mcn-short-video/.workbuddy/skills/nuwa-skill-main/scripts/quality_check.py

#!/usr/bin/env python3
"""
自动检查生成的SKILL.md是否通过Phase 4质量标准。
对照通过标准表格逐项检查，输出通过/不通过和具体原因。

用法:
    python3 quality_check.py <SKILL.md路径>

示例:
    python3 quality_check.py .claude/skills/elon-musk-perspective/SKILL.md
"""

import sys
import re
from pathlib import Path


def check_mental_models(content: str) -> tuple[bool, str]:
    """检查心智模型数量（3-7个）"""
    # 匹配 ### 模型N: 或 ### N. 等模式
    models = re.findall(r'^###\s+(?:模型|Model|心智模型)\s*\d', content, re.MULTILINE)
    if not models:
        # fallback: 数「### 」开头的行在心智模型section中
        in_section = False
        count = 0
        for line in content.split('\n'):
            if re.match(r'^##\s+.*心智模型|Mental Model', line, re.IGNORECASE):
                in_section = True
                continue
            if in_section and re.match(r'^##\s+', line) and '心智模型' not in line:
                break
            if in_section and re.match(r'^###\s+', line):
                count += 1
        if count > 0:
            passed = 3 <= count <= 7
            return passed, f"{count}个心智模型 {'✅' if passed else '❌ (应为3-7个)'}"

    count = len(models)
    if count == 0:
        return False, "未检测到心智模型section"
    passed = 3 <= count <= 7
    return passed, f"{count}个心智模型 {'✅' if passed else '❌ (应为3-7个)'}"


def check_limitations(content: str) -> tuple[bool, str]:
    """检查每个模型是否有局限性"""
    has_limitation = bool(re.search(r'局限|失效|不适用|盲区|limitation|blind spot', content, re.IGNORECASE))
    return has_limitation, "有局限性标注 ✅" if has_limitation else "❌ 未找到局限性描述"


def check_expression_dna(content: str) -> tuple[bool, str]:
    """检查表达DNA辨识度"""
    dna_section = bool(re.search(r'表达DNA|Expression DNA|表达风格', content, re.IGNORECASE))
    if not dna_section:
        return False, "❌ 未找到表达DNA section"

    # 检查是否有具体的风格描述（句式、词汇等）
    style_markers = len(re.findall(r'句式|词汇|语气|幽默|节奏|确定性|引用|口头禅', content))
    passed = style_markers >= 3
    return passed, f"表达DNA特征: {style_markers}项 {'✅' if passed else '❌ (应≥3项)'}"


def check_honest_boundary(content: str) -> tuple[bool, str]:
    """检查诚实边界（至少3条）"""
    # 找诚实边界section
    boundary_match = re.search(r'(?:##\s+.*诚实边界|## Honest Boundary)(.*?)(?=\n##\s|\Z)', content, re.DOTALL | re.IGNORECASE)
    if not boundary_match:
        return False, "❌ 未找到诚实边界section"

    boundary_text = boundary_match.group(1)
    # 计算列表项
    items = re.findall(r'^[-*]\s+', boundary_text, re.MULTILINE)
    count = len(items)
    passed = count >= 3
    return passed, f"诚实边界: {count}条 {'✅' if passed else '❌ (应≥3条)'}"


def check_tensions(content: str) -> tuple[bool, str]:
    """检查内在张力（至少2对）"""
    tension_markers = len(re.findall(r'张力|矛盾|tension|paradox|一方面.*另一方面|既.*又', content, re.IGNORECASE))
    passed = tension_markers >= 2
    return passed, f"内在张力: {tension_markers}处 {'✅' if passed else '❌ (应≥2处)'}"


def check_primary_sources(content: str) -> tuple[bool, str]:
    """检查一手来源占比"""
    # 找调研来源section
    source_section = re.search(r'(?:##\s+.*来源|## Source|## Reference)(.*?)(?=\n##\s|\Z)', content, re.DOTALL | re.IGNORECASE)
    if not source_section:
        return True, "未找到来源section（跳过检查）"

    source_text = source_section.group(1)
    primary = len(re.findall(r'一手|primary|本人著作|原始', source_text, re.IGNORECASE))
    secondary = len(re.findall(r'二手|secondary|转述|评论', source_text, re.IGNORECASE))
    total = primary + secondary
    if total == 0:
        return True, "未标记来源类型（跳过检查）"

    ratio = primary / total
    passed = ratio > 0.5
    return passed, f"一手来源占比: {primary}/{total} ({ratio:.0%}) {'✅' if passed else '❌ (应>50%)'}"


def check_terminology_adaptation(content: str) -> tuple[bool, str]:
    """检查面向用户交付物中是否有内部术语残留。
    检查section标题和正文，确保没有学术黑话。"""
    # SKILL.md文件是给AI agent用的，保留术语是合理的
    is_skill_file = bool(re.search(r'^---\s*\nname:\s*', content, re.MULTILINE))
    if is_skill_file:
        return True, "SKILL.md文件（给AI agent用），术语保留 ✅"

    # 完整的禁用术语列表（含正文检查）
    jargon_terms = [
        '心智模型', '决策启发式', '表达DNA', '诚实边界',
        '内在张力', '智识谱系', '反模式',
        '三重验证', '跨域复现', '生成力', '排他性'
    ]

    # 检查section标题（## 开头的行）
    section_headers = re.findall(r'^##\s+.*$', content, re.MULTILINE)
    jargon_in_headers = []
    for header in section_headers:
        for term in jargon_terms:
            if term in header:
                jargon_in_headers.append(f'标题"{header.strip()}"含"{term}"')

    # 检查正文中的术语残留（排除在代码块/引用块中的情况）
    jargon_in_body = []
    lines = content.split('\n')
    in_code_block = False
    for i, line in enumerate(lines, 1):
        if line.strip().startswith('```'):
            in_code_block = not in_code_block
            continue
        if in_code_block:
            continue
        for term in jargon_terms:
            if term in line:
                # 排除在"禁用词列表"中提到术语本身的情况
                if '禁用' in line or '不出现' in line or '❌' in line:
                    continue
                jargon_in_body.append(f'第{i}行含"{term}": {line.strip()[:60]}...')

    all_issues = jargon_in_headers + jargon_in_body[:5]  # 最多显示5条正文问题

    if all_issues:
        return False, f"❌ 术语残留: {'; '.join(all_issues[:5])}"
    return True, "面向用户交付物，无术语残留 ✅"


def check_persona_card_format(content: str) -> tuple[bool, str]:
    """检查短视频人设卡是否符合12段标准结构。
    当文件名包含'人设'或'账号卡片'时触发此检查。"""
    # 判断是否是人设卡
    if not re.search(r'人设|账号卡片|persona', content[:200], re.IGNORECASE):
        return True, "非人设卡文件（跳过12段结构检查）"

    expected_sections = [
        '账号核心定位', '账号内容基因', '角色原型库', '情绪结构',
        '视听风格', '五大内容打法', '内容规则', '系列内容规划',
        '差异化壁垒', '红线清单', '当前需解决的问题', '人设底线'
    ]

    found_sections = []
    missing_sections = []
    for section in expected_sections:
        if re.search(rf'##\s+.*{section}', content):
            found_sections.append(section)
        else:
            missing_sections.append(section)

    if missing_sections:
        return False, f"❌ 缺少段落: {', '.join(missing_sections)}"
    return True, f"12段结构完整 ✅ ({len(found_sections)}/12)"

# 检查学术化表达
def check_academic_writing(content: str) -> tuple[bool, str]:
    """检查正文中是否有学术化表达（人设卡场景）。"""
    # SKILL.md文件不检查
    is_skill_file = bool(re.search(r'^---\s*\nname:\s*', content, re.MULTILINE))
    if is_skill_file:
        return True, "SKILL.md文件（跳过学术化检查）"

    academic_patterns = [
        '认知框架', '降维打击', '范式', '策略性', '结构化',
        '方法论体系', '底层逻辑架构', '思维范式', '认知操作系统'
    ]

    found = []
    lines = content.split('\n')
    in_code_block = False
    for i, line in enumerate(lines, 1):
        if line.strip().startswith('```'):
            in_code_block = not in_code_block
            continue
        if in_code_block:
            continue
        for pattern in academic_patterns:
            if pattern in line:
                found.append(f'第{i}行: "{pattern}"')

    if found:
        return False, f"❌ 学术化表达: {'; '.join(found[:3])}"
    return True, "无学术化表达 ✅"


def main():
    if len(sys.argv) < 2:
        print("用法: python3 quality_check.py <文件路径>")
        sys.exit(1)

    skill_path = Path(sys.argv[1])
    if not skill_path.exists():
        print(f"❌ 文件不存在: {skill_path}")
        sys.exit(1)

    content = skill_path.read_text(encoding='utf-8')

    # 判断文件类型
    is_skill_file = bool(re.search(r'^---\s*\nname:\s*', content, re.MULTILINE))
    is_persona_card = bool(re.search(r'人设|账号卡片|persona', content[:500], re.IGNORECASE))

    if is_skill_file:
        # SKILL.md文件：运行全部检查（内部术语是合理的）
        checks = [
            ("心智模型数量", check_mental_models),
            ("模型局限性", check_limitations),
            ("表达DNA辨识度", check_expression_dna),
            ("诚实边界", check_honest_boundary),
            ("内在张力", check_tensions),
            ("一手来源占比", check_primary_sources),
            ("术语通俗化", check_terminology_adaptation),
        ]
        file_type = "SKILL.md（AI agent Skill文件）"
    elif is_persona_card:
        # 人设卡：只运行面向用户的检查项
        checks = [
            ("术语通俗化", check_terminology_adaptation),
            ("人设卡12段结构", check_persona_card_format),
            ("学术化表达", check_academic_writing),
        ]
        file_type = "人设卡（面向编导/创作者）"
    else:
        # 其他面向用户的文档
        checks = [
            ("术语通俗化", check_terminology_adaptation),
            ("学术化表达", check_academic_writing),
        ]
        file_type = "面向用户文档"

    print(f"质量检查: {skill_path.name}")
    print(f"文件类型: {file_type}")
    print("=" * 50)

    passed_count = 0
    total = len(checks)

    for name, check_fn in checks:
        passed, detail = check_fn(content)
        status = "✅ PASS" if passed else "❌ FAIL"
        print(f"  {name:<12} {status}  {detail}")
        if passed:
            passed_count += 1

    print("=" * 50)
    print(f"结果: {passed_count}/{total} 通过")

    if passed_count == total:
        print("🎉 全部通过，可以交付")
    elif passed_count >= total - 1:
        print("⚠️ 基本通过，建议修复不通过项后交付")
    else:
        print("❌ 多项不通过，建议回到Phase 2迭代")

    sys.exit(0 if passed_count == total else 1)


if __name__ == '__main__':
    main()