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。

Apirevise — 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
A

dsh-apirevise

Apirevise

用于 DeepSeek Harness 的 API 回归与验收工作台。

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-apirevise#1ac43c49c5e0f7ca9009913434068c6e124d60a9
README兼容性版本

兼容性与来源证明

Apirevise 以 dsh-apirevise 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 developer-tools 分类下经过校验的插件。

Cc Safety Netcc-safety-net编码代理 CLI 钩子——阻止破坏性命令和访问机密文件Vision Toolkit@anionex/dsh-vision-toolkit面向 Harness 原生集成的 DeepSeek 与 agent-vision-toolkit:图像问答、OCR、定位、UI 还原、像素差异、Artifacts 和 Web UI。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 syBridge@wenbin_wb/dsh-bridgeDSH 远程连接/远程控制插件(remote control):手机扫码即可在移动端/公网继续用 DeepSeek Harness,人不在电脑前也能接着干。一键局域网二维码、Cloudflare 公网隧道、自建隧道与微信 / QQ / 飞书 / Telegram Bot(多工作区/会话持久化/媒体/卡片审批/流式输出),无需自己搭公网服务器。

README

DSH API Revise

Snapshot your APIs. Rerun after the agent edits. Accept each difference — don't guess.

Pre-release v0.1.0. Verified with DSH Web 0.2.0-rc.2 on Windows, Node 24. UI and reports are Chinese. The real-model closed loop passed on 2026-10-01: baseline → real DeepSeek V4-Flash edited the backend → rerun → structured diffs → per-item acceptance → report. Model steps were never mocked; see docs/real-model-validation.md.

中文说明 · Compatibility & limits

Real DeepSeek run: register endpoints, baseline, model edit, structured diff, per-item acceptance, report

This 64-second walkthrough uses stills from the actual run on 2026-10-01 (DSH Web 0.2.0-rc.2 + real DeepSeek V4-Flash). It is not a continuous recording. Details: docs/real-model-validation.md.

Why: your DSH agent changes backend code, and you no longer know whether endpoint behavior changed. DSH API Revise records response baselines before the edit and shows structured, field-level differences after it — status codes, fields, and numeric values with old → new values — side by side with a source-code diff. You accept every difference item by item, and export a report.

What you get

  • Register endpoints: method + URL + optional headers/body, all typed by hand in the DSH Web panel (no traffic capture — that is v2).
  • Baseline snapshot: each endpoint is requested once and its response (status / headers / JSON or text body) is stored in the project's .dsh-apirevise/ directory, together with a bounded snapshot of backend source files.
  • Current-session handoff: the change goal plus the monitored endpoint list is sent to the current DSH session through DSH's normal prompt queue.
  • Rerun & compare: after the model edits, one click re-requests every endpoint and produces a structured diff — JSON field-level differences (added / removed / type / value, numeric delta), status-code changes highlighted, text patch for plain-text bodies.
  • Per-item acceptance: accept each endpoint, or mark it “needs work” with feedback that travels back into the next request. Comparisons go stale when the source changed after the rerun: you must rerun before accepting. Only when every endpoint is accepted against the current rerun can the round be completed.
  • Portable reports: Markdown plus a single-file HTML report with embedded diffs, XSS-escaped and served with a strict CSP. Facts only — diffs never claim “the API is fine”; acceptance is the human's decision.

Install

Use Node 22.19+ (22.x) or Node 24. Install and configure DSH according to its official documentation. DSH handles your model credentials; API Revise has no separate API key.

From npm

dsh plugin --profile web add dsh-apirevise

Published as dsh-apirevise@0.1.0. A pre-release tarball is also attached to v0.1.0.

Source checkout

git clone https://github.com/Lostforest7/dsh-apirevise.git
cd dsh-apirevise
npm ci
npm run build

Stop DSH, then register the checkout in its Web profile and restart:

dsh plugin --profile web add /absolute/path/to/dsh-apirevise
dsh web

The repository includes built lib/ files so DSH can load the plugin without running a package build script.

Use it

  1. Start your backend locally (e.g. node server.mjs from examples/api-server, port 3000).
  2. Open a DSH project session whose directory is the backend project. Click API 接口验收台 at the bottom right.
  3. Create a round: title (what to change), goal (sent to the session), and the endpoints to monitor. Creating the round snapshots the baseline immediately.
  4. Click 生成请求, review it, and 发送到当前会话. The model edits the backend code.
  5. After the model finishes, click 重跑对比. Review the structured diffs and the source-code diff; accept each endpoint or attach feedback and send the session back to work.
  6. When every endpoint is accepted, click 完成本轮验收, then 导出报告.

Add .dsh-apirevise/ to your project's .gitignore.

What it is not

  • Not a traffic-capture proxy (v2 direction); not a frontend tool (that is dsh-revise); not a test-framework runner (that is dsh-test-workbench territory).
  • It never claims “the API is definitely fine.” Diffs present facts — status / fields / numeric old-and-new values — and acceptance remains a human decision.
  • The model step is never mocked in validation; see docs/real-model-validation.md.

Development

npm ci
npm run check        # typecheck + unit tests + build
npm run dev          # standalone workbench preview (examples/api-server) on http://127.0.0.1:4320

Unit tests are fixture-based and run offline (loopback HTTP only), no DSH and no network required: tests/unit.

License

MIT