TL;DR
Between August 13 and August 24, 2026, the DeepSeek Harness (DSH) plugin ecosystem exploded. In twelve days, 7,870 new repositories appeared — a rate of 254 per day. That's 24x the pace of the preceding six months combined.
This report analyzes all 10,762 repositories currently indexed by DSH Universe: their languages, categories, activity levels, and what the surge tells us about where this ecosystem is heading.
The surge: 12 days that changed everything
| Period | Repositories created | Daily rate |
|---|
| Feb–Jul 2026 (6 months) | 379 | ~2/day |
| Aug 2026 | 7,870 | 254/day |
| Sep 2026 (through Sep 25) | 1,680 | 67/day |
The single busiest day was
August 14, 2026 — 1,169 new repositories in 24 hours.
Peak days:
2026-08-14 1,169
2026-08-20 1,116
2026-08-15 1,006
2026-08-16 995
2026-08-17 592
2026-08-19 534
This is not organic growth. It's a platform moment — a toolchain becoming popular enough that thousands of developers build on it in the same fortnight. The September rate (67/day) suggests the initial land rush has stabilized into a steadier cadence.
Ecosystem scale
| Metric | Value |
|---|
| Indexed repositories | 10,762 |
| Total GitHub stars | 2,032,949 |
| Projects with ≥1,000 stars | 217 |
| Projects updated in the last 7 days | 1,407 |
| Updated in the last 30 days | 4,261 (39.6%) |
| Distinct topics in use | 6,914 |
Nearly
40% of the ecosystem saw activity in the past month, and 14% in the past week. For a young ecosystem, that's a high churn rate — plugins are being written, forked, and abandoned quickly.
What are people building?
By category
| Category | Count |
|---|
Uncategorized (other) | 5,800 |
| UI | 1,046 |
| Development | 948 |
| Agent sessions | 878 |
| Model / MCP | 509 |
| Data | 458 |
| Operations | 317 |
| Security | 246 |
| Communication | 243 |
| Lifestyle | 227 |
The dominance of
other (54%) is itself a finding:
more than half of new plugins don't fit an established category yet. The ecosystem is still in a "throw things at the wall" phase — taxonomy hasn't caught up with practice.
By language
| Language | Repositories |
|---|
| JavaScript | 5,938 |
| TypeScript | 3,195 |
| Python | 565 |
| HTML | 120 |
| Shell | 114 |
| Rust | 87 |
| PowerShell | 77 |
JavaScript + TypeScript account for 85% of the ecosystem. New entrants skew even harder: among repositories created after August 1, JS/TS make up
91%. This is a web-first plugin culture — consistent with plugins being UI extensions to a desktop harness.
By license
| License | Count | Share |
|---|
| MIT | 8,208 | 76% |
| No assertion | 1,913 | 18% |
| Apache-2.0 | 400 | 3.7% |
| BSD-3-Clause | 117 | 1.1% |
| AGPL-3.0 | 46 | 0.4% |
MIT dominates at 76%. Permissive licensing is the default here — good news for anyone building commercial tooling on top of the ecosystem.
Star distribution: a long tail with a few giants
| Star range | Projects |
|---|
| ≥10,000 | 34 |
| 1,000–9,999 | 183 |
| 100–999 | 324 |
| 10–99 | 540 |
| <10 | 9,681 |
90% of projects have fewer than 10 stars. The ecosystem is overwhelmingly small, hobbyist-scale work — but the top 34 projects carry a combined 1.5M+ stars and define the ecosystem's public face.
The 15 most-starred projects
| Stars | Project | Language |
|---|
| 175,901 | skills | Python |
| 117,078 | graphify | Python |
| 105,112 | caveman | Go |
| 91,602 | open-design | TypeScript |
| 86,384 | taste-skill | JavaScript |
| 74,897 | awesome-claude-skills | Python |
| 51,723 | cherry-studio | TypeScript |
| 48,193 | obsidian-skills | — |
| 47,057 | humanizer | Python |
| 46,919 | CowAgent | Python |
| 46,304 | agentic-awesome-skills | Python |
| 42,553 | i-have-adhd | Python |
| 38,921 | awesome-copilot | JavaScript |
| 31,402 | awesome-gpt-image-2 | JavaScript |
| 30,949 | cli | Rust |
Note the pattern in the top ranks:
"skills" collections and curated lists dominate, not single-purpose plugins. The ecosystem's most valuable artifacts are
collections of capabilities, not individual tools.
Topic analysis: how the ecosystem names itself
| Topic | Repositories |
|---|
dsh-plugin | 8,503 |
deepseek-harness | 8,176 |
dsh | 4,101 |
deepseek | 1,510 |
cordis | 1,196 |
dsh-plugins | 804 |
plugin | 766 |
ai-agent | 504 |
ai-agents | 479 |
mcp | 328 |
79% of repositories tag themselves dsh-plugin — a remarkably high self-identification rate. Developers are consistently labelling their work as belonging to this ecosystem, which makes discovery tractable. The
mcp count (328) shows Model Context Protocol integration is a growing but still minority concern.
What this means
1. The ecosystem had its "moment" in August 2026. The 12-day surge wasn't gradual adoption — it was a step change. If you're evaluating whether to build for DSH, the answer is settled: thousands of developers already decided.
2. It's a web-developer ecosystem. 85% JS/TS, UI as the second-largest category. These are frontend developers extending a desktop harness, not systems programmers.
3. Permissive licensing makes it commercially viable. 76% MIT means tooling, indexes, and commercial products can be built on top without legal friction.
4. Discovery is the unsolved problem. 90% of projects have <10 stars. The ecosystem produces far more software than it can surface. Curation — not creation — is the bottleneck.
5. Activity is high but churn is higher. 40% of the ecosystem updated in the past month, yet the majority are sub-10-star projects. Many are experiments that will be abandoned.
Methodology
Data collected from the GitHub API by DSH Universe, a continuously-updated index of the DeepSeek Harness plugin ecosystem. Repository metadata (stars, forks, language, license, topics, creation and push timestamps) is synced daily. Categorization uses topic-based classification with LLM-assisted fallback.
Corrections and additions welcome — contact@duink.com