插件目录 / Developer / dsh-live-stats
dsh-live-stats
已验证 · 实测可装 Proton1917
功能简介
dsh-live-stats — DSH 插件(模型)
可用 — 实测通过,早期项目
dsh-live-stats — DSH 插件(模型) 实测能干净安装、正常启动。早期项目,但功能可用。
「已验证」表示我们的自动化 CI 在干净 profile 里实际执行了 dsh plugin add 并启动成功——仅此而已。功能描述与版本兼容性均为作者声明。这不是安全审计,也不代表对第三方代码的背书。
README
DSH Live Stats
External plugin for DSH Web. It owns the composer's stats slot cell, renders the complete session status row with live input, output, and unabridged total token counts, and adds a second row for generation throughput:
Input ~7.9K tok · Output ~12 tok · Total ~7,912 tok
TPS 31.4 tok/s
~ marks a value that still includes the current request's input estimate. Streamed DeepSeek output is counted with the official tokenizer; provider usage replaces all provisional values when it arrives. Exact cache accounting continues to come from DSH's durable token-usage projection. A retry replaces the prior estimate for that step, and an aborted turn removes its unsettled estimate.
TPS is the number of tokens added after the first output sample divided by the elapsed time between the first and latest token-bearing stream events. Final usage, block-end, and assistant-message events correct counts but do not extend that interval. A one-sample response therefore has no TPS value instead of an artificial terminal-event rate.
The plugin replaces the built-in conversation.composer.dock entry through the slot registry's documented same-id priority mechanism. It reads public sessionStats and tokenUsage projections, owns liveTokenUsage, and leaves the Harness checkout untouched. A Harness update therefore no longer needs to replay a StatsLine source patch.
The package is a profile bundle and mounts itself without a separate home-level patch:
- insert:
- id: live-stats
name: '@proton1917/dsh-live-stats'
The bundled tokenizer comes from DeepSeek's official tokenizer download. See THIRD_PARTY_NOTICES.md for its source, checksums, and license references.
Requirements
- DeepSeek Harness
0.1.0-rc.7or later. - A local checkout when developing through
link:, or a GitHub/npm package spec for a managed installation. - Node.js and pnpm versions accepted by the harness repository.
Build and test
pnpm install
pnpm run typecheck
pnpm test
pnpm run build
Load in DSH Web
Build the plugin and add its bundle to the Web profile once:
dsh plugin --profile web add @proton1917/dsh-live-stats@link:$HOME/Projects/dsh-live-stats
dsh web
The profile records the bundle in dsh.profile.bundles; normal Harness installations and upgrades keep the profile and its plugin dependency outside the Harness source checkout.
For client and server hot updates, run the Web app in development mode and keep the plugin watcher in a second terminal:
dsh web --dev
cd ~/Projects/dsh-live-stats
pnpm run dev
The plugin watcher rebuilds lib/index.js and lib/client.js; DSH Web reloads the client contribution through HMR without replacing the page.
Compatibility updates
The repository builds against published DSH release packages rather than relative links into one Harness checkout. Dependabot groups new @deepseek-ai/* releases into one compatibility pull request, and GitHub Actions runs typecheck, unit tests, and both client and host builds on every update. A new Harness release therefore produces a concrete compatibility result without relying on a manual reminder.
For a locally linked checkout, merge the generated compatibility pull request, fast-forward this checkout, and rebuild it. The Web profile keeps the same link and bundle registration.
License
BSD-3-Clause. See LICENSE.
安装
装一次目录插件,之后本站所有插件都能让 DeepSeek Harness 自动找、自动装:
dsh plugin add dshbase-catalog 然后对 agent 说「帮我装 dsh-live-stats」,它会在目录里找到并自动安装。文档:dshbase-catalog · 已验证场景包。
该插件是 GitHub 源码(未发 npm)——直接从仓库装:
Web profile:
dsh plugin --profile web add github:Proton1917/dsh-live-stats Headless(CLI)profile:
dsh plugin --profile headless add github:Proton1917/dsh-live-stats 实测报告
验证通过:从 GitHub 源码完成 L1 安装 + L2 加载 + L3 运行(dsh 0.1.0-rc.6)。
使用场景
扩展 agent 的编码能力面——给它一个新工具、工作流或集成,让它接手以前做不了的开发任务。
适合谁
想让 dsh 在真实代码库上像队友一样干活的开发者——能改、能跑、能验证,而不只是回答问题。
二次开发建议
工具/命令面就是缝:暴露更多 SDK 能力、加更聪明的上下文接线,或收紧改代码与验证之间的循环。