zhihu-report-polisher

Zhihu Report Polisher

Use this skill to turn a local investment-prediction report Markdown into a reader-facing Zhihu article draft, and optionally paste it into Zhihu's article editor for preview. Do not publish anything unless the user explicitly confirms publishing after reviewing the generated draft and editor preview.

Output

• Write the public draft under zhihu/.
• Name it from the source file stem, ending in -zhihu.md.
• zhihu/ is ignored by git; do not force-add generated drafts unless the user explicitly asks.
• If the draft references local images, copy them into zhihu/assets/ and use relative Markdown image links.
• Before any publish-related browser step, send the generated Markdown file and every referenced image to the user through the available IM/file tools so they can review the exact content.

Required Opening

Put 分析对象 immediately after the article title. Then add a short project introduction in the style of:

本文由开源项目 `fluencer-predict-track` 辅助生成和整理。项目地址：

https://github.com/wzj998/fluencer-predict-track

这个项目用于采集公开投资/财经账号的历史预测，结合公开数据做复盘，并在可映射到交易标的时生成回测结果。本文不构成投资建议。

Then continue with method notes and the main content.

Rewrite Rules

The Zhihu draft is a polished public article, not an engineering report.

• Keep the useful reader-facing content: headline, target account, sample size, verification cutoff, qualitative conclusions, backtest result, backtest assumptions, limitations, tables needed to understand the backtest, and external sources.
• If backtest metrics are included, preserve position usage information such as average exposure, maximum exposure, average cash ratio, and cash drag. State clearly that annualized return and Sharpe are affected by position usage: accounts with more frequent tradable signals may become more fully invested, while sparse-signal accounts may keep more cash.
• In 结论先行, include profit/loss contribution analysis when a backtest signal table has per-operation contribution. Summarize the main positive and negative contributors by ticker, sector/theme, and operation type, then translate them into reader-facing language such as which areas were more accurate and which areas dragged returns. Do not leave the opening conclusion as only total return, win rate, CAGR, or Sharpe.
• If a backtest signal table is included, place a 信号准确表 immediately after it. This table should map the sampled public predictions to their qualitative verification result, using public evidence and labels such as 准确, 部分准确, 不准确, 尚未到期, or 不计分.
• Remove or rewrite implementation details: local file paths, candidate-pool file names, internal report paths, exact CDP endpoint URLs, cookies, browser state, script names, zip/package names, and workspace-specific paths.
• It is acceptable to say "通过本地 Chrome CDP 采集知乎公开回答页", but do not include a concrete local URL such as 127.0.0.1:15166.
• Do not expose iteration wording such as "上一版", "旧版", "新版", "重新跑", "rerun", or "candidate file". Use "本次样本", "定性复盘", "回测结果", and "预测样本".
• If the source report contains a net-value/equity-curve image, copy the image into zhihu/assets/ and reference it with Markdown image syntax, for example ![净值曲线](assets/<chart>.png). Do not replace generated charts with vague placeholders such as "此处插入".
• Keep public URLs that support verification, including the GitHub project URL, Zhihu answer links, and external data/source URLs.
• Keep financial disclaimers concise: include that the article is a sample review and not investment advice.