Repository navigation
Releases: LydiaTools/keep-every-cell
Release list
v0.1.0 - Review the extra cell before Excel export
Three headers. Four cells. One explicit repair.
Keep Every Cell reviews uneven Markdown table rows before Excel export. Extra cells stay visible; missing cells are not silently padded. You choose adjacent cells to join or edit the source. IDs start as Text, and numeric columns are a deliberate choice.
Open the app · 中文界面 · Method
Offline: extract keep-every-cell-v0.1.0.zip, enter the keep-every-cell folder and open index.html. No build, package installation, account or model key is needed. Save .md provides a source copy for later; review JSON is an audit record.
Verified: 17 core checks, an actual Chromium repair/download workflow, desktop and mobile captures, English/Chinese switching, keyboard file import, invalid UTF-8 preservation, new-file type reset, and offline file entry. Actual generated workbooks were read with openpyxl 3.1.5. IDs stayed strings, selected Qty cells became numeric, and =2+2 remained text. A local ZIP/XML readback independently checked the earlier downloaded artifacts.
Runtime CI 39512243293 passed at 09c1166. The release commit 769bfde only adds README text, recorded verification and unedited PNG captures; parser, app, writer, styles and tests match that verified runtime exactly. README/PNG-only updates do not rerun the browser pipeline.
The source ZIP was compared byte for byte with all 21 tracked files. The example workbook and two raw reports come from that CI run. SHA256SUMS.txt covers all four assets. The optional tablemd comparison is one original synthetic case of documented normalization versus strict review, not a general ranking or bug claim.
Early preview: one bounded pipe-delimited Markdown table. This is not PDF/OCR extraction, full GFM rendering or verification that AI-generated facts are true. Native Excel, Numbers, Safari and screen-reader acceptance have not been performed. The app does not upload source or load analytics.
中文版:错列先看清、修法自己选,再交给 Excel。额外内容保留可见,编号默认保留文本,需要计算的列再选数字。支持中英文、离线双击打开及原文另存;截图和导出样例均来自实际测试,不是生成的产品示意图。