生态洞察 · 2026-W46
重点关注 · 8 个异常信号
中本周新增 9529,近 3 周均值 2102——突增,可能有批量导入/刷库
中账号 uckkk 有 1286 个插件、均分仅 56——疑似批量刷库/模板复制
中账号 xushilianzhuren 有 52 个插件、均分仅 56——疑似批量刷库/模板复制
中账号 ben7am1n 有 10 个插件、均分仅 58——疑似批量刷库/模板复制
中账号 CAI-MH 有 9 个插件、均分仅 57、1 个无 README——疑似批量刷库/模板复制
低最大账号 uckkk 独占 1286 个插件(占全生态 5%)、均分 56——模板批量号拉高规模数字,解读增长时需剔除水分
低10 个插件仓库版本领先 npm 发布版本——安装侧拿到的是旧版
高dsh 官方本周发布 2 个版本,其中 1 个含 breaking 变更(dsh-v0.2.2-rc.2)——插件作者需评估适配
DSH 生态洞察 · 2026-W46(2026/11/09~2026/11/15)
由 DSH Insights 管线 + AI 生成 · 数据快照 2026-11-16 · 启发式评估,非安全审计
1. 核心结论
- 增长被批量号污染:信号显示 uckkk 独占 1286 个插件(均分 56),weeklyDiff 相对基线 2026-10-15 的 9529 新增中需剔除水分;真实社区活跃度要看 active7 仅 15.1%。
- 质量与采用严重背离:billion-context(5170 stars, B 级)周下载 84.8 万,而 sunyuhuirong/deepseek-account 仅 1 star 却 51 万周下载;star 不再是可信的质量代理指标。
- 头部作者集中但头部占比低:top10Share 仅 2.2%,Ebony-Vinyl 靠单插件 75756 star 领跑,生态长尾极厚,但质量分化严重。
- 新队列质量正在回升:cohortTrend 平均分从 9 月初 74.1 升至 11 月 77.1,新作者入场质量改善,但需警惕批量号拉低均值。
- 发布链路存在系统性漂移:npm 发布率 44.1%,且有 10 个插件仓库版本领先 npm(如 dsh-market repo 1.43.0 vs npm 1.73.20),用户安装侧可能长期拿旧版。
2. 头部插件深潜
- ranxianglei/billion-context:状态/监控/用量类,stars 5170,score 86,B 级,weekly 848080。desc 提到 100K tokens context-compression proxy。为何关注:零星作者也能做出高下载工具,说明下载量由功能需求驱动而非 star;但存在 manifest.no-client-export 和 npm.version-drift 警告,发布质量待提升。
- dsh-market/dsh-market:官方市场,stars 12524,A 级,weekly 293468,npm 包名 dshmarket。赢在权威分发地位,但 npm 版本漂移(repo 1.43.0 vs npm 1.73.20)且 missing files,需警惕用户通过 npm 安装获得不一致体验。
- Ebony-Vinyl/dsh-our-free-model:stars 75756 全生态最高,A 级,weekly 10812,提供免登录免费模型调用。高 star 但周下载仅 1 万,说明热度未完全转化为安装;desc 被截断但 category 为状态/监控/用量,需确认其核心价值是否与分类一致。
- alchaincyf/nuwa-skill:stars 38349,B 级,published false,weekly null。未发布 npm 但 star 极高,说明用户通过源码/其他渠道安装;published false 且 weekly null 表示官方统计可能未覆盖其真实使用,生态测量存在盲区。
- sunyuhuirong/deepseek-account:stars 1,B 级,weekly 510593,desc 为空。极低 star 极高下载,疑似包名抢注或集中分发;desc 缺失和 docs.zh-missing 等 flags 表明用户可能在无文档保障下安装,风险高。
3. 作者格局
- 集中度:top10Share 仅 2.2%,total 15084 作者,multi 3778(约 25%);说明星力分散,但头部 10 人只占 2.2% 是因为总 star 基数大且长尾极多。
- 头部作者:Ebony-Vinyl(5 plugins, 78424 stars, avg 79.8)靠 dsh-our-free-model 单点引爆;alchaincyf(3 plugins, 68392 stars, avg 76.3)两个未发布插件占大头;dsh-market(1 plugin, 90 avg)质量最高但依赖官方身份。
- 多产作者:xby-skill 175 个插件但总 star 仅 2、avg 62.2,典型模板批量号;PerryLink 155 个插件 avg 84.7 高质多产;feely0208 104 个插件 avg 86.1 也优质,说明多产不等于低质。
- 连接者:cursoragent(5 plugins, 9984 stars)和 Ebony-Vinyl 出现在 connectors 列表,反映其生态协作网络中的枢纽地位;zixin947 3 plugins 78463 stars 但未出现在 topByStars,可能因为合作项目 star 归属。
- 污染:uckkk 1286 个插件(占 5%)均分 56,加上 xushilianzhuren、ben7am1n 等批量号,作者榜的规模数字被注水。判断:生态同时存在“社区驱动”的真实创作和“流量驱动”的批量灌库,官方需用质量分和活跃度做分层治理。
4. 质量 × 采用度交叉发现
- SenmuuuuW/dsh-whale-report:S 级 100 分,0 star,周下载 175,提供会话事件日报/周报。高质低用典型,官方应将其收录进 awesome 或市场推荐,扩大高质量工具曝光。
- Fish-under-sea/dsh-better-reasoning-effort-fish:S 级 100 分,2 stars,周下载 203,是上游 Fork 的增强版。Fork 改进被埋没,说明版本发现机制需支持 fork 溯源与推荐。
- 0x7a76/dsh-extra-paths:S 级 100 分,0 star,周下载 317,配置额外可写根目录。这类系统级增强插件对高级用户极有价值,但 star 为零,需通过场景化推荐触达。
- LuniteGlaze/dsh-plugin-catalog:C 级 64 分,1 star,周下载 103410;desc 为空,用户靠猜测安装,且无 star 验证,可能遇到未维护代码。
- Creakono/dsh-cost-meter:C 级 64 分,1 star,周下载 38681;成本计算类插件一旦逻辑错误会误导用户预算,低质量高用量风险极高。
- ben7am1n/dsh-browser:D 级 56 分,7 stars,周下载 2679;D 级意味着工程健康度差,浏览器自动化插件涉及系统控制,建议用户谨慎使用并检查源码。
5. 生态主题地图
- 会话与记忆是核心战场:categoryMap 中“会话管理”2406 个、平均 75.5,“记忆/知识”749 个、平均 75.7;topTags 中 session、memory 分别有 95 和 68 个标注。大量插件围绕上下文管理、记忆增强竞争,但平均分中上,仍有优化空间。
- UI/侧栏改造拥挤:“侧栏/工作区”2132 个、平均 75.6;“主题/视觉美化”958 个、平均 73.8;topTags 中 ui 73 个、sidebar 49 个。入门门槛低导致供给过剩,平均分被低质模板拉低,新作者应避免单纯 UI 包装。
- 模型/代理与 MCP 集成是上升通道:“模型/代理”2582 个、平均 75.8;topTags 中 mcp 56 个、agent 57 个。DeepSeek 自身模型生态带动插件集成需求,但能力标注覆盖率仅 1823/25046(7.3%),大量插件主题不明,官方需推动标注。
- 空白机会:topTags 中 vision 55、image 54,但“图像生成/视觉”未出现在 categoryMap 前 10,可能散落在“其它”8356 个中;通过主题标注可挖掘这些隐性机会。
- 风险方向:“其它”8356 个平均分仅 72,是最大类别且质量最低;大量未分类插件意味着元数据治理缺口,官方应强制分类以减少用户搜索成本。
6. 生态健康度评估
| 维度 | 本期数值 | 解读 |
|---|---|---|
| 规模 | authoritative 25046;作者 15084;medianStars 0 | 长尾极厚,一半以上插件零 star |
| 增长 | weeklyDiff 相对 2026-10-15 累计 +9529;byWeek 2026-11-02 2643,2026-11-09 327 | 基线口径与周口径背离,需核实数据管道;真实周增长可能 <1000 |
| 质量分布 | S 0.95%、A 4.34%、B 46.1%、C 37.2%、D 11.4%;avgScore 74.2 | B/C 占 83%,中低质量是主体 |
| 活跃度 | active7 15.1%(3780);active30 51.2%(12830) | 近半数插件 30 天内有活动,但 7 日活跃仅 15%——大部分是一次性提交 |
| npm 发布率 | 44.1%(11042/25046) | 超半数插件未通过 npm 分发,安装链路碎片化 |
| 文档覆盖 | readme 99.8%;zh 48.3%;none 60 | readme 几乎全覆盖,但中文文档渗透不到一半,对国内用户不友好 |
cohortTrend 解读:9 月首周平均 74.1,10 月中旬起升至 76-77,11 月 9 日 77.1(仅 327 个样本);新队列质量在改善,但样本量骤降可能掩盖批量号退出后的真实增量,需持续观察。
7. 异常与应对
growth-anomaly (mid)
- 可能原因(假设):批量导入/刷库;或基线口径导致累计值虚高,无法区分周内真实增量。
- 影响面:增长指标失真,会误导官方资源分配和用户对生态活力的判断。
- 应对建议:官方应分离“累计新增”与“周新增”,并增加去重、活跃度过滤后再呈现增长。
owner-cluster (mid) - uckkk
- 可能原因(假设):模板批量生成插件,无实质功能差异。
- 影响面:拉低作者榜可信度,1286 个低分插件污染搜索结果与市场推荐。
- 应对建议:对 uckkk 等批量号做标记或限流,对均分 <60 且插件数 >50 的账号启动人工审核。
owner-cluster (mid) - xushilianzhuren
- 可能原因(假设):模板批量生成插件,无实质功能差异。
- 影响面:52 个低分插件增加长尾噪声,降低用户搜索效率。
- 应对建议:同样纳入批量号治理,对均分 <60 的账号限制发布频率。
owner-cluster (mid) - ben7am1n
- 可能原因(假设):批量发布低质插件,可能为测试或刷库行为。
- 影响面:10 个均分 58 的插件造成低质印象,影响作者生态声誉。
- 应对建议:要求作者补充功能说明和测试,否则对低分插件降权或下架。
owner-cluster (mid) - CAI-MH
- 可能原因(假设):批量复制模板,部分未完成 README。
- 影响面:1 个无 README 的插件让用户无法了解用途,增加安装风险。
- 应对建议:强制 README 缺失插件在市场中隐藏,直至补齐文档。
dominant-owner (low) - uckkk
- 可能原因(假设):同一账号批量发布导致作者集中度异常。
- 影响面:规模数字被单一账号抬升 5%,生态规模统计需剔除。
- 应对建议:在公开报告中使用“剔除 top 批量号后的有效插件数”作为补充指标。
npm-stale (low)
- 可能原因(假设):发布流水线未同步 repo 最新 tag;或 npm 包被抢注/错配。
- 影响面:10 个插件 npm 安装用户拿到旧版,可能缺失新功能或存在已知 bug。
- 应对建议:官方提供版本漂移检测工具,并通知作者修复;对官方市场 dsh-market 优先修复。
breaking-release (high)
- 可能原因(假设):rc.2 引入破坏性 API 变更,用于推进架构升级。
- 影响面:依赖旧接口的插件可能出现兼容故障,用户升级后面临失效风险。
- 应对建议:插件作者应立即对照 rc.2 变更适配;官方应提前发布迁移指南并延长 rc 测试期。
8. 下一阶段重点与建议
插件作者:
- 检查 npm 发布状态,确保版本与 repo 一致;利用官方提供的漂移检测。
- 避免进入“侧栏/工作区”等拥挤方向,转向记忆/知识、MCP 集成等平均分较高且供给相对少的领域。
- 补充中文文档,抓住 zhPct 仅 48.3% 的增量用户;完善描述和 topics 以提升可发现性。
生态用户:
- 不依赖 star 选插件,结合 score 和 weekly downloads 交叉验证;对 C/D 级高下载插件(如 LuniteGlaze/dsh-plugin-catalog)优先查源码。
- 关注 S 级 gems(如 SenmuuuuW/dsh-whale-report),主动安装并反馈,帮助高质量插件获得曝光。
- 升级 DSH 前留意 breaking-release 公告,确认常用插件已适配。
DSH 官方:
- 治理批量号:对 uckkk 等账号实施限流、标记,在增长统计中剔除模板号。
- 修复 npm 发布链路:优先解决 dsh-market 的版本漂移,建立发布健康度看板。
- 提升能力标注覆盖率:当前仅 7.3%,推动作者打标,为空白机会识别提供数据基础。
- 建立“高质低用”推荐机制:将 quadrant.gems 自动推送至市场首页或 awesome 列表。
9. 官方动态解读
- dsh 官方仓库 star 281645,最新 release dsh-v0.2.2-rc.3(2026-11-13,非 breaking):说明官方在快速迭代,但 rc 阶段提示稳定性优先,插件作者可暂缓大规模适配直到正式版。
- breaking release dsh-v0.2.2-rc.2 已发布:高风险信号,建议作者立即加入 beta 测试群或查看 changelog,避免用户升级后插件失效。
- 官方市场 dsh-market npm 版本 1.73.20 领先仓库 1.43.0:可能意味着 npm 侧已由其他维护者推进,或仓库更新滞后;官方需明确发布主权,防止用户安装到非官方维护版本。
- 平台侧 deepseek-ai 生态项目 star 高企(如 DeepSeek-V3 104842):模型热度持续引流,但插件生态需从“蹭热度”转向“做深工具”,否则批量号将继续稀释价值。
DSH Ecosystem Insights · 2026-W46(2026/11/09~2026/11/15)
Generated by the DSH Insights pipeline + AI · snapshot 2026-11-16 · heuristic evaluation, not a security audit
1. Core Conclusions
- Growth polluted by batch accounts: signal shows uckkk alone holds 1286 plugins (avg score 56); the 9529 cumulative additions relative to baseline 2026-10-15 need moisture removal; real community activity is only 15.1% active7.
- Quality and adoption severely diverge: billion-context (5170 stars, grade B) has 848k weekly downloads, while sunyuhuirong/deepseek-account with 1 star sees 510k weekly downloads; stars are no longer a reliable quality proxy.
- Top author concentration but low top share: top10Share only 2.2%; Ebony-Vinyl leads with a single 75756-star plugin; the long tail is extremely thick but quality is highly polarized.
- New cohort quality is improving: cohortTrend average rises from 74.1 in early September to 77.1 in November; new author quality is improving, but batch accounts still drag the average down.
- Publishing pipeline has systematic drift: npm publish rate 44.1%; 10 plugins have repo versions ahead of npm (e.g. dsh-market repo 1.43.0 vs npm 1.73.20); users may consistently get outdated builds.
2. Head Plugin Deep Dive
- ranxianglei/billion-context: category Status/Monitoring/Usage; stars 5170, score 86, grade B, weekly 848080; desc mentions 100K tokens context-compression proxy. Zero-star-author can produce high-download tool; downloads are driven by functional demand, not stars; but manifest.no-client-export and npm.version-drift warnings indicate publishing quality issues.
- dsh-market/dsh-market: official market; stars 12524, grade A, weekly 293468, npm package dshmarket. Wins on authoritative distribution, but npm version drift and missing files could cause inconsistent user experience via npm.
- Ebony-Vinyl/dsh-our-free-model: highest stars 75756, grade A, weekly 10812; offers free login-free model calls. High stars but only 10k weekly downloads: heat didn't fully convert to installs; truncated desc and category mismatch warrant review.
- alchaincyf/nuwa-skill: stars 38349, grade B, published false, weekly null. Extremely high stars but not on npm; users install via source/other channels; published false and weekly null indicate measurement blind spots.
- sunyuhuirong/deepseek-account: stars 1, grade B, weekly 510593, desc empty. Extremely low stars with extremely high downloads suggests package squatting/centralized distribution; missing desc and docs flags mean users install without documentation.
3. Author Landscape
- Concentration: top10Share only 2.2%, total 15084 authors, multi 3778 (~25%); star power is diffuse because total star base is huge and long tail is massive.
- Head authors: Ebony-Vinyl (5 plugins, 78424 stars, avg 79.8) exploded via one plugin; alchaincyf (3 plugins, 68392 stars, avg 76.3) relies on two unpublished plugins; dsh-market (1 plugin, avg 90) highest quality but depends on official identity.
- Prolific authors: xby-skill 175 plugins but only 2 stars total and avg 62.2, typical template farm; PerryLink 155 plugins avg 84.7 high-quality prolific; feely0208 104 plugins avg 86.1 also quality; prolific does not equal low quality.
- Connectors: cursoragent (5 plugins, 9984 stars) and Ebony-Vinyl appear in connectors list, showing their hub role; zixin947 3 plugins 78463 stars not in topByStars, possibly star attribution from collaborations.
- Pollution: uckkk 1286 plugins (5% of ecosystem) avg 56, plus xushilianzhuren, ben7am1n batch accounts; author-based scale is inflated. Judgment: ecosystem has both community-driven real creation and traffic-driven batch farming; official must layer governance by quality score and activity.
4. Quality × Adoption Cross-Findings
- SenmuuuuW/dsh-whale-report: S grade 100, 0 star, weekly 175; provides session event daily/weekly reports. Classic high-quality low-adoption; official should list it in awesome or market recommendation.
- Fish-under-sea/dsh-better-reasoning-effort-fish: S grade 100, 2 stars, weekly 203; enhanced fork. Fork improvements buried; version discovery should support fork attribution and recommendation.
- 0x7a76/dsh-extra-paths: S grade 100, 0 star, weekly 317; configures extra writable roots. System-level enhancement very valuable for advanced users, but zero stars; needs scenario-based recommendation.
- LuniteGlaze/dsh-plugin-catalog: C grade 64, 1 star, weekly 103410; empty desc, users install by guess; no star validation; risk of unmaintained code.
- Creakono/dsh-cost-meter: C grade 64, 1 star, weekly 38681; cost calculation errors can mislead budget; high risk at high usage.
- ben7am1n/dsh-browser: D grade 56, 7 stars, weekly 2679; D grade poor engineering; browser automation involves system control; users should inspect source.
5. Ecosystem Theme Map
- Session and memory are core battleground: categoryMap "Session Management" 2406 plugins avg 75.5, "Memory/Knowledge" 749 avg 75.7; topTags session 95, memory 68. Lots of context and memory plugins with upper-middle average; still room for optimization.
- UI/sidebar crowded: "Sidebar/Workspace" 2132 avg 75.6; "Theme/Visual" 958 avg 73.8; topTags ui 73, sidebar 49. Low entry barrier leads to oversupply and lower averages; new authors should avoid pure UI packaging.
- Model/Agent and MCP integration rising: "Model/Agent" 2582 avg 75.8; topTags mcp 56, agent 57. DeepSeek model ecosystem drives plugin integration needs; but capability tag coverage only 1823/25046 (7.3%), many unclassified; official must push tagging.
- Blank opportunities: topTags vision 55, image 54, but no image/vision category in top 10; likely scattered in "Other" 8356; tagging can reveal hidden opportunities.
- Risk direction: "Other" 8356 avg only 72, largest and lowest quality; missing metadata governance; official should enforce categorization to reduce search cost.
6. Ecosystem Health Assessment
| Dimension | Current Value | Interpretation |
|---|---|---|
| Scale | authoritative 25046; authors 15084; medianStars 0 | Very thick long tail; over half plugins have zero stars |
| Growth | weeklyDiff cumulative +9529 vs 2026-10-15; byWeek 2026-11-02 2643, 2026-11-09 327 | Baseline and weekly metrics diverge; real weekly growth may be <1000 |
| Quality Distribution | S 0.95%, A 4.34%, B 46.1%, C 37.2%, D 11.4%; avgScore 74.2 | B/C account for 83%; medium-low quality dominates |
| Activity | active7 15.1% (3780); active30 51.2% (12830) | Nearly half active within 30 days, but only 15% in 7 days—mostly one-shot commits |
| npm Publish Rate | 44.1% (11042/25046) | Over half not distributed via npm; fragmented installation |
| Documentation Coverage | readme 99.8%; zh 48.3%; none 60 | README nearly universal, but Chinese docs below half—unfriendly to domestic users |
cohortTrend interpretation: early September average 74.1, rising to 76-77 from mid-October, November 9 at 77.1 (only 327 samples); new cohort quality improving, but sample size collapse may hide real increment after batch account exit; continuous monitoring needed.
7. Anomalies and Responses
growth-anomaly (mid)
- Possible cause (hypothesis): batch import/farming; or cumulative baseline inflates value, obscuring real weekly increment.
- Impact: growth metric distorted, misleading official resource allocation and user perception of ecosystem vitality.
- Suggestion: official should separate cumulative vs weekly additions, and apply dedup and activity filtering before showing growth.
owner-cluster (mid) - uckkk
- Possible cause (hypothesis): template-generated plugins with no substantive differences.
- Impact: undermines author leaderboard credibility; 1286 low-score plugins pollute search and market recommendations.
- Suggestion: mark or rate-limit uckkk-like batch accounts; manual review for accounts with avg <60 and plugins >50.
owner-cluster (mid) - xushilianzhuren
- Possible cause (hypothesis): template-generated plugins.
- Impact: 52 low-score plugins add long-tail noise, reducing user search efficiency.
- Suggestion: include in batch governance; restrict publishing frequency for avg <60 accounts.
owner-cluster (mid) - ben7am1n
- Possible cause (hypothesis): batch low-quality releases, possibly testing/farming.
- Impact: 10 plugins avg 58 create poor impression, affecting author ecosystem reputation.
- Suggestion: require functional descriptions and tests, otherwise demote or delist low-score plugins.
owner-cluster (mid) - CAI-MH
- Possible cause (hypothesis): batch template copying, some incomplete README.
- Impact: 1 README-less plugin leaves users blind, increasing install risk.
- Suggestion: hide README-less plugins from market until docs provided.
dominant-owner (low) - uckkk
- Possible cause (hypothesis): single account batch publishing causing abnormal concentration.
- Impact: scale inflated by 5% from one account; ecosystem size statistics need exclusion.
- Suggestion: use "effective plugins after excluding top batch accounts" as supplementary metric in public reports.
npm-stale (low)
- Possible cause (hypothesis): publishing pipeline not synced with repo tags; or npm package squatted/mismatched.
- Impact: 10 plugins deliver outdated builds to npm users, missing features or known bugs.
- Suggestion: official provide version drift detection tool and notify authors; prioritize dsh-market fix.
breaking-release (high)
- Possible cause (hypothesis): rc.2 introduced breaking API changes for architecture upgrade.
- Impact: plugins relying on old interfaces may break; users face failures after upgrade.
- Suggestion: plugin authors should adapt to rc.2 immediately; official should publish migration guide early and extend rc testing.
8. Next Focus and Recommendations
Plugin Authors:
- Check npm publish status, ensure version consistent with repo; use official drift detection.
- Avoid crowded directions like sidebar/workspace; turn to memory/knowledge or MCP integration with higher average and less supply.
- Add Chinese docs to capture incremental users (zhPct only 48.3%); improve desc and topics for discoverability.
Ecosystem Users:
- Do not rely on stars; cross-validate score and weekly downloads; inspect source for C/D high-download plugins (e.g. LuniteGlaze/dsh-plugin-catalog).
- Pay attention to S-grade gems (e.g. SenmuuuuW/dsh-whale-report), install and provide feedback to boost quality plugins.
- Before upgrading DSH, watch breaking-release announcements; confirm commonly used plugins are adapted.
DSH Official:
- Govern batch accounts: rate-limit and mark uckkk-like accounts; exclude template farms from growth stats.
- Fix npm publishing pipeline: prioritize dsh-market version drift; build publishing health dashboard.
- Improve capability tag coverage: currently only 7.3%; push authors to tag, providing data foundation for opportunity identification.
- Establish "high-quality low-adoption" recommendation mechanism: auto-push quadrant.gems to market homepage or awesome list.
9. Official Dynamics Interpretation
- dsh official repo 281645 stars, latest release dsh-v0.2.2-rc.3 (2026-11-13, non-breaking): official iterates fast but rc stage signals stability priority; plugin authors can postpone large-scale adaptation until stable.
- Breaking release dsh-v0.2.2-rc.2 published: high-risk signal; authors should join beta or check changelog immediately to avoid user plugin failures after upgrade.
- Official market dsh-market npm version 1.73.20 ahead of repo 1.43.0: possibly npm side advanced by other maintainers or repo lagging; official must clarify publishing sovereignty to prevent users installing non-official versions.
- Platform-side deepseek-ai ecosystem projects high stars (e.g. DeepSeek-V3 104842): model popularity continues to attract traffic, but plugin ecosystem must shift from bandwagon to deep tools, otherwise batch accounts will continue to dilute value.