菜单
Use when raw close, adjusted close, cash-dividend, and split-factor data must be checked for invalid event inputs, unexplained price jumps, or adjusted-return mismatches before equity research or backtesting.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/XinAloha/skills --skill corporate-action-adjustment-auditor命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use when an agent needs a standard cross-sectional long-only backtest protocol with T+1 open execution, top-bucket equal weighting, fees, limit-up or limit-down exclusions, benchmark comparison, NAV curves, drawdown, IC, and diagnostic charts.
Brinson-Fachler / BHB sector attribution with interaction, concentration (HHI), top contributors, and optional Carino multi-period linking. Use to explain active return as allocation vs selection.
Run and explain cross-sectional trading-factor backtests on schema-compatible factor signals and market data. Use when Codex needs to validate backtest inputs; calculate returns, NAV, drawdown, holdings, turnover, benchmark or hedged performance; generate a PDF diagnostic report; or explicitly verify the checked-in test fixture. Also use for Chinese requests such as 因子回测、收益验证、净值、回撤、持仓、换手、对冲收益或回测报告.
正在显示 SKILL.md
| name | corporate-action-adjustment-auditor |
| description | Use when raw close, adjusted close, cash-dividend, and split-factor data must be checked for invalid event inputs, unexplained price jumps, or adjusted-return mismatches before equity research or backtesting. |
| license | GPL-3.0-only |
| metadata | {"organization":"QuantSkills","organization_url":"https://github.com/quantskills","repository":"skill-corporate-action-adjustment-auditor","repository_url":"https://github.com/quantskills/skill-corporate-action-adjustment-auditor","project_type":"skill","collection":"corporate-action-adjustment","creator":"adennng","creator_url":"https://github.com/adennng","maintainer":"adennng","maintainer_url":"https://github.com/adennng"} |
| quantSkills | {"project_type":"skill","category":"tooling","tags":["corporate-action","adjusted-price","dividend","split","pandadata"],"platforms":["claude-code","codex","openclaw","cursor"],"language":"zh-en","status":"draft","validation_level":"runnable","maintainer_type":"community","requires":["skill-pandadata-api"],"summary_zh":"核对现金分红与拆并股场景下原始价格、复权价格和总收益的一致性。","summary_en":"Check raw and adjusted prices for cash-dividend and split consistency before research or backtesting.","license":"GPL-3.0-only"} |
把输入研究制品当作需要验证的证据，而不是默认可信的结果。先冻结口径，再运行确定性检查，最后把“已证实的问题”和“缺失证据”分开报告。
python scripts/audit_adjustments.py --demo 做离线烟雾测试；处理真实数据时用 --input <csv> --out <report.json>。references/output-contract.md 输出机器可读 JSON 和简洁中文结论。至少包含 symbol、date、close、adj_close、split_factor、cash_dividend 的 CSV。
symbol 必须为非空且不含首尾空白的证券标识符；date 必须使用 ISO-8601 YYYY-MM-DD 格式。symbol 至少需要两个日期不同的观测值，否则只能返回证据不足。split_factor 表示当前行事件中“每一旧股对应的新股数”，无拆并股时填 1，且必须大于 0。cash_dividend 表示当前行除息事件按每一旧股计的现金分红，无分红时填 0，且不得为负。close 与 adj_close 必须使用相同币种和价格单位并且大于 0。字段名不一致时先显式建立映射，不要猜测。缺少关键字段时停止定量结论，并列出补数清单。
--demo：使用内置样例；与 --input 互斥且二者必须提供一个。--input <csv>：读取 UTF-8 CSV；与 --demo 互斥且二者必须提供一个。--return-tolerance <float>：原始总收益与复权收益允许的绝对偏差，默认 0.02，必须是有限非负数。--jump-threshold <float>：无事件原始价格跳点阈值，默认 0.40，必须是有限正数。--out <json>：可选输出路径；省略时把 JSON 写到标准输出。JSON 异常清单，包含原始总收益、复权收益和冲突原因。
读取 references/output-contract.md 获取统一的证据等级、结论状态和报告字段。输出至少包含：输入规模、参数/假设、逐项发现、限制和下一步修复。
在修改阈值、公式或解释前读取 references/methodology.md。保留数据版本、时区、样本窗、随机种子和所有降级项，使另一位研究员可以复现结果。
references/pandadata-integration.md，再使用兄弟 Skill pandadata-api 查询：get_stock_daily、get_stock_daily_pre、get_stock_daily_post、get_adj_factor、get_stock_dividend、get_stock_split、get_stock_cash_dividend、get_stock_allotment。它验证数据处理，不预测公司行动，也不提供事件交易信号。