Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →AI Berkshire skill: 去劣筛选：7条指标快速排除非一流公司. Source: skills/quality-screen.md.
.claude/skills/xbtlin-quality-screen/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-21 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-10 | ✓→✗ | ▼ Worse | 8% | 0% |
| case-11 | ✓→✗ | ▼ Worse | 7% | 0% |
| case-12 | ✓→✗ | ▼ Worse | -29% | 0% |
| case-02 | ✓→✗ | ▼ Worse | -40% | 0% |
This skill is generated from skills/quality-screen.md so Claude Code and Codex users share one canonical workflow.
$ARGUMENTS as the user's request in the current Codex thread.tools/ in this repository. Prefer running commands from the repository root with paths like python3 tools/financial_rigor.py ...; if the current thread starts outside the repo, locate the actual checkout path first instead of assuming a fixed home-directory path.date command to confirm today's date; treat it as the baseline for "latest" data and state the data cutoff date in the report header. Never assume the current date from training data.AGENTS.md: cross-check financial data, use exact arithmetic tools for valuation/math, and clearly label uncertainty and source gaps.对 $ARGUMENTS 执行去劣指标筛选，快速排除不符合一流公司标准的标的。
支持输入格式：
| 输入方式 | 示例 | 说明 | |---------|------|------| | 个股 | 腾讯, 美团, 英伟达 | 逐家筛选 | | 行业 | 中国啤酒行业 全球云计算 港股运动品牌 | 先搜索该行业主要上市公司（10-20家），再逐家筛选 | | 市场/指数 | 恒生指数成分股 沪深300 纳斯达克100 | 拉取成分股列表，逐家筛选 | | 主题 | 中国高股息50强 全球AI算力链 | 先搜索主题相关公司，再逐家筛选 |
行业/市场/主题模式下，输出额外包含：通过率统计、行业内排名、板块对比总结。
| # | 指标 | 排除条件 | 衡量的是什么 | |---|------|---------|-------------| | 1 | 10年平均ROE | < 8% | 资本效率——股东的钱能不能跑赢机会成本 | | 2 | 5年累计自由现金流 | 为负 | 真金白银——利润是不是"纸面富贵" | | 3 | 利息覆盖倍数（EBIT/利息） | < 2倍 | 偿债安全——还利息的能力 | | 4 | 长期毛利率 | < 15% | 定价权——产品/服务有没有差异化 | | 5 | 经营现金流 / 净利润（5年均值） | < 0.7 | 利润质量——赚到的利润能不能收回现金 | | 6 | 长期净利率 | < 5% | 抗风险能力——收入波动时利润是否归零 | | 7 | 5年总股本膨胀 | > 20%（非并购原因） | 股东利益——管理层是否在稀释你的权益 |
如果同时满足以下3个条件，可豁免第1条ROE不达标：
逻辑：高毛利率+现金流转正说明商业模式没问题，ROE低只是因为还在投入期。典型案例：美团。
如果同时满足以下2个条件，可豁免第6条净利率不达标：
逻辑：毛利率高说明有定价权，净利率低是战略选择（再投资）而非能力缺失。典型案例：亚马逊。
如果同时满足以下3个条件，可豁免第4条毛利率和第6条净利率不达标：
逻辑：有些一流公司的利润不藏在毛利率里，而是藏在会员费、周转效率或平台抽成中。它们的毛利率和净利率天然很低，但ROE极高说明资本效率一流。典型案例：Costco（毛利率12%、净利率2.5%，但ROE 25%+、会员续费率90%+）。
模式判断：
对每家公司确定全称、代码、交易所。
为每家公司启动独立后台Agent，使用 WebSearch 搜索以下数据：
数据来源优先级：公司年报 > 券商研报 > 财经数据平台
对每家公司，逐条检验7个指标：
如果触犯某条，检查是否满足对应豁免条件。
markdown# 去劣筛选结果 **筛选日期**：{当天日期} **公司数量**：{N}家 ## 汇总表 | 公司 | ①ROE | ②FCF | ③利息覆盖 | ④毛利率 | ⑤OCF/NI | ⑥净利率 | ⑦稀释 | 结果 | |------|------|------|----------|---------|---------|---------|-------|------| | xxx | ✅ 24% | ✅ | ✅ | ✅ 56% | ✅ | ✅ 30% | ✅ | **通过** | | yyy | ❌ 3% | ❌ | ❌ | ✅ 20% | ✅ | ❌ 2% | ✅ | **排除** | | zzz | ⚠️→✅ | ✅ | ✅ | ✅ 35% | ✅ | ⚠️→✅ | ✅ | **豁免通过** | ## 通过的公司（N家） [列表] ## 排除的公司（N家） | 公司 | 触犯指标 | 具体数据 | 排除理由 | |------|---------|---------|---------| ## 豁免通过的公司（N家） | 公司 | 豁免条款 | 具体数据 | 豁免理由 | |------|---------|---------|---------| ## 边界争议（如有） [对处于阈值附近的公司做补充说明] ## 板块总结（行业/市场模式专用） **通过率**：{通过数}/{总数} = {百分比} **行业质量判断**：[根据通过率给出行业整体质量评价] | 质量分层 | 公司 | 共同特征 | |---------|------|---------| | 一流（全过+高ROE） | xxx, yyy | ... | | 合格（全过但指标平庸） | aaa, bbb | ... | | 淘汰 | ccc, ddd | ... | **行业选股结论**：[一句话总结该行业值不值得深挖，最值得关注的2-3家是谁]
这套指标能排除"确定不好"的公司，但通过筛选不等于"确定好"。通过的公司仍需进一步研究：
去劣是第一步，不是最后一步。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-08 | fail→fail | 22,800 | 8,052 | -65% | 1 | 1 | 0% | 3,112 | 2,954 | -5% | 0 | 0 | — |
case-09 | fail→fail | 20,753 | 22,064 | +6% | 1 | 1 | 0% | 3,098 | 2,920 | -6% | 0 | 0 | — |
case-10 | pass→fail | 14,373 | 22,679 | +58% | 1 | 1 | 0% | 2,364 | 2,560 | +8% | 0 | 0 | — |
case-11 | pass→fail | 14,009 | 5,041 | -64% | 1 | 1 | 0% | 2,410 | 2,588 | +7% | 0 | 0 | — |
case-12 | pass→fail | 23,592 | 4,730 | -80% | 1 | 1 | 0% | 3,920 | 2,764 | -29% | 0 | 0 | — |
case-01 | fail→fail | 25,037 | 5,167 | -79% | 1 | 1 | 0% | 3,934 | 2,570 | -35% | 0 | 0 | — |
case-02 | pass→fail | 48,864 | 19,342 | -60% | 1 | 1 | 0% | 4,335 | 2,586 | -40% | 0 | 0 | — |
case-03 | fail→fail | 31,070 | 14,695 | -53% | 1 | 1 | 0% | 4,745 | 2,714 | -43% | 0 | 0 | — |
case-04 | fail→fail | 24,118 | 36,300 | +51% | 1 | 1 | 0% | 3,423 | 2,543 | -26% | 0 | 0 | — |
case-05 | pass→fail | 16,974 | 7,339 | -57% | 1 | 1 | 0% | 2,386 | 2,612 | +9% | 0 | 0 | — |
case-06 | pass→fail | 19,730 | 6,455 | -67% | 1 | 1 | 0% | 2,833 | 2,785 | -2% | 0 | 0 | — |
case-07 | fail→fail | 26,971 | 5,152 | -81% | 1 | 1 | 0% | 4,085 | 2,656 | -35% | 0 | 0 | — |
case-13 | pass→pass | 17,908 | 11,707 | -35% | 1 | 1 | 0% | 3,000 | 4,283 | +43% | 0 | 0 | — |
case-14 | fail→fail | 14,437 | 4,682 | -68% | 1 | 1 | 0% | 2,458 | 2,612 | +6% | 0 | 0 | — |
case-15 | pass→fail | 26,600 | 6,130 | -77% | 1 | 1 | 0% | 4,681 | 2,704 | -42% | 0 | 0 | — |
case-16 | pass→fail | 30,057 | 4,730 | -84% | 1 | 1 | 0% | 5,573 | 2,544 | -54% | 0 | 0 | — |
case-17 | fail→fail | 18,513 | 24,141 | +30% | 1 | 1 | 0% | 3,207 | 2,607 | -19% | 0 | 0 | — |
case-18 | pass→fail | 29,154 | 4,858 | -83% | 1 | 1 | 0% | 6,191 | 2,723 | -56% | 0 | 0 | — |
case-19 | pass→pass | 25,475 | 44,428 | +74% | 1 | 1 | 0% | 3,630 | 8,556 | +136% | 0 | 0 | — |
case-20 | pass→fail | 19,942 | 5,189 | -74% | 1 | 1 | 0% | 2,978 | 2,629 | -12% | 0 | 0 | — |
case-21 | fail→pass | 15,170 | 7,751 | -49% | 1 | 1 | 0% | 2,513 | 3,593 | +43% | 0 | 0 | — |
case-22 | pass→pass | 19,797 | 21,463 | +8% | 1 | 1 | 0% | 2,604 | 4,548 | +75% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 4 counted toward the lift figure. The other 18 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -41 percentage points is the difference between those two pass rates over the 4 comparable cases. 14 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.