What 900 plugins taught me about an ecosystem's first week
What 900 plugins taught me about an ecosystem’s first week
Summary. DeepSeek Harness’s plugin registry crossed 908 entries within its first week of tracking — up from ~554. This post reads that data the way an ecosystem map-maker should: what the category distribution says, what the verification gap quantifies, and what happens next. All numbers are live registry data, not estimates.
Growth
Week 1 registry: ~554
Current registry: 908 (+~64%)
For a developer-preview harness whose upstream PR channel is closed, that is not a hype curve — it is people building against the current API while the API is still moving. Every one of those 908 entries is a bet on the platform.
Category distribution
| Category | Count | Share |
|---|---|---|
| tools | 208 | 23% |
| ui | 208 | 23% |
| dev | 108 | 12% |
| workflow | 66 | 7% |
| notify | 59 | 6% |
| session | 58 | 6% |
| memory | 48 | 5% |
| fun | 36 | 4% |
| theme | 35 | 4% |
| skill | 32 | 4% |
| model | 30 | 3% |
| market | 19 | 2% |
Two readings:
- The platform is past “experiment”. tools + ui = 46% of the ecosystem. The community is shipping capability, not just demos.
- The agent-native categories are forming. memory (48), workflow (66), session (58), skill (32) — these are not ported tools, they are harness-shaped primitives. That is the sign of a platform, not a wrapper.
fun (36) and theme (35) are the health signal: an ecosystem that people personalize is an ecosystem people stay in.
The verification gap
verified: 21 / 908 (2.3%)
This is the number that matters. Discovery is solved — 908 entries, browseable and searchable. Trust is not: 2.3% of the registry carries the narrow “curator exercised CI + install + runtime smoke” claim. The gap is not a criticism of the 887 unverified entries; it is the quantified reason the doctor / poison-guard / registry-contract work exists. An ecosystem with 900 plugins and no verification layer is a discovery win and a trust liability at the same time.
What happens next
- The verified fraction needs to climb, and the
verifiedclaim must stay narrow (CI + install + runtime smoke, never “audited”). - The agent-native categories (memory, workflow, session) will keep growing — they are where the harness differs from a CLI wrapper.
- Registry interop (one contract, RFC #1846) determines whether 908 entries in one storefront become 908 entries in every storefront.
The weekly map continues at dsh-ecosystem, with every number traceable to a live source.
中文摘要
DSH 插件注册表第一周从约 554 涨到 908（+64%）。分类分布说明平台已经过了 “实验期”：tools+ui 占 46%，社区在交付能力而不只是 demo；memory/workflow/ session/skill 这些 agent 原生类别正在成型——这是平台的信号不是包装器的信号。 最关键的数是验证缺口：908 个里只有 21 个 verified（2.3%）。发现的问题解决 了，信任没有——这正是 doctor/投毒扫描/注册表契约存在的量化理由。下一步： verified 比例要涨且定义保持窄、agent 原生类别继续长、注册表互操作（#1846） 决定 908 个条目是”一个商店”还是”所有商店”。