Introduction

DSH adopts a “plugins for everything” architecture, with community plugins continuously growing under the dsh-plugin topic. When developers search for plugins, common methods include browsing GitHub search results, scanning awesome lists, or judging based on star counts in READMEs—these approaches either lack comprehensive coverage, miss a unified dimension, or make it difficult to query or install directly within DSH.

Below introduces the online tool plugin dsh-recommend maintained by zp-home. It automatically scrapes all dsh-plugin topic repositories on GitHub every 5 hours, sorts them using a public scoring model, and supplies the same data to a static ranking site, DSH model tools, and the settings page. The SkillHub directory classifies it as an “online tool,” with the current version being v0.3.0.

What Is This

dsh-recommend is a transparent ranking and recommendation service for the DSH plugin ecosystem, with its repository at github.com/zp-home/dsh-recommend. Maintainer zp-home has publicly disclosed the scoring formula, weights, and raw data in the repository; anyone can clone it and run node scripts/sync.mjs to recalculate the rankings.

Data uses data/registry.json as the single source of truth. GitHub Actions performs a full recalculation every 5 hours and commits the data/ directory, including rankings, historical snapshots, badges, and derived development rankings. The current pipeline covers over 2,200 dsh-plugin topic repositories. After excluding placeholders, empty repositories, the official core, and items where deep scans did not detect plugin characteristics, approximately 1,900+ are included in the rankings.

Core Features

Public Scoring and Filtering

The composite score formula is: 0.35×Maintainability + 0.30×Popularity + 0.20×Quality + 0.15×Ecosystem. Details are in the repository’s docs/scoring.md. Excluded items remain in data/registry.json with reasons attached (fork, archived, empty repository, no description, placeholder characteristics, official core, deep scan did not detect plugin characteristics, etc.).

The top 200 items in the rankings undergo deep scanning via scripts/scan.mjs to detect dsh declarations, @deepseek-ai/* dependencies, cordis configurations, and skills characteristics. Repositories where characteristics are not detected are excluded from the rankings and transparently annotated. The official core and non-plugin deny list are located in scripts/exclude-list.json.

Static Ranking Site (No Installation Required)

You can browse rankings without installing the plugin. Open the composite ranking homepage:

https://zp-home.github.io/dsh-recommend/site/

The page supports search, category filtering, four sorting methods (composite score / popularity / last updated / latest release), pagination, detail expansion, and a composite score trend chart over the last N days. It also allows one-click copy of installation commands.

The development ranking is a separate page, covering star growth (7/30/90 days), ranking ascent, npm downloads, new entries this week, and featured certifications:

https://zp-home.github.io/dsh-recommend/site/rankings.html

Raw JSON data is also public: data/rankings.json (updated every 5 hours), data/history.json (daily trends), data/trends.json (derived from development rankings).

Model Tools and Settings Page within DSH

After installation, five types of model tools are available in DSH:

Tool Purpose
rank_plugins Query rankings, filterable by category and dimensions
search_plugins Search for plugins
recommend_plugins Recommend based on goals, supporting Chinese-English synonyms and the keywords parameter
trend_plugins Development rankings (star growth, ranking ascent, downloads, new entries, featured)
sync_registry Refresh local data, report hub and deep scan health status

The settings page path is “Settings → Plugins → Plugin Rankings,” providing a compact ranking list, one-click refresh, one-click installation, installation command copying, detail expansion, certification badges, and trend charts. It supports both zh/en languages and adapts to DSH’s light/dark themes.

The plugin also checks the Git/NPM direct dependency versions of the current profile, with default reminders only for updates. Clicking “Update” executes the official dsh plugin update, after which DSH must be manually restarted.

Capabilities for Plugin Authors

Ranked projects automatically generate shields badges every 5 hours (data/badges/<owner>__<name>.json), which can be referenced in READMEs. Featured certification is applied for by submitting a collection/certification Issue. After approval, a 🏅 identifier is displayed; certification is a display-layer incentive and does not affect scoring.

Installation and Activation

DSH plugins run with the permissions of the current DSH process. Before installation, you should read the repository source code and MIT license to determine suitability for your environment.

dsh plugin --profile web add dsh-recommend

After installation, restart dsh web for it to take effect.

Method B: Direct GitHub Installation

dsh plugin --profile web add github:zp-home/dsh-recommend
dsh --profile web --dump-config

The dump-config output should include the # == dsh-recommend configuration layer. Restart dsh web as well.

Method C: Local Directory Installation

dsh plugin --profile web add /your-path/dsh-recommend

Suitable for scenarios where the plugin directory is copied offline.

Network Tips for China

Plugin ranking data is fetched by default from raw.githubusercontent.com (via the sync_registry tool). If this domain is unreachable, you can edit cordis.patch.yml in the installed package (path like node_modules/dsh-recommend/cordis.patch.yml) and change dataUrl to the jsDelivr CDN:

https://cdn.jsdelivr.net/gh/zp-home/dsh-recommend@main/data/registry.json

After modification, restart DSH. CDN data may have a cache delay of several hours.

Typical Usage

Browse Rankings and Copy Installation Commands in the Browser

  1. Open https://zp-home.github.io/dsh-recommend/site/.
  2. Use search or category filtering to locate the target plugin.
  3. Expand details, copy the one-click installation command provided on the page, execute it in the terminal, and restart dsh web.

Use the Model to Query Rankings or Get Recommendations in Conversations

After installation and restart, you can directly call tools in DSH conversations, such as querying the composite ranking, searching by keywords, recommending plugins based on scenarios, or viewing development indicators like star growth. To refresh the local cache, call sync_registry.

Recalculate the Data Pipeline Yourself

If you need to audit or debug ranking logic locally, run the following in a Node 18+ environment:

node scripts/sync.mjs
node scripts/validate.mjs
node scripts/smoke.mjs

The deep scan step requires GITHUB_TOKEN; if not set, it uses GitHub API’s unauthenticated rate limit, which is sufficient for one run but automatically skips deep scanning. The token is automatically injected in CI.

Applicable Scenarios and Notes

Who Is It For

  • Developers who want to quickly query rankings, search, and get goal-based plugin recommendations within DSH.
  • Users who need to compare plugins’ maintainability, popularity, quality, and ecosystem signals, rather than just looking at star counts.
  • Plugin authors who wish to display transparent scoring badges or apply for featured certification.

What to Note

  • This plugin and the SkillHub directory are part of the community ecosystem and have no official affiliation with DeepSeek / High-Flyer; they are not an official app store.
  • The repository README mentions static security scanning (see docs/security-scanning.md). Not hitting rules does not equate to security; the prompts are for reference only and do not affect scoring or installation eligibility.
  • The plugin only reads registry.json and displays it; it does not execute the code of collected plugins. However, the dsh-recommend you install itself runs with DSH process permissions, so please review the source code before installation.
  • Features on the settings page, such as ranking labels, data routing, and one-click installation, are for --profile web; host-side model tools can be used in other profiles, as specified in the repository’s cordis.patch.yml.

Conclusion

dsh-recommend transforms the collection, scoring, filtering, and display of the DSH plugin ecosystem into an auditable automated pipeline, using the same registry.json to serve the static site, model tools, and settings page. If you are expanding your DSH plugin stack, you can first browse the ranking site and then install as needed to your web profile.