AdamPlatin123/dsh-plugin-radar

DSH Plugin Radar – What It Is

DSH Plugin Radar is an open‑source “radar” that continuously discovers, tests, and catalogs plugins for DeepSeek Harness (DSH) – a platform for building AI agents and LLM‑powered applications. The project automatically:

1. Scans GitHub every 6 hours for repositories that look like DSH plugins (by topic and keywords).
2. Deduplicates and filters out private or black‑listed repos.
3. Validates each candidate (checks package.json for the required DSH entry points).
4. Runs a lightweight Kubernetes pod for each plugin (10‑pod concurrency) to see if it can start and talk to a Qwen model in streaming mode.
5. Classifies the result into four status buckets – ✅ compatible, ⚠️ needs adaptation, ⬜ awaiting test, ❌ failed – and records the verdict in a machine‑readable snapshot every 15 minutes.

All of this data is turned into human‑friendly artifacts:

• PLUGINS-ALL.md – a full list of every discovered repo with its status, star count, and category.
• 精选插件榜 – a curated “featured” list of 56 high‑quality plugins, grouped into 11 functional areas.
• 整合包 – ready‑made bundles that pre‑install a set of plugins for specific use‑cases (e.g., UI skins, agent skill packs, desktop distributions).
• JSON API – two stable endpoints (plugins-all.json and plugins.json) that downstream tools (marketplaces, community sites) can consume without any registration.

The radar itself is split into a Radar Engine (engine/) that handles discovery, aggregation, and rendering, and a Test Engine (still being open‑sourced) that runs the Kubernetes validation pods.

---

Who Might Use It?

• Plugin developers – get automatic visibility; just add the dsh-plugin topic to their repo and the radar will pick them up within hours.
• DSH power users – browse the curated lists to find plugins that are already proven to run, instead of trial‑and‑error.
• Marketplace operators – pull the JSON verdicts to display compatibility badges or filter listings.
• Researchers – study the health of the DSH ecosystem (growth, failure rates, categories) via the public snapshots.

---

How It Works (Simplified)

flowchart TB
  A[GitHub Search] --> B[Dedup & Filter]
  B --> C[Static Check (package.json)]
  C -->|plugin| D[K8s Runtime Test]
  D --> E{Verdict}
  E -->|compatible| F[✅]
  E -->|needs adapt| G[⚠️]
  E -->|awaiting| H[⬜]
  E -->|failed| I[❌]
  F & G & H & I --> J[Snapshot (data/snapshots/)]
  J --> K[Render Markdown & JSON]

• Discovery runs every 6 h, with a finer‑grained probe every 15 min.
• Testing uses a 20 s driver that streams a Qwen model (de‑stream) inside a pod.
• Results are stored as immutable snapshots, then fed to the renderers that produce the markdown catalogs and the JSON API.

---

Getting Started

1. Explore the data – open PLUGINS-ALL.md or fetch the JSON via the URL shown in the README.
2. Add your own plugin – add the dsh-plugin topic to your repo and push; the radar will pick it up automatically.
3. Consume the API – example Python snippet is provided in the README to map repo names to their compatibility verdicts.
4. Watch the roadmap – Phase 2 (engine source) is already public; Phase 3 (lightweight and full‑scale test engines) will be released once stable.

---

Quick Glossary

• DSH – DeepSeek Harness, an open‑source framework for building LLM‑driven agents.
• Radar Engine – the part of the project that discovers repos and builds the catalog.
• Verdict buckets – ✅ compatible, ⚠️ needs adaptation, ⬜ awaiting test, ❌ failed.
• 整合包 (Bundle) – a pre‑defined collection of plugins that can be installed together.

---

Bottom Line

DSH Plugin Radar is a real, continuously‑operating service that maps the health and usability of the DeepSeek Harness plugin ecosystem. It provides both a developer‑friendly pipeline and ready‑to‑use curated lists, making it a valuable piece of infrastructure for anyone building or consuming AI‑agent plugins on the DSH platform.