dsh-recommend: Transparent Plugin Rankings and Recommendations for DeepSeek Harness

dsh-recommend is a transparent ranking and recommendation plugin (DSH plugin) for the DeepSeek Harness (DSH) plugin ecosystem, maintained by zp-home under the MIT license. It solves the problem of discovering quality plugins in the DSH ecosystem by automatically crawling all GitHub repositories tagged with dsh-plugin every 2 hours, scoring and sorting them with a public scoring model, and providing both a web-based leaderboard and a DSH plugin as consumption methods. Core capabilities include a transparent and reproducible scoring mechanism, deep-scan plugin validation, automated data updates, and one data source reused across multiple endpoints (static site, DSH plugin, external tools).

How does the dsh-recommend DSH plugin ensure transparent and trustworthy scoring?

The scoring formula, weights, and all raw data are public in the repository; anyone can clone and run node scripts/sync.mjs to reproduce the results. Additionally, the top 200 entries undergo deep-scan plugin validation (scripts/scan.mjs checks for dsh declarations, @deepseek-ai/* dependencies, cordis config, skills features), and repositories without detected features are excluded and transparently annotated. An official denylist for non-plugins is also public, ensuring the credibility of the rankings.

How to install dsh-recommend into DeepSeek Harness?

Three installation methods are supported: npm install (recommended for users in China), direct GitHub install, and local directory install (fully offline). After installation, restart dsh web to take effect. Here is the command sequence for GitHub direct install:

dsh plugin --profile web add github:zp-home/dsh-recommend
dsh --profile web --dump-config   # should show "# == dsh-recommend" layer
# restart dsh web to take effect

What features does the dsh-recommend web leaderboard offer?

The web version (https://zp-home.github.io/dsh-recommend/site/) features a Neo-Brutalism high-contrast design with prominent top-three medals, four-dimensional signal scores, and 🏅 certified badges. It supports search, category filtering, four sorting options (overall score / popularity / recent update / latest release), pagination, detail expansion (topic tags / license / release date / deep-scan status), and a trend chart for overall score over the last N days, plus one-click copy of install commands. The separate development leaderboard page shows fastest star growth (7/30/90 days), fastest ranking rise, most npm downloads, new entries this week, and certified badges, each with a growth curve sparkline.

How does the dsh-recommend data update mechanism work?

GitHub Actions recalculates and commits the data/ directory every 2 hours, including deep-scan results, historical snapshots, badges, and monthly reports; data is never manually maintained. data/registry.json is the single source of truth, shared by the static leaderboard site, the DSH plugin (model tools + settings page tabs), and external tools; data/history.json provides daily trends, and data/trends.json provides derived development rankings. If the hub directory fetch fails, CI turns red, making signal source health visible at all times.

Which DeepSeek Harness versions and platforms does dsh-recommend support?

As an official dsh.bundle, it depends on @deepseek-ai/dsh-tools, @deepseek-ai/dsh-host-webserver, @deepseek-ai/dsh-client-locale, @deepseek-ai/dsh-client-runtime, @deepseek-ai/dsh-client-ui-settings, and @deepseek-ai/dsh-client-ui-slots, deeply integrated with the DeepSeek Harness web client. The repository was last updated on 2026-08-16, current version 0.3.0, licensed under MIT, free to use and modify.

See more DSH plugins at dsh-plugin.org