DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in
← Plugins

dsh-fast-mode

Fast Mode

DeepSeek Harness’s own fast mode: the same model, half the output, up to twice as fast. Switches combine freely and stay prompt-side — never changes the model, reasoning effort, or the request body. Best measured 0.505x wall clock, −42% output tokens, 438

The plugin will be installed here. Keep web if you are unsure.

npx -y @deepseek-ai/dsh plugin --profile web add dsh-fast-mode@1.7.28
READMECompatibilityVersions

Compatibility and provenance

Fast Mode is published as dsh-fast-mode and currently resolves to version 1.7.28. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
web
Release source
npm
Registry updated
11/4/2026

Versions

1.7.28stable
11/4/2026
1.7.27stable
11/4/2026
1.7.26stable
11/4/2026
Show 73 more versionsCollapse versions
1.7.25stable
11/4/2026
1.7.24stable
11/4/2026
1.7.23stable
11/3/2026
1.7.22stable
11/3/2026
1.7.21stable
11/3/2026
1.7.20stable
11/3/2026
1.7.19stable
11/3/2026
1.7.18stable
11/3/2026
1.7.17stable
11/3/2026
1.7.16stable
11/3/2026
1.7.15stable
11/3/2026
1.7.14stable
11/3/2026
1.7.13stable
11/3/2026
1.7.12stable
11/3/2026
1.7.11stable
11/2/2026
1.7.10stable
11/2/2026
1.7.9stable
11/2/2026
1.7.8stable
11/2/2026

Related plugins

Loading related plugins…

Latest
1.7.28
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
1.1 MB
Files
56
Surface
web
License
MIT
Source
npm
GitHub
★ 8
Weekly downloads
9,523
Security scan
✓ v1.7.28 scan passed
Last push
11/4/2026
View source ↗Project homepage ↗
README badge

Click the badge to copy Markdown for your README.

Do you maintain this Plugin?Claim benefit · Priority security scan

Verify the GitHub repository declared in package.json to manage this listing. After you claim it, Hub will prioritize a security scan of the current version and publish the result when it passes.

Claim this Plugin →
Report an issue
1.7.7stable
11/2/2026
1.7.6stable
11/2/2026
1.7.5stable
11/2/2026
1.7.4stable
11/2/2026
1.7.3stable
11/2/2026
1.7.2stable
11/2/2026
1.7.1stable
11/1/2026
1.6.2stable
10/29/2026
1.6.1stable
10/29/2026
1.6.0stable
10/29/2026
1.5.35stable
10/29/2026
1.5.34stable
10/29/2026
1.5.33stable
10/28/2026
1.5.32stable
10/28/2026
1.5.31stable
10/28/2026
1.5.30stable
10/28/2026
1.5.29stable
10/28/2026
1.5.28stable
10/28/2026
1.5.27stable
10/28/2026
1.5.26stable
10/28/2026
1.5.25stable
10/28/2026
1.5.24stable
10/28/2026
1.5.23stable
10/28/2026
1.5.22stable
10/28/2026
1.5.21stable
10/28/2026
1.5.20stable
10/28/2026
1.5.19stable
10/28/2026
1.5.18stable
10/28/2026
1.5.17stable
10/28/2026
1.5.16stable
10/28/2026
1.5.15stable
10/28/2026
1.5.14stable
10/28/2026
1.5.13stable
10/28/2026
1.5.12stable
10/27/2026
1.5.11stable
10/27/2026
1.5.10stable
10/27/2026
1.5.9stable
10/27/2026
1.5.8stable
10/27/2026
1.5.7stable
10/26/2026
1.5.6stable
10/26/2026
1.5.5stable
10/26/2026
1.5.4stable
10/26/2026
1.5.3stable
10/26/2026
1.5.2stable
10/26/2026
1.5.1stable
10/26/2026
1.5.0stable
10/25/2026
1.4.4stable
10/25/2026
1.4.3stable
10/24/2026
1.4.2stable
10/24/2026
1.4.1stable
10/24/2026
1.4.0stable
10/24/2026
1.2.0stable
10/24/2026
0.8.0stable
10/24/2026
0.5.2stable
10/24/2026
0.5.1stable
10/24/2026

Related plugins

More verified plugins in models-usage.

Experimental Cot Translation Bundle@deepseek-ai/dsh-experimental-cot-translation-bundleExperimental machine translation of expanded reasoningWhale Widgetdsh-whale-widgetDeepSeek balance whale widget in the bottom-right corner of the DSH Web interface: balance/today’s usage/peak-off-peak pricing, customizable bubble click sequence (text/balance/today/peak-off-peak/image/random phrases and parallel weighted selection), per-line styles and fonts, floating quick editinOur Free Model@ebony-vinyl/dsh-our-free-modelOur Free Model — Install this plugin in dsh to use cutting-edge models including DeepSeek V4.1 Flash and Kimi K3, with no login, registration, or API Key required. The desktop version also offers an EAC co-payment channel; starting with 2.0, it supports ten account channels: LobsterAI, Qoder, TRAE, Usage@linxin666/dsh-usageUsage statistics plugin for the dsh web GUI: per-provider balance and coding-plan quota detection plus a live token usage ledger, with a compact price glance card above the sidebar's Settings row

README

dsh-fast-mode

dsh-fast-mode

DeepSeek Harness 自己的 fast mode:同一个模型,只消耗一半的输出,最快快一倍。
多种开关自由组合,全在提示词侧:不换模型、不降思考强度、不写任何一个请求字段。
实测最快 0.505× 耗时、输出 token −42%、判分 438/438 全绿 —— benchmark 与复现。

English


一眼看完

30 对完整任务、同一模型同一设置,只差三个开关:

指标结果
墙钟耗时0.505×(快约一倍,30 对 26 对更快)
输出 token−42%(16037 → 9380)
思考字符−51%(42380 → 20872)
任务质量判分 438/438 全绿,30/30 运行通过

三个开关全开 + 面板默认项,是这套东西最快的一档;只开闪电、不开三个新开关,全池 138 对是 0.823×(省 18 %)。

怎么装

dsh plugin install dsh-fast-mode

装完在会话里点闪电。右键闪电打开八行面板,只影响当前会话。

八个按钮在做什么

八个开关都是纯提示词侧:不加请求参数、不改模型、不改完成判定。前两个管输出,最后一个管上下文体积,中间五个管「怎么走」。

按钮原理默认
精简输出不预告调用、不复述结果、不写进度,最终答案不缩水开
步数上限硬上限,到点先汇报现状,不截断任务不限
开局附文件清单开局把工作目录交给模型,省掉先跑一遍 ls/find关
批量读文件互不依赖的读取合成一批发出,少几个来回关
截短工具输出只留头尾和报错行,长输出不再整段进上下文关
不确定先查证拿不准就读一次;不是纯提速,省的是猜错后的重来关
先执行后解释动手前最多一行计划;思考过 4000 字符时提醒「按证据动手」关
重复时换方法同一调用重发、或连续三次结果一样时,提醒换条路关

中间五个是 1.7.0 起的新机制,默认关,理由见下面的边界。截短工具输出 1.7.12 起进面板——它是唯一会改变模型所见证据的开关,所以默认关,且关掉就是原文。

Benchmark

三开关全开(trio17)vs 全关,同一批任务逐对比较,配对 bootstrap 4000 次:

实验块对数墙钟几何均值95% 区间更快
DP17120.647×[0.501, 0.893]10/12
DP1890.491×[0.314, 0.788]8/9
DP20b(发布版)90.372×[0.240, 0.563]8/9
合池300.505×[0.401, 0.635]26/30

发布版那一块里,步数 13.2 → 6.7、墙钟 464.9 s → 173.1 s,三开在每一块都是唯一区间不跨 1 的臂。复现命令与原始数据在 BENCHMARK.md。

为什么能快,以及哪里不会快

一步的耗时 ≈ 往返开销 × 步数 + 解码时间 × 输出 token + 工具等待。模型不在这个式子里,所以这里只动程序量:少几次往返、少几轮回头读、少写重复的编辑。

  • 有些情况会变慢,别指望它全场景加速。 1–5 步的小改修和数据整形收不回成本(perfrefactor 1.18×、csvpipe 1.12×),实现类任务偶尔多绕一步(DP20 有一对 1.04×)。
  • 质量优先。 每个采纳的批次里两臂隐藏断言逐格相同;安全校验、数据删除、无障碍这些场景永远走完整验证,不做提速。
  • 开关在你手里。 默认只开「精简输出」;三个新开关默认关,因为它们的收益按任务形状波动,不适合替所有人默认决定。

推荐:大多数时候三个全开。想稳一点只开「不确定先查证」;最在意思考链过长就开「先执行后解释」;「重复时换方法」单开实测等于没变(1.007×,十二个会话只触发 3 次),放在组合里无害。

自己量一遍

上面的数字都来自本仓库的 bench,换到你的机器和你的任务上未必一样。先 node audit/run.mjs --dry-run --cwd <仓库> --tasks <清单> 看它要展开成什么命令,去掉 --dry-run 就是配对真跑(每任务每臂 3 轮),产物落在 audit-out/<时间戳>/:原始读数 pairs.jsonl 和带几何均值比与区间的人看报告 report.md。配对数不足 6 时报告只写「样本不足,不做结论」。

方法与结论边界见 docs/AUDIT.md,清单怎么写见 audit/tasks.example.jsonl,agent 操作手册见 skills/fast-mode-audit/SKILL.md。

许可

MIT。参考手册见 docs/REFERENCE.md,逐版改动见 CHANGELOG.md。