生态洞察 · 2026-W45

重点关注 · 7 个异常信号

中本周新增 6469，近 3 周均值 1546——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 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 6%）、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 2 个含 breaking 变更（dsh-v0.2.2-rc.1, dsh-v0.2.2-alpha.3）——插件作者需评估适配

DSH 生态洞察 · 2026-W45（2026/11/02～2026/11/08）

由 DSH Insights 管线 + AI 生成 · 数据快照 2026-11-09 · 启发式评估，非安全审计

核心结论

生态规模虚高，真活跃不足两成：权威插件 21,986 个，但 7 日活跃仅 16.3%，30 日活跃 48.6%——大量插件为一次性提交或冷启动后弃坑。
头部效应极端，一个账号占 6%：作者 uckkk 独占 1286 个插件，均分 56，纯模板批量号拉高规模数据；剔除后生态实际作者集中度更低（top10 作者 star 占比仅 2.6%）。
质量与采用度严重倒挂：下载榜前三中 billion-context（周 84.8 万次）与 dsh-windows-mcp（71.5 万次）均为 B 级且存在 no-client-export 等 warning，而 S 级宝藏插件如 dsh-whale-report 周下载仅 221。
官方 breaking 发布，适配压力陡增：本周两个 release 均含 breaking 变更（dsh-v0.2.2-rc.1、dsh-v0.2.2-alpha.3），大量插件存在 npm 版本漂移（10 个仓库版本领先 npm），安装侧可能拿到旧版兼容性差。
队列质量在回升，但有刷库污染：新插件队列平均分从 9 月初的 74.9 升至 11 月初的 77.3，但同时新增 6469（基线 10/15 累计）中夹杂批量低分号，需持续清理。

头部插件深潜

Ebony-Vinyl/dsh-our-free-model：66,195 star、A 级、周下载 12,704，主打免登录免费调用 DeepSeek V4.1 Flash 等模型。在 DSH 生态中它是流量入口级插件，star 数一骑绝尘，但周下载远低于下载榜头部，说明 star 与真实使用可能脱节。
ranxianglei/billion-context：下载榜第一，周 84.8 万次，B 级（86 分），定位通用上下文压缩代理。高下载与 B 级的落差不只来自 no-client-export 等工程问题，更反映用户对“无限上下文”的刚需压倒了对完美实现的挑剔——这是典型的实用主义选择。
dsh-market/dsh-market：11,502 star、A 级、周下载 32.1 万，DSH 官方插件市场入口。其周下载量被多个同名/仿冒插件（如 KokuYu-sysu/dsh-market-desktop、Sun-0306/dshmarket）分流，说明没有官方命名保护，生态内已出现品牌混淆风险。
helloHub1212/dsh-windows-mcp：227 star、B 级（86 分）、周下载 71.5 万，Windows 桌面自动化 MCP 客户端。极低的 star 与极高的下载极不匹配，可能是通过预装、模板分发或渠道推广获得大量安装，但社区认可度极低——需警惕供应链风险。
T-Auto/dsh-ops：A 级、周下载 5.26 万，星数 5,591，提供 bash/PowerShell 7 等工具。作为基础设施类插件，稳定增长，是少数同时兼顾质量与采用的头部插件，其模式值得复制。

So-what：头部格局显示生态尚未形成“明星插件 + 活跃社区”的正循环，下载量被少数实用工具垄断，而真正高质量插件却无人问津。

作者格局

作者总数 13,081，多插件作者 3,282，但 top10 作者的 star 占比仅 2.6%——这是一个极度去中心化、长尾显著的生态。
Ebony-Vinyl：5 个插件总 star 67,969，平均分 79.8，其旗舰 dsh-our-free-model 拿下 66k star，是当之无愧的头部作者，但平均分未达 80，量产型特征明显。
Minglink：6 个插件总 star 6,390，平均分仅 73，头部作者中质量偏弱，属于堆量但打磨不足。
Fish-under-sea：4 个插件平均分 96.5，总 star 28，是典型的高质低星作者，其 fork/补充版插件（如 dsh-better-reasoning-effort-fish）质量极高但缺乏曝光。
xby-skill：175 个插件，总 star 仅 2，平均分 62.2，是批量模板生产者的极端案例；同时 uckkk 以 1286 个插件均分 56 污染作者榜。这些批量号使“多产”指标失去意义。
判断：生态是 流量驱动而非社区驱动 ——头部作者靠单个爆款吸星，协作网络（connectors）中 cursoragent、Ebony-Vinyl 等节点虽有连接，但整体贡献者网络稀疏，真正持续打磨的优质作者被淹没。

So-what：官方如果要培育社区，必须把推荐位和流量给到 Fish-under-sea 这类高质量作者，并系统性抑制批量号。

质量 × 采用度交叉发现

被低估的宝藏（gems）：
- SenmuuuuW/dsh-whale-report：S 级满分（100）、周下载 221、0 star，从会话日志生成日报/周报的深度代理分析工具——这是生态内稀缺的数据可视化/复盘能力，值得官方收录。
- Shane-Jay/dsh-self-update：S 级满分、周下载 27、2 star，提供 git 安装版自更新能力，解决非 npm 用户的升级痛点，是 DSH 分发渠道的重要补充。
- 0x7a76/dsh-extra-paths：S 级满分、周下载 1,302、0 star，允许配置额外可写根目录，是进阶用户和开发者刚需，却完全未被发现。
低质高用（risks）：
- LuniteGlaze/dsh-plugin-catalog：C 级 64 分、周下载 10.5 万、1 star，无描述，类别“开发/数据”，但作为目录插件被大量安装——用户可能装上了一个不透明、无维护的入口。
- Creakono/dsh-cost-meter：C 级 64 分、周下载 3.9 万、1 star，成本计算插件存在 eng.no-ci 等风险，却因功能刚需被广泛采用。
- Bald0Wang/dsh-imggenerate：C 级 62 分、周下载 2.4 万，图像生成工具，多个同名/相似插件（如 haimuhaimu/dsh-image-gen）分食同一需求，用户容易选到实现差的版本。

So-what：高风险低质插件集中在“目录”和“成本/图像生成”等刚需类别，用户无暇甄别，官方应该在安装前给出更显著的质量警示。

生态主题地图

核心主题聚类（基于 capabilityTags 和 categoryMap）：
- 会话与记忆：session（95）、memory（68）标签密集，但缺乏头部旗舰插件；dsh-session-superagent 是下载不错的 A 级会话管理插件，但未形成品牌。
- 搜索与 UI 增强：search（85）、ui（73）、sidebar（49）表明用户高度关注界面效率，dsh-market 本身就是搜索入口，但侧栏类专用插件（如 omdsh-dev/DSH-better-sidebar）虽 S 级却下载一般。
- 模型/代理编排：agent（57）、mcp（56）、model / 代理 类别插件 2,283 个平均分 75.9，方向拥挤但质量尚可，yjh051108/dsh-routing-suite 和 QingYunA/answer-me-with-html 是 star 较高的代表。
- 状态监控与用量：balance（50）、usage（50）直接对应刚需，T-Auto/dsh-ops、DeepSeek-Balance-Whale-Widget 均在此类且下载不俗，但低质模仿者多。
- 空白机会：记忆 / 知识 类别仅 626 个插件但平均分 76.1 属中上，缺乏杀手级长期记忆插件；开发 / 数据 类别平均分 74.5，插件数 630，工具链分散，有整合机会。
拥挤低质方向：其它 类别有 7,444 个插件，平均分仅 71.8，是最大的“垃圾场”；主题 / 视觉美化（850 个，平均 73.9）也充满低质模板。

So-what：生态正在从单纯“接入模型”转向“会话/记忆/效率增强”，但热门方向鱼龙混杂，真正的空白是“记忆知识管理”和“开发工作流整合”。

生态健康度评估

维度	本期数值	解读
规模	权威插件 21,986	总量庞大，但 head 虚假繁荣，剔除批量号后真实规模需打折
增长	相对 10/15 基线累计 +6,469，3 周均值 1,546	突增主要由批量导入拉动，正常增长有限
质量分布	S 224 / A 957 / B 10336 / C 7880 / D 2589	B 级占近半，C+D 合计 42%，整体中低质量偏多
活跃度 7日	16.3%	极低，大部分插件为一次性提交，生态留存差
活跃度 30日	48.6%	近半插件 30 天内有活动，但其中包含模板号批量操作
npm 发布率	44.1%	不足一半，且 npm 漂移/陈旧率高，安装侧体验差
文档覆盖	zh 覆盖率 48.5%，readme 几乎全覆盖	中文文档覆盖接近一半，但仍有大量无 zh 文档插件

cohortTrend 解读：最近几周新插件队列平均分从 8 月末的 74.9 稳步升至 11 月初的 77.3，说明新加入的插件整体质量在提升（可能因清理或模板号被相对稀释），但需注意 10/19 和 10/26 两周新增数高达 2381、2236，其中可能混入批量低分号，真实高质量作者的增长速度远低于总量增长。

异常与应对

增长异常：本周新增 6469（severity: mid）

可能原因（假设）：有批量导入或自动化刷库行为，从 10/15 基线到 11/2 的新增远超此前周均值。
影响面：生态规模虚高，排行榜被批量号污染，用户检索体验下降。
应对建议：官方应启用创建频率限制和异常模式检测，并对 10 月中旬以来的新增插件进行回溯审核。

作者集群：`uckkk`（severity: mid）

可能原因（假设）：模板复制或脚本批量生成 1286 个低质插件，均分 56。
影响面：占全生态 6%，严重拉低作者榜可信度，使长尾效应被进一步放大。
应对建议：冻结或封禁该账号，清理其插件，并在作者榜单中引入“去批量”过滤逻辑。

作者集群：`xushilianzhuren`（severity: mid）

可能原因（假设）：类似的模板批量号，52 个插件均分 56。
影响面：虽规模远小于 uckkk，但同属批量生产，影响多产作者榜的真实性。
应对建议：评估其插件内容，若确认无实质功能则集中降权或移除。

作者集群：`CAI-MH`（severity: mid）

可能原因（假设）：9 个插件均分 57，1 个无 README，疑似小规模模板复制。
影响面：属于典型长尾污染之一，降低用户对中小作者的信任度。
应对建议：提高发布门槛，例如强制 README 和基础元数据完整，否则不纳入统计。

主导作者：`uckkk`（severity: low）

可能原因（假设）：前述批量号的规模效应，导致单一账号插件数占生态 6%。
影响面：生态健康度统计（如活跃度、增长）被该账号扭曲，需在指标中剔除。
应对建议：在官方仪表盘和报告中标注“排除批量账号”口径，提供净化后的数据。

npm 版本漂移：仓库领先于 npm（severity: low）

可能原因（假设）：插件作者更新了仓库但未同步发布到 npm，或使用了非标准发布流程。
影响面：用户安装到旧版，可能带来兼容性问题和安全风险（尤其 breaking 版本）。
应对建议：插件市场增加“仓库/npm 版本一致性”检查，并在安装时提示版本落后；同时建议作者启用 CI 自动发布。

breaking 发布：dsh 官方两个 release 含 breaking 变更（severity: high）

可能原因（假设）：DSH 核心框架进行 API 调整或安全更新，需要插件适配新接口。
影响面：大量未及时适配的插件可能出现兼容性问题，用户升级后可能功能失效。
应对建议：官方应提供迁移指南并给插件作者充足缓冲期；插件作者需尽快对照 changelog 测试并发布新版本。

下一阶段重点与建议

插件作者：

针对 dsh-v0.2.2-rc.1 的 breaking 变更，阅读官方迁移文档，优先修复 manifest 导出和依赖兼容性。
利用 dsh-market 的 inImsai/covered 标识提升曝光，规范 README 和中文文档，提高 discoverability。
关注队列平均分上升趋势，避免发布无差别模板插件；认真维护 1 个高质量插件胜过 10 个低分填充。

生态用户：

安装前优先选择 A/S 级且 npm 未漂移的插件，对 billion-context 等热门但 B 级的插件保持谨慎，关注其后续修复。
警惕下载量高但 star 极低的插件（如 dsh-windows-mcp），可能来自非透明渠道，检查其源码是否公开可审计。
利用 suggested 列表和 quadrant.gems 发现高质量小众插件（如 dsh-whale-report、dsh-self-update），支持良性生态。

DSH 官方：

立即清理 uckkk 等批量账号，并建立创建频率和内容查重机制，从源头抑制刷库。
在插件市场加入分级警示，对 C/D 级高下载插件展示风险标签，引导用户分流到更优质替代品。
发布 breaking 变更前，提前向头部插件作者发送适配通知，并预留至少 2 周的兼容过渡期。
定期更新 cohortTrend 并公开其构成，让开发者理解生态真实质量走向。

官方动态解读

本周发布 dsh-v0.2.2-rc.1 和 dsh-v0.2.2-alpha.3，均含 breaking 变更，说明官方正在加速核心迭代，插件 API 将面临调整。
官方仓库 deepseek-ai/dsh-superagent-prototype 有 21,256 star，显示官方对 agent 方向的重点投入，可能影响后续插件能力边界。
平台生态中 deepseek-ai/DeepSeek-V3 star 破 10 万，模型影响力持续外溢，插件集成需求会继续增长。
官方未在数据中体现对批量号的直接治理动作，但 cohortTrend 上升可能暗示已有一定过滤机制，需观察后续。

DSH Ecosystem Insights · 2026-W45（2026/11/02～2026/11/08）

Generated by the DSH Insights pipeline + AI · snapshot 2026-11-09 · heuristic evaluation, not a security audit

Core Conclusions

Ecosystem scale is inflated; real activity is under 20%: Authoritative plugins 21,986, but 7-day active only 16.3%, 30-day 48.6% — many plugins are one-time submissions or abandoned after cold start.
Extreme head concentration: one account holds 6% of all plugins: Author uckkk owns 1286 plugins with average score 56, a pure template farm inflating scale; top10 authors account for only 2.6% of total stars, indicating a highly decentralized long tail.
Quality and adoption are severely inverted: Top downloads billion-context (weekly 848k) and dsh-windows-mcp (715k) are B-grade with warnings like no-client-export, while S-tier gems like dsh-whale-report get only 221 weekly downloads.
Official breaking releases raise adaptation pressure: Two releases this week both contain breaking changes (dsh-v0.2.2-rc.1, dsh-v0.2.2-alpha.3), and 10 plugins have repo versions ahead of npm, risking outdated installs.
Cohort quality is recovering, but batch pollution persists: New plugin cohort average score rose from 74.9 in early Sept to 77.3 in early Nov, yet cumulative +6469 since 10/15 baseline includes low-score batch submissions.

Head Plugin Deep Dive

Ebony-Vinyl/dsh-our-free-model: 66,195 stars, grade A, weekly downloads 12,704, offering free model access without login. It's a traffic-entry plugin but weekly downloads far lower than download leaders, suggesting stars and actual usage may be disconnected.
ranxianglei/billion-context: #1 in downloads, weekly 848k, grade B (86), a universal context-compression proxy. The gap between high downloads and B-grade reflects user demand for "infinite context" overriding engineering perfection — a pragmatic choice.
dsh-market/dsh-market: 11,502 stars, grade A, weekly 321k, the official plugin market entry. Its downloads are fragmented by copycats like KokuYu-sysu/dsh-market-desktop and Sun-0306/dshmarket, showing brand confusion risk without official namespace protection.
helloHub1212/dsh-windows-mcp: 227 stars, grade B (86), weekly 715k, Windows desktop automation MCP client. Extremely low stars vs high downloads suggests pre-installation or channel promotion with low community trust — a supply-chain risk.
T-Auto/dsh-ops: grade A, weekly 52.6k, 5,591 stars, provides bash/PowerShell 7 tools. A rare head plugin balancing quality and adoption; its model is worth replicating.

So-what: The head landscape shows no virtuous cycle of "star plugins + active community"; downloads are monopolized by a few utility tools while high-quality plugins go unnoticed.

Author Landscape

Total authors 13,081, multi-plugin authors 3,282, yet top10 authors' star share is only 2.6% — an extremely decentralized, long-tail ecosystem.
Ebony-Vinyl: 5 plugins, total stars 67,969, avg 79.8, flagship dsh-our-free-model with 66k stars; a top author but avg below 80, indicating mass production.
Minglink: 6 plugins, total stars 6,390, avg only 73, weak quality among head authors.
Fish-under-sea: 4 plugins, avg 96.5, total stars 28, a classic high-quality low-star author; forks like dsh-better-reasoning-effort-fish are excellent but lack exposure.
xby-skill: 175 plugins, total stars only 2, avg 62.2, extreme template producer; uckkk with 1286 plugins avg 56 pollutes author rankings.
Verdict: Ecosystem is traffic-driven, not community-driven — head authors rely on viral stars; connector nodes like cursoragent and Ebony-Vinyl exist but contributor network is sparse, drowning quality maintainers.

So-what: To cultivate community, DSH must feature high-quality authors like Fish-under-sea and systematically suppress batch accounts.

Quality × Adoption Cross-Discovery

Underrated gems:
- SenmuuuuW/dsh-whale-report: S-tier perfect score (100), weekly 221, 0 stars, generates daily/weekly reports from session logs — a rare visualization/review capability worth official curation.
- Shane-Jay/dsh-self-update: S-tier perfect, weekly 27, 2 stars, provides self-update for git installs, solving non-npm users' upgrade pain.
- 0x7a76/dsh-extra-paths: S-tier perfect, weekly 1,302, 0 stars, allows extra writable roots — an advanced user/dev essential completely undiscovered.
Low-quality high-use risks:
- LuniteGlaze/dsh-plugin-catalog: C-grade 64, weekly 105k, 1 star, no description, but widely installed as a catalog — users may get an opaque, unmaintained entry.
- Creakono/dsh-cost-meter: C-grade 64, weekly 39k, 1 star, cost calculator with eng.no-ci risk, adopted due to rigid demand.
- Bald0Wang/dsh-imggenerate: C-grade 62, weekly 24k, image generation with multiple similar clones (e.g., haimuhaimu/dsh-image-gen), users easily pick poor implementations.

So-what: High-risk low-quality plugins concentrate in "catalog" and "cost/image generation" categories; official should show clearer quality warnings before install.

Ecosystem Theme Map

Core theme clusters (based on capabilityTags and categoryMap):
- Session & Memory: session (95), memory (68) tags dense but lack a flagship plugin; dsh-session-superagent is a decent A-grade session manager but no brand.
- Search & UI Enhancement: search (85), ui (73), sidebar (49) show high interest in interface efficiency; dsh-market itself is a search entry, but dedicated sidebar plugins like omdsh-dev/DSH-better-sidebar (S-tier) get low downloads.
- Model/Agent Orchestration: agent (57), mcp (56), category "Model/Proxy" has 2,283 plugins avg 75.9 — crowded but okay quality; yjh051108/dsh-routing-suite and QingYunA/answer-me-with-html are high-star representatives.
- Status Monitoring & Usage: balance (50), usage (50) map to hard needs; T-Auto/dsh-ops, DeepSeek-Balance-Whale-Widget perform well but attract many low-quality imitators.
- Opportunity gaps: Memory/Knowledge category has only 626 plugins with avg 76.1 (above medium), lacking a killer long-term memory plugin; Development/Data category avg 74.5, 630 plugins, fragmented toolchain, ripe for integration.
Crowded low-quality directions: Others category has 7,444 plugins, avg only 71.8, the largest dumping ground; Theme/Visual (850, avg 73.9) also full of low-quality templates.

So-what: The ecosystem is shifting from pure "model access" to "session/memory/efficiency enhancement", but hot areas are mixed; real gaps are "memory knowledge management" and "development workflow integration".

Ecosystem Health Assessment

Dimension	Current Value	Interpretation
Scale	Authoritative 21,986	Huge total, but head is fake prosperity; real scale needs discount after removing batch accounts
Growth	Cumulative +6,469 vs 10/15 baseline; 3-week avg 1,546	Spike driven by batch imports, limited organic growth
Quality Distribution	S 224 / A 957 / B 10336 / C 7880 / D 2589	B-grade nearly half; C+D 42%, overall mid-low quality
7-day Activity	16.3%	Extremely low; most plugins are one-time submissions, poor retention
30-day Activity	48.6%	Nearly half active in 30 days, but includes batch operations
npm Publish Rate	44.1%	Less than half; npm drift/stale high, poor install experience
Documentation	zh coverage 48.5%, readme near full	Chinese docs cover half, but many lack zh docs

cohortTrend interpretation: Recent new plugin cohorts show average score rising from 74.9 (late Aug) to 77.3 (early Nov), indicating overall quality of new plugins is improving (possibly due to cleanup or dilution), but weeks 10/19 and 10/26 had huge additions (2381, 2236) likely mixed with batch low-score accounts; real quality-author growth far slower than total growth.

Anomalies and Responses

Growth anomaly: 6469 added this week (severity: mid)

Possible cause (hypothesis): Batch import or automated farming; additions since 10/15 far exceed prior weekly average.
Impact: Ecosystem scale inflated, rankings polluted, user search experience degraded.
Recommendation: Enforce creation rate limits and anomaly detection; retroactively audit new plugins since mid-Oct.

Owner cluster: `uckkk` (severity: mid)

Possible cause (hypothesis): Template copying or scripted generation of 1286 low-quality plugins, avg 56.
Impact: Occupies 6% of ecosystem, severely lowers author ranking credibility and amplifies long tail.
Recommendation: Freeze/ban the account, clean up its plugins, and introduce batch-filtering in author rankings.

Owner cluster: `xushilianzhuren` (severity: mid)

Possible cause (hypothesis): Similar template batch account, 52 plugins avg 56.
Impact: Smaller scale but still batch production, affecting prolific author ranking authenticity.
Recommendation: Evaluate plugin content; if no substantial function, downgrade or remove centrally.

Owner cluster: `CAI-MH` (severity: mid)

Possible cause (hypothesis): 9 plugins avg 57, 1 without README, suspected small-scale template copying.
Impact: Typical long-tail pollution, reducing user trust in small/medium authors.
Recommendation: Raise publishing threshold, e.g., require README and basic metadata completeness.

Dominant owner: `uckkk` (severity: low)

Possible cause (hypothesis): Scale effect of batch account, single account holding 6% of plugins.
Impact: Health metrics (activity, growth) distorted; need to exclude from indicators.
Recommendation: Label "batch-account excluded" in official dashboards and reports; provide cleaned data.

npm version drift: repo ahead of npm (severity: low)

Possible cause (hypothesis): Authors updated repo but not npm, or non-standard release process.
Impact: Users install outdated versions, risking compatibility and security issues (especially with breaking changes).
Recommendation: Add repo/npm version consistency check in plugin market; warn users; encourage CI auto-publish.

Breaking release: two DSH releases with breaking changes (severity: high)

Possible cause (hypothesis): DSH core framework API adjustments or security updates requiring plugin adaptation.
Impact: Many unadapted plugins may break compatibility; user upgrades may cause failures.
Recommendation: Official should provide migration guides and grace period; plugin authors must test against changelog and release new versions promptly.

Next-Stage Priorities and Recommendations

Plugin Authors:

For dsh-v0.2.2-rc.1 breaking changes, read migration docs and fix manifest exports and dependencies first.
Use dsh-market inImsai/covered flags to boost discoverability; standardize README and Chinese docs.
Avoid publishing undifferentiated template plugins; maintain one high-quality plugin rather than ten low-score fillers.

Ecosystem Users:

Prefer A/S-tier plugins without npm drift; be cautious with popular but B-grade plugins like billion-context, monitor fixes.
Be wary of high-download low-star plugins like dsh-windows-mcp; check if source is public and auditable.
Use suggested list and quadrant.gems to discover high-quality niche plugins (e.g., dsh-whale-report, dsh-self-update).

DSH Official:

Immediately clean up batch accounts like uckkk; establish creation frequency and content dedup mechanisms.
Add tier warnings in plugin market for C/D high-download plugins, guiding users to better alternatives.
Before breaking releases, notify top plugin authors in advance and reserve at least 2-week transition period.
Regularly update and publish cohortTrend composition so developers understand real quality trends.

Official Dynamics Interpretation

Released dsh-v0.2.2-rc.1 and dsh-v0.2.2-alpha.3 this week, both with breaking changes, signaling accelerated core iteration and upcoming plugin API adjustments.
Official repo deepseek-ai/dsh-superagent-prototype has 21,256 stars, indicating strong investment in agent direction, potentially affecting plugin capability boundaries.
Platform deepseek-ai/DeepSeek-V3 exceeded 100k stars, model influence spilling over, plugin integration demand will continue to grow.
No direct batch-account governance visible in data, but cohortTrend rise may hint at existing filtering; monitor further.