DeepSeek Harness Plugin Hub

发布与管理完整 Harness Profiles,发现适合你的插件。

探索

插件目录环境预设文档中心动态

社区

发布插件联系我们报告问题

相关链接

Plugin Hub GitHubDeepSeek Harness 官方项目系统状态隐私说明
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

独立、非官方社区项目,与 DeepSeek 官方无隶属、授权或背书关系。

Sheetflow — DeepSeek Harness 插件(DSH Plugin)
DeepSeek Harness Plugin Hub
ProfilesPlugins分类动态文档登录管理 Profiles
ProfilesPlugins分类动态文档登录
← Plugins
S

dsh-sheetflow

Sheetflow

填写重复使用的 Excel 模板,审查对账异常,并在 DeepSeek Harness 中追踪每个数值。

插件会安装到这里;不确定时保持 web。

npx -y @deepseek-ai/dsh plugin --profile web add github:Lostforest7/dsh-sheetflow#6761ae184a6ab0af61b7fe372a42c2027bf33827
README兼容性版本

兼容性与来源证明

Sheetflow 以 dsh-sheetflow 发布,当前版本为 0.1.0。Plugin Hub 会校验它的 manifest,并保存精确安装来源,便于复现安装结果。

DSH 兼容范围
*
运行环境
web
发布来源
github
Registry 更新时间
2026/10/1

版本

0.1.0stable
2026/10/1

相关插件

正在加载相关插件…

最新版
0.1.0
DSH
*
HMR
重启进程
Tree shaking
未声明可安全裁剪
解包体积
未提供
文件数
未提供
Surface
web
许可证
MIT
发布源
github
GitHub
★ 0
周下载
0
最近提交
2026/10/1
查看源码 ↗
README Badge

点击下方 Badge 复制 Markdown,粘贴到 README 即可。

这是你的 Plugin?认领权益 · 优先安全扫描

验证 package.json 声明的 GitHub 仓库,即可管理这个公开页面。认领后,Hub 会优先安排当前版本的安全扫描,并在通过后公开展示结果。

认领这个 Plugin →
报告问题

相关插件

继续浏览 productivity-workflow 分类下经过校验的插件。

Codex Ui@michengai/dsh-codex-uiDSH Codex UI — 为 DeepSeek Harness Web 提供 Codex 风格侧栏、工作区会话树、全局搜索和轮次导航Remotedsh-remoteDeepSeek Harness 的远程工作助手:连接 SSH(密码/密钥/代理/键盘交互/代理跳转),选择远程工作区,使用 22 个 rw_* 工具操作该工作区(包括 rw_edit/rw_stat/rw_mkdir/rw_remove/rw_move/rw_forward),具备冲突感知的镜像 syAi Novel Writer@ethanyoq/dsh-ai-novel-writer面向 DeepSeek Harness 的本地优先小说项目插件Mneme@modusensus/dsh-mneme适用于 DeepSeek Harness 的跨会话记忆插件,支持 autoDream 整合:SQLite 存储、Markdown 镜像、9 个模型工具、自动注入、会话摘要、用户配置文件/规则、自定义斜杠命令、向量(语义)搜索,以及 Web GUI pa

README

DSH SheetFlow

Fill recurring Excel templates, review reconciliation exceptions, and trace every value to its source.

Financial disclaimer: SheetFlow assists with data preparation and reconciliation review. It does not provide accounting, tax or financial advice and is not an accounting system. AI mappings and all exported amounts, formulas and exceptions require independent review before business use. Acknowledging an exception does not resolve it or approve a payment.

v0.1.0-pre preview: tested on Windows with Node 24.19.0 and DSH Web 0.2.0-rc.2. Chinese / English workbench. Three examples plus one newly created synthetic blind dataset verified. A real deepseek-v4-pro / max run proposed the correct columns but needed two source-header-row corrections before export. This is one assisted success, not proof of unattended mapping or production workbook compatibility. See the real model review. GitHub Pre-release downloads; no npm publication for this preview.

中文 · Compatibility · Validation evidence · Recipe format

The blind export also passed a delegated native Microsoft Excel UI check: forced recalculation, a live formula edit/undo, all eight rows and three merged regions. Screenshots and results. This was operated by Codex on synthetic data, not a human accounting sign-off.

Synthetic example data: reviewed month-end records, exceptions and evidence

Demo screenshot: synthetic example data.

An independent community plugin for DeepSeek Harness. Start with an orders table, a settlements table, and an existing business template. Confirm the mapping once, preview every write, review every exception, and export a new template copy with an HTML report and JSON trace.

  • Explicit mappings: file, sheet, header row, match key, amount, and reserved template cells.
  • Exact decimal comparison: two-decimal amounts, optional tolerance, explicit split-payment aggregation. Blank amounts remain unavailable.
  • Explainable exceptions: missing records, duplicate keys, invalid inputs, and amount differences retain source file / sheet / row / cell evidence.
  • Template copies: retains original ZIP parts and cell styles, protects formula cells and merges, and previews clearing old unused rows.
  • Repeatable recipes: save/import/export rules; changed headers stop a reused rule. Local histories are isolated by workspace and DSH session.
  • Human review: every exception must be acknowledged before export. Changed inputs invalidate a review. Acknowledgement records inspection, not resolution.
  • DSH tools: AI can inspect bounded samples and submit a proposed recipe. Execution, acknowledgement, and export remain workbench actions.

Try locally in three steps

Use Node 24, or Node 22.19+ within the 22.x line. No Microsoft Excel or model API key is needed for this local sample.

# Run in the extracted source folder
npm ci --cache .cache/npm
npm run build
npm run dev

Open the local URL printed in the terminal. Choose Load sample mapping, confirm the mapping, then Generate preview. The sample has eight records: two matched and six exceptions. Review evidence and acknowledge all exceptions to export. New .xlsx, .html, and .json files are saved alongside the inputs in examples/month-end/.

The standalone preview verifies rule execution. AI proposals are available inside DSH.

Use in DSH

  1. Install and configure DSH using its official instructions. Build this source folder using the commands above, then register the plugin:

    dsh plugin --profile web add /absolute/path/to/dsh-sheetflow
    # A built .tgz can be registered with the same command.
    dsh web
    
  2. In DSH, select a workspace containing the two source tables and template. To try the example, select the plugin's examples/month-end folder. Click SheetFlow 填表台 and switch to English if desired.

  3. Confirm mappings and the writable region, generate a preview, inspect exceptions, and export a new copy. For AI help, send mapping instructions, wait for the current DSH session to finish, then choose Load proposal and confirm it.

Add .dsh-sheetflow/ and generated sheetflow-*.xlsx, sheetflow-*.html, sheetflow-*.json to the business workspace's .gitignore. Histories and reports contain business data. Input files remain local during deterministic processing; using AI sends the supplied instructions, bounded samples, and inspection results through your configured DSH model provider.

Scope

Ordinary .xlsx templates and .xlsx / UTF-8 comma-separated .csv tables; one key and one two-decimal amount per source, one shared currency and business period, a fixed reserved writable region. Sources may be separate sheets of the same workbook. Recipes are per DSH session; use JSON to transfer them.

This preview does not calculate Excel formulas, perform currency conversion, fuzzy matching, many-to-many reconciliation, OCR/PDF ingestion, or insert template rows. Formulas are preserved and caches invalidated; open the copy in Excel and recalculate before relying on formula outputs. Macros, digital signatures, external workbook links and protected target sheets are rejected. Complex workbook fidelity and remote/sandbox-only filesystem backends are unverified. Full limits.

Develop

npm run check
npm run test:e2e

Browser tests use Windows Google Chrome when available; otherwise run npx playwright install chromium. SHEETFLOW_CHROMIUM_PATH can select a browser. .npmrc uses legacy peer resolution to work around the tested DSH prerelease dependency graph; DSH itself should be installed separately using its official instructions.

For another local fixture:

$env:SHEETFLOW_WORKSPACE = "examples/next-month"
npm run dev

Import that folder's recipe.json through Advanced. The next-month example demonstrates 0.10 + 0.20 = 0.30; the supplier example uses another sheet, row range and prefilled template.

MIT licensed. Preview acceptance, integration boundaries, and the next validation gate are documented in validation.md.