DEEPSEEK HARNESS · ECOSYSTEM MONTHLY · 2026-11-16

DSH 插件生态月报

对目录全部 5,643 个条目实测 npm 周下载(窗口 this week),可复现。这页回答一个问题:「我的插件的下载数,在生态里算什么水平」。

目录规模 5,643 条,较 基线测量(2026-11-15,5,643 条)净增 +NaN(0%)。生态在加速膨胀。

底噪带稳定性(跨时间点)

测量日样本日中位数底噪带真实日用爆发形
2026-11-10(窗口止 2026-11-08)1983558.6%6.1%19.7%
2026-11-15(窗口止 2026-11-13)1983460.1%8.1%28.3%
2026-11-15(窗口止 2026-11-13)19437925.8%69.6%10.8%
2026-11-15(窗口止 2026-11-13)1983460.1%8.1%28.3%
2026-11-16(窗口止 2026-11-13)1993263.8%6.5%20.6%

窗口滑动但结构不动:约六成条目停在底噪带,发版爆发形常驻约两成。这就是「下载列需要一个不误读的口径」的结构性依据,也是 信号提案 的趋势底座。

采用漏斗(自用 App 状态)

测量日插件版本工件Session人工决策最近决策
2026-11-10unknown75100从未
2026-11-150.1.386140从未

人工决策至今为零:回填在增长,但采用未发生。待办深度不是采用。 三个数为什么只有一个是采用信号,见 采用漏斗的方法论。

本周坐标(来自 weekly 测量文件)

项值
上游状态pr13722: MERGED (first upstream merge);pr13729: OPEN, held:maintainer; reviewer fact-correction applied 11-11; zero-review ping sent 11-16;issue13819: CLOSED by maintainer: display/sort unchanged by decision; concrete alternative-signal proposals with reproducible measurement explicitly invited (new issue);discussion11870: peer author zuwenyin independently confirmed the floor; quantified release-day overhead at 8.3x; answered their version-change question 11-15;research-report#27: CLOSED;signal-proposal#1207: dsh-market/dsh-market — filed 11-15, maintainer declined (badge fatigue 20-28% hit rate + number already positioned as relative size); reopening condition = ranking-error data; rank-stratified measurement ran same day (TOP-200 burst 10.8% vs catalog 28.0%, ranking NOT distorted), proposal withdrawn with data per the promised protocol;p0-1-evidence#1208: dsh-market/dsh-market — acted on within hours: PR #1212 (github-install-target.ts pre-install guard) shipped in harness v1.73.20; maintainer confirms our case is covered, three evidence-less edge cases deliberately not added; our catalog entries are npm-source so the guard layer does not apply to them;catalog-listing: awesome-dsh-plugin PRs #15594/#15655/#15656 MERGED (8 subpackage entries live, per-package pages 200); PR #15951 (packages 4-5: meeting-notes, resume) check + Submission gate both SUCCESS, awaiting merge; npm catalog feed frozen at 2026.1114.20961 pending the upstream dsh-dot-mcp duplicate fix (#15579 review names the root cause);nightly-blocker: the scheduled build fails on a duplicate npm name: Tests4gptplay/dsh-dots root package.json declares name dsh-dot-mcp (dsh.bundle present, gate-legitimate) and its subpackage entry resolves to the same npm package, so format-1 whole-doc uniqueness throws and dsh-plugin-catalog has not published since 2026.1114.20961 (11-14 05:27Z) — the in-app Discover feed is frozen at the 11-14 snapshot; known upstream via #15569 (frozen versions) and #15579 (de-dup fix PR, our review comment 11-16 names the root-manifest slip and which row survives);discover-impact: our 8 entries are visible on the site and its pages but not yet in the in-app Discover until the nightly publishes again; npm-side installs are unaffected;vertical-kit: packages 4-5 shipped 11-16: dsh-meeting-notes-check + dsh-resume-check, 0.1.1 live, site 6 pages, kit at the 5-package cap for this wave

排名数字仍是发版突刺的形状,不作采用证据;分解与口径见 底噪基线。

分布(周下载,全体目录)

分位下载
median362
p25234
p75706
p901,438
p952,335
p9910,932
max848,080

中位数 362 是「正常水平」。实测里 0.4% 的条目周下载不足 100,但其中大部分仍能稳定录得每天几十次——底噪带与是否有人安装无关。

头部榜单(Top 15)

#包周下载
1billion-context848,080
2dshmarket293,468
3dsh-context133,085
4dsh-pet95,892
5dsh-better-sidebar95,414
6dsh-agent-compare92,759
7dsh-remote79,516
8dsh-ops73,476
9@liustack/modlens70,692
10dsh-mobile-plugin58,280
11dsh-whale-widget50,556
12cc-safety-net49,410
13@vectorize-io/hindsight-coding-agents41,375
14dsh-windows-mcp41,368
15dsh-free-search39,301

分类分布(按中位数排序,条目数 ≥5)

类别条目中位数p75头部包
agi454981,187@suifei/dsh-ai-team(32,083)
memory244472953@vectorize-io/hindsight-coding-agents(41,375)
remote1794461,067dsh-mobile-plugin(58,280)
model300422883dsh-connect-workbuddy(34,342)
identity71415781dsh-fish-soul(13,774)
workflow369412785dsh-agent-compare(92,759)
tools827402735dsh-remote(79,516)
skill166401798@linxin666/dsh-client-ui-skill-explorer(30,620)
browser134383826dsh-free-search(39,301)
docs113382823@tt-a1i/archify-dsh(16,124)
vision100382683@liustack/modlens(70,692)
theme181381784dsh-plugin-wallpaper-engine(28,333)
security155377758cc-safety-net(49,410)
market753711,061dshmarket(293,468)
git104370666@linxin666/dsh-client-ui-git-graph(7,891)
notify185362614@xmanrui/dsh-im(26,691)
voice80351881dsh-music-panel(19,809)
session435341640billion-context(848,080)
usage286340575dsh-context(133,085)
fun162318588dsh-pet(95,892)
ui1,081307569dsh-better-sidebar(95,414)
dev337306619@michengai/dsh-skills-manager(37,213)
wsl12306524dsh-wsl-suite(1,339)

读法:中位数高 = 需求已被验证;条目多而中位数低 = 拥挤;条目少而中位数高 = 空位。中位数仍含底噪(每类都如此),只做相对比较。

本项目坐标

本项目 dsh-review-outbox 周下载 1,997,排序 #352 / 5,641(前 6.2%)。这个排名是发版突刺的形状:本周前两天的下载占 82%,安静日 36–81 落在底噪带。排名数字不作采用证据。

给目录/市场的信号提案(带可复现测量):单日下载占周总量 ≥50% 的包,其周数字应显示为「发布日」标记而非裸数字——实测 19.7% 的抽样条目是这种形状,头部榜单里也常驻(见基线文档)。这不是要求改排序,是给读者一个不误读的口径。

复现

node scripts/ecosystem-snapshot.mjs --package <name>(全量周分布)· node scripts/ecosystem-snapshot.mjs --shapes 200 --seed 20261110(抽样 30 天形状)· 月报由 scripts/ecosystem-monthly.mjs 从同一缓存渲染。

原始数字:docs/metrics/ · 方法论:ECOSYSTEM-BASELINE · 测量时间 2026-11-16