Financial Report Analyzer · 财报智能解读

Pull every metric that matters from a 200-page annual report in under 60 seconds, with traceable line-item provenance, peer benchmarking, and red-flag detection.

60 秒从 200 页年报抽出全部关键指标，每个数字可溯源到原文，附同行对标与风险信号识别。

🎯 When to Use · 何时使用

Trigger keywords (中文): 分析财报、解读年报、半年报解读、利润表分析、资产负债表、现金流量表、ROE 分析、毛利率、商誉风险、关联交易、同行对比、行业对标、财报对比、招股说明书

Trigger keywords (EN): parse 10-K, analyze annual report, extract financials, ratio analysis, peer comparison, red flag detection, earnings quality

Supported inputs:

• PDF（A 股年报/半年报、招股书、Form 10-K/10-Q/20-F、IFRS annual reports）
• XBRL / iXBRL
• Excel 财务摘要（巨潮、东方财富、Wind 导出格式）
• 纯文本财务披露

Do NOT use when:

• User asks for stock price prediction or investment advice (this skill is descriptive analysis only)
• Input is a single number or one-line text
• Input is non-financial content

📋 Output Sections · 输出结构

The skill always produces a 5-section report:

🔄 Analysis Protocol · 分析流程

Step 1: Document ingestion · 文档摄入

python3 scripts/ingest.py --input \x3Cpdf-or-xbrl> --out /tmp/raw_extract.json

• Auto-detect report type (10-K / 年报 / 半年报 / 招股书)
• Identify reporting period and accounting standard
• Extract text + tables with layout preservation (uses bundled pdfplumber)

Step 2: Table normalization · 报表归一化

scripts/normalize_statements.py maps raw line items to a standardized chart of accounts (knowledge/coa_master.csv), handling:

• 中文/英文双语科目名
• CAS ↔ IFRS ↔ US-GAAP 科目对照
• 单位归一（元/千元/百万元/亿元 → 元）
• 期间对齐（年度/季度/累计）

Step 3: KPI & ratio computation · 指标计算

python3 scripts/compute_ratios.py --input normalized.json --out ratios.json

Computes 40+ standard ratios documented in knowledge/ratio_definitions.md, with formulas matching CFA Institute conventions and CSRC disclosure requirements.

Step 4: Red-flag detection · 红旗扫描

scripts/detect_red_flags.py runs 18 heuristic checks, including:

• 应收账款增速显著高于营收增速 (Receivables growth >> Revenue growth)
• 经营性现金流持续低于净利润 (OCF/NI \x3C 0.5 for 3+ years)
• 商誉/净资产 > 30%
• 关联方交易占营收 > 20%
• 审计意见非"标准无保留"
• 频繁更换会计师事务所
• 存货周转率突降
• 期间费用大幅波动
• 应付账款异常下降
• 在建工程长期挂账
• 其他应收款异常增长
• 大额非经常性损益占比
• 短期借款激增
• 商誉减值历史
• 大股东质押比例
• 控股股东净利润占比异常
• 利息覆盖倍数 \x3C 1.5
• Z-score \x3C 1.8

Each flag is documented with severity (🟢🟡🔴), evidence span, and recommended follow-up question.

Step 5: Peer benchmarking · 同行对标

If FINREPORT_PEER_DB is set (or built-in industry medians available), produces percentile ranks. Otherwise outputs absolute values with industry context notes.

Step 6: Report rendering · 报告渲染

python3 scripts/render_report.py --analysis ratios.json --flags flags.json --format md|json|html

For web channels with web-prism-artifact available, the skill automatically emits a structured artifact for inline rendering.

📤 Output Format · 输出格式

JSON structure:

{
  "company": { "name": "...", "ticker": "...", "industry": "..." },
  "period": { "fiscal_year": 2024, "reporting_basis": "CAS" },
  "executive_summary": "...",
  "kpis": { "revenue": {...}, "net_income": {...}, ... },
  "ratios": { "roe": {...}, "gross_margin": {...}, ... },
  "red_flags": [ { "code": "RF03", "severity": "🟡", "title": "...", "evidence": "...", "page": 47 } ],
  "peer_benchmark": { "industry": "...", "rankings": {...} },
  "provenance": { /* page+line offset for every number */ }
}

For human consumption, --format md produces a Markdown report with embedded tables and red-flag callouts.

⚠️ Safety & Compliance · 安全合规

1. Descriptive only, not predictive — never output buy/sell/hold recommendations or stock price targets.
2. Source attribution — every number includes {page, table, row} pointing back to the source document.
3. No silent imputation — missing data is explicitly reported, never extrapolated.
4. Red flags are signals, not verdicts — output always frames flags as "warrant further investigation," not "company is fraudulent."
5. Audit log — full extraction log saved to \x3Coutput>.audit.log for compliance.

本技能仅做描述性分析，绝不输出买卖评级或股价预测；所有数字可溯源；缺失数据明确标注不做估算；红旗信号仅作为"需进一步核查"的提示，不作定性结论。

🚀 Usage Examples · 使用示例

Example 1: Single annual report analysis

python3 scripts/run_pipeline.py --input 600519_2024_annual.pdf --output report.json
python3 scripts/render_report.py --input report.json --format md > 茅台2024分析.md

Example 2: YoY comparison

python3 scripts/compare_periods.py \
  --current 2024_annual.pdf \
  --prior 2023_annual.pdf \
  --output yoy_comparison.md

Example 3: Peer benchmark (white-liquor industry)

python3 scripts/peer_benchmark.py \
  --target 600519_2024_annual.pdf \
  --peers 000858,000568,000799 \
  --output peer_report.md

Example 4: Red-flag scan only (fast mode)

python3 scripts/run_pipeline.py --input report.pdf --mode red-flag-only --output flags.json

🧪 Testing · 测试

cd tests && python3 -m pytest -v

Test fixtures include:

• 10 real anonymized A-share annual reports (5 industries)
• 3 Form 10-K samples
• Edge cases: restated financials, going-concern qualifications, mid-year auditor change

📚 References · 参考资料

• CSRC《公开发行证券的公司信息披露内容与格式准则第 2 号》
• IFRS Standards: https://www.ifrs.org/issued-standards/
• US GAAP: https://asc.fasb.org/
• Beneish M-Score, Altman Z-Score original papers
• CFA Institute Financial Statement Analysis curriculum

🏷️ Tags · 标签

finance accounting 10-K annual-report ratio-analysis red-flag IFRS GAAP CAS 财报 年报 财务分析