^#+\s*.*(?:和先验不一样|先验).*$', content, re.MULTILINE)
if not m:
return None
tail = content[m.end():]
nxt = re.search(r'^#{1,2}\s+\S', tail, re.MULTILINE)
seg = tail[:nxt.start()] if nxt else tail
items = [l for l in seg.split('\n') if re.match(r'^\s*(?:[-*+]|\d+[.、)])\s+\S', l)]
if not items:
# 写成表格的:数表头和分隔线之外的行
rows = [l for l in seg.split('\n') if l.strip().startswith('|')]
items = [l for l in rows[2:] if not re.match(r'^\s*\|[\s:|-]+\|\s*$', l)]
return {
'推翻': sum(1 for l in items if '推翻' in l),
'加深': sum(1 for l in items if '加深' in l),
'新增': sum(1 for l in items if '新增' in l),
'total': len(items),
}
def find_contradictions(files: dict[str, str]) -> list[str]:
"""简单检测跨文件矛盾(同一关键词出现不同判断)"""
contradictions = []
# 检测「但是」「然而」「相反」「矛盾」等矛盾标记
for name, content in files.items():
matches = re.findall(r'(?:矛盾|相反|但实际上|然而.*?不同|争议).{0,100}', content)
for m in matches:
contradictions.append(f"{AGENTS.get(name, name)}: {m[:80]}")
return contradictions[:5] # 最多5条
def main():
if len(sys.argv) < 2:
print("用法: python3 merge_research.py <人物Skill目录>(含 references/research/ 的那一层)")
sys.exit(1)
skill_dir = Path(sys.argv[1])
research_dir = skill_dir / 'references' / 'research'
# 直接传了 research 目录也认
if not research_dir.exists() and skill_dir.name == 'research' and skill_dir.is_dir():
research_dir = skill_dir
if not research_dir.exists():
print(f"❌ 目录不存在: {research_dir}")
sys.exit(1)
files = {}
rows = []
surprises = {'推翻': 0, '加深': 0, '新增': 0, 'total': 0}
no_surprise_section = []
total_sources = 0
total_primary = 0
total_secondary = 0
missing = []
for key, label in AGENTS.items():
md_file = research_dir / f"{key}.md"
if not md_file.exists():
missing.append(label)
rows.append(f"│ {label:<12} │ {'❌ 缺失':<8} │ {'—':<24} │")
continue
content = md_file.read_text(encoding='utf-8')
files[key] = content
stats = count_sources(content)
findings = extract_key_findings(content)
sp = count_surprises(content)
if sp is None:
no_surprise_section.append(label)
else:
for k in surprises:
surprises[k] += sp[k]
total_sources += stats['unique_urls']
total_primary += stats['primary_markers']
total_secondary += stats['secondary_markers']
findings_str = ', '.join(findings) if findings else '—'
if len(findings_str) > 40:
findings_str = findings_str[:37] + '...'
rows.append(f"│ {label:<12} │ {stats['unique_urls']:<8} │ {findings_str:<24} │")
# 升级或补搜时新加的底稿(07-xxx.md 及以后):只把「和先验不一样的地方」计入
for md_file in sorted(research_dir.glob('[0-9][0-9]-*.md')):
if md_file.stem in AGENTS or md_file.stem.startswith('00-'):
continue
sp = count_surprises(md_file.read_text(encoding='utf-8'))
if sp is not None:
for k in surprises:
surprises[k] += sp[k]
# 矛盾检测
contradictions = find_contradictions(files)
# 输出
print("┌──────────────┬──────────┬──────────────────────────┐")
print("│ 调研分路 │ 来源数量 │ 关键发现 │")
print("├──────────────┼──────────┼──────────────────────────┤")
for row in rows:
print(row)
print("├──────────────┼──────────┼──────────────────────────┤")
primary_ratio = f"{total_primary}/{total_primary + total_secondary}" if (total_primary + total_secondary) > 0 else "未标记"
print(f"│ 总来源数 │ {total_sources:<8} │ 一手占比: {primary_ratio:<15} │")
if contradictions:
print(f"│ 矛盾点 │ {len(contradictions)}处 │ {contradictions[0][:24]:<24} │")
else:
print(f"│ 矛盾点 │ 0处 │ {'—':<24} │")
if missing:
print(f"│ 信息不足维度 │ {len(missing)}个 │ {', '.join(missing):<24} │")
else:
print(f"│ 信息不足维度 │ 无 │ {'—':<24} │")
print("└──────────────┴──────────┴──────────────────────────┘")
print(f"\n和先验不一样的地方:共 {surprises['total']} 条(推翻 {surprises['推翻']} / 加深 {surprises['加深']} / 新增 {surprises['新增']})")
if no_surprise_section:
print(f"⚠️ 这几路底稿末尾没有「和先验不一样的地方」一段: {', '.join(no_surprise_section)}")
if surprises['total'] < 8:
print("⚠️ 不到 8 条,说明还在复述常识:换更深的来源(长访谈原文、早期文章、当事人回忆、批评文章)再找")
if not (research_dir / '00-prior.md').exists():
print("⚠️ 没有 00-prior.md(先验快照),无从判断哪些是增量")
if not (research_dir / 'holdout.md').exists():
print("提醒:还没有 holdout.md。提炼之前挑 3–5 件一手材料封存,留作验证的考题")
# 总结
if total_sources < 10:
print("\n⚠️ 总来源数 <10,建议降低期望或补充调研")
if missing:
print(f"\n⚠️ 缺失维度: {', '.join(missing)},建议补充或在诚实边界中标注")
if __name__ == '__main__':
main()