A plugin trust & discovery registry for DSH. We answer four questions about every plugin: what it is, whether it's safe, whether you can install it, and whether it's still maintained. Safety is answered in two layers: an automated static scan across the registry, plus a model-driven dynamic security audit of the entries flagged as needing review.

How do I install a plugin?

Copy the install command from any card (dsh plugin --profile web add ) and run it in your terminal. Commands are generated from the plugin's manifest and verified at L3 or above.

What do the verification levels (L1-L5) mean?

L1: found on GitHub. L2: manifest is valid. L3: install spec parsed. L4: installed in an isolated sandbox. L5: plugin loads and runs (smoke-tested). The rules are public on the Verification page.

What does "Security: pending" mean?

An automated static scan (dependencies, permissions, secrets, supply chain) has not run for this artifact yet. Results are shown on the detail page after the next nightly scan. Beyond that first layer there is a second, model-driven dynamic security audit that reads source context for entries flagged as needing review — it runs only on that subset, and when it reaches a conclusion that conclusion is authoritative: the published tier follows the finding counts of that audit — any critical finding makes it critical, otherwise any high finding makes it high, otherwise any medium finding makes it medium, otherwise low, so a softer conclusion lowers the tier too — with the static scan tier (vet primary, dsh backstop) applying only when there is no conclusion. Exemptions stay manual (the audit never writes the suppression list). Human review can withdraw a conclusion it considers wrong, which reverts the tier to the static value immediately (no pipeline rerun needed); entries outside the scanning scope follow the existing policy — they are skipped in the registry-wide risk statistics and their badge reads unknown. Both layers are automated, not a manual review.

How can I submit my plugin?

Submit a GitHub URL — it enters the review queue, gets L1-L2 checks within 24h, and appears on the site. Add the dsh-plugin or dsh-pack topic to your repo and dsh.so discovers it automatically too.

Is the site a marketplace?

No. dsh.so is a verification and intelligence layer over the official registry — no payments, no transactions, no ratings. We verify and inform; you install from the official sources.

Plugin Trust & Discovery Registry

Discover useful dsh plugins

We answer four questions about every plugin: What is it? Is it safe? Can I install it? Is it still maintained? Verified against the GitHub API, plus automated security scans and sandbox install testing — nothing padded.

DSH is an open-source agent harness developed by DeepSeek AI. It uses an architecture where everything is a plugin, and is powered by Cordis. Source: DSH README (https://github.com/deepseek-ai/deepseek-harness) · Architecture (https://github.com/deepseek-ai/deepseek-harness/blob/master/docs/architecture.md)

Browse all 28,644 artifacts How verification works

Browse:Developer workbench Creative & visual Finance & ops Study buddy Teaching Long-form writing Research Editor's picks

What is it?

28,644

artifacts indexed · browse →

Is it safe?

28,181

security scanned · 16,735 gold trust

Can I install it?

26,429

install-tested (L4) · 17,817 also run-tested (L5)

Is it maintained?

15,103

active repos · ecosystem →

Last updated 2026-11-12 16:20:54 UTC+8 · security scan 2026-11-12 07:40:14 UTC+8

DSH v0.2.1-alpha.4+ · custom index sources

Add dsh.so as a DSH plugin index source

Add either URL and dsh’s Discover tab lists, searches and installs them; both can be added.Full guide →

Full marketinstall-tested + low risk

https://cdn.dsh.so/market/index.json

Curatedtop 200 × 6 boards

https://cdn.dsh.so/market/top-rank.json

Setup:①Enable “Custom plugin indexes” under Experimental②Discover → Add index source③Search and install

What can developers do with dsh plugins?

Developers use dsh plugins to extend DSH with tools, terminal commands, memory, vision, notifications, databases, and integrations. Counts come from public repository metadata. Charts list plugins that passed L5 run testing — skins & themes from L4 sandbox-install verification — plus ecosystem apps (desktop / web shells, and entries that install but mount no plugin capability): ecosystem plugins still show their install command because the package installs, while ecosystem apps link to their source repository because they are not installed via dsh plugin add.

View all →

Files & data

Files & dataAll 2323 →

Includes File · Storage · Database

Memory & knowledge

Memory & knowledgeAll 3585 →

Includes Memory

Models & agents

Models & agentsAll 2195 →

Includes Agent orchestration

Dev workflow

Dev workflowAll 1953 →

Includes GitHub integration · Code review

Connect & integrate

Connect & integrateAll 2123 →

Includes Notifications

Look & feel

Look & feelAll 2012 →

Includes Skins & Themes

Office & docs

Office & docsAll 2318 →

Includes Office automation · Vision / OCR

Runtime & terminal

Runtime & terminalAll 1021 →

Includes Terminal / Shell

Creation & media

Creation & mediaAll 1403 →

Includes Creator & social

Web & search

Web & searchAll 2100 →

Includes Web search

Interface add-ons

Interface add-onsAll 627 →

Includes

Security & audit

Security & auditAll 777 →

Includes

Finance & trading

Finance & tradingAll 584 →

Includes

Fun & life

Fun & lifeAll 311 →

Includes

Desktop & clients

Desktop & clientsAll 722 →

Includes

Workbench

WorkbenchAll 396 →

Includes Desktop / web / mobile clients, launchers

Ecosystem plugins

Ecosystem pluginsAll 4350 →

installs but mounts no plugin capability (L4 passed, missing from the L5 manifest) — shown as L5 · ecosystem plugin, install slot still shows the install command

What was recently updated?

Plugins with recent commits, ordered by last commit

View all

These plugins had the most recent commits at the time of the registry refresh, ordered by last commit.

@KongFangXun/sofagent

sofagent is an open-source FDE Harness for AI coding agents — on entry, write your business judgment into files (workflows, ontology, AI nodes); after departure, audit every change against them. 28 codified audit rules, tamper-evident chain, snapshots, and an MCP toolset (104 tools, 12 plugins). Everything can be FDEing.

★ 55L5 · Ecosystem pluginOut of scopeUpdated 11-12

@qishuilalala/dsh-voice-mode

Full-duplex voice mode for DeepSeek Harness: streaming ASR to editable drafts, wake word, Edge TTS with subtitles, barge-in, no API key.

★ 32L5 install verifiedLowRiskUpdated 11-12

@vshulcz/deja-vu

Memory for coding agents that indexes past sessions on disk and recalls them locally without embeddings.

★ 1518L5 · Ecosystem pluginLowRiskUpdated 11-12

@better-er/dsh-peak-block

Automatically blocks or reroutes expensive official DeepSeek API requests during specified peak hours, with configurable target provider.

★ 8L5 install verifiedLowRiskUpdated 11-12

@better-er/dsh-write-rule-guard

DSH plugin that intercepts edit/write tool content matching configurable regex rules, default full-width parentheses, with pwsh lock.

★ 4L4 install testedLowRiskUpdated 11-12

@zhu1090093659/dsh-web

Web plugin aggregation ecosystem for DeepSeek Harness, distributing everything as plugins via a creative workshop.

★ 10365L5 · Ecosystem appLowRiskUpdated 11-12

@harness-home/harness-ai-plugins

Community plugin catalog for Harness AI that scans the public npm registry and publishes a snapshot.

★ 0L5 · Ecosystem pluginLowRiskUpdated 11-12

@zbc0315/dsh-synomega

AI-driven organic reaction prediction, retrosynthesis, and route planning with in-chat visualization for DeepSeek Harness.

★ 0L5 install verifiedLowRiskUpdated 11-12

@coeasy/dsh-go

全自动同步的 DeepSeek Harness 插件市场导航站，提供 RESTful API 和 0 元部署。

★ 23L5 · Ecosystem pluginLowRiskUpdated 11-12

What is newly listed?

Recently published plugins

View all

These plugins were added to the registry most recently.

@maverikhe-cpu/zhuangzai-dsh-workbench

Zhuangzai construction delivery workbench bundling the Yiqizhuang MCP server, 10 management skills and scenario entry points for DSH.

Added 2026-11-10L3 install specLowRisk

@Tinxyoo/dsh-file-mention

DSH Web/Desktop plugin that lets you right-click a file in the directory tree or quote selected file content straight into the input box.

Added 2026-11-10L3 install specLowRisk

@mznxbcvsdc/dsh-shot-send

One-click desktop region selection that drops the screenshot straight into the current DSH chat input, all handled in memory.

Added 2026-11-10L3 install specLowRisk

@yuxiangwhn/dsh-chat-wallpaper

Puts an image wallpaper behind the DSH chat UI with frosted composer and bubbles, auto-blur when a conversation has content, and frosted backdrops behind overlays.

Added 2026-11-10L3 install specLowRisk

@retweve/dsh-plugin-pinky

Connects Pinky's phone-control MCP server so the agent can operate a real Android or iPhone, asking you to approve risky steps.

Added 2026-11-10L3 install specLowRisk

@tyrantblue/dsh-trek-mcp

Connects DeepSeek Harness to a remote TREK travel-planner MCP server with a settings page, automatic OAuth tokens and idle-session recovery.

Added 2026-11-10L3 install specLowRisk

@HimLeoing/dsh-task-progress

Tracks task progress in DeepSeek Harness with local execution evidence, teammate progress and optional AI observations.

Added 2026-11-10L3 install specLowRisk

@yan92626/dsh-task-butler

Task queue and execution board for DeepSeek Harness with batch publishing, auto-claiming, AI daily reports, webhooks and scheduled reviews.

Added 2026-11-10L3 install specLowRisk

@bob868335-rgb/dsh-lightvela

DSH plugin that registers the LightVela OpenAI-compatible gateway as a model provider with model discovery, streaming and multi-key failover.

Added 2026-11-10L3 install specLowRisk

Not sure what to install? Start with what you want to do

Tell us your goal, we point at the plugin — every scene links to install-tested options

🔔

I want DSH to notify me

Add a Telegram / email notification plugin

Plug into IM or email, get reports when jobs finish

Browse this scene

🔍

I want it to search the web

Add a web-search plugin

Give your agent live internet access

Browse this scene

🧠

I want it to remember things

Add a long-term memory plugin

Persist preferences and context across sessions

Browse this scene

💻

I want it to run commands

Add a terminal / remote-ops plugin

Run commands, manage files, connect over SSH

Browse this scene

👁️

I want it to understand images

Add a vision plugin

Read screenshots and recognize image content

Browse this scene

🎨

I want a different look

Add a skin / theme plugin

Swap colors, wallpapers and sidebar chrome

Browse this scene

🖥️

I want a workbench client

Browse workbench clients (desktop · web)

Desktop and web client shells for DSH, plus sidebar and workbench plugins

Browse this scene

📊

I want it to handle office files

Add a Word / Excel / PowerPoint plugin

Generate spreadsheets, batch-edit documents, read and write slides

Browse this scene

📣

I want to run social media

Add a creator / multi-platform publishing plugin

Publish and track content on WeChat, Xiaohongshu, Douyin and Bilibili

Browse this scene

How does dsh.so verify plugins?

Five levels, from "found" to "run-tested" — rules are public.

Read the rules

dsh.so checks each entry against the GitHub API and applies a public five-level scale from repository found (L1) to run tested (L5) — verification currently covers install-tested (L4) and, for smoke-tested entries, run-tested (L5).

100%

Found

Repo accessible, non-empty, README exists

28,644 artifacts reached

100%

Structured

Manifest / package.json valid, key fields complete

28,644 artifacts reached

100%

Install Spec

Install command parsed, DSH version declared

28,644 artifacts reached

92%

Install Tested

Installed in an isolated sandbox, no dep errors

26,429 artifacts reached

62%

Run Tested

Plugin loads and runs (smoke test passed)

17,817 artifacts reached

What do the security scans say?

The full answer to "is it safe" — risk distribution across the registry

Security audits

The first layer is a registry-wide static scan: automated static analysis (dependencies, permissions, secrets, supply chain) for every indexed entry — not a manual review, with verdicts re-computed registry-wide as rules evolve. The second layer is a model-driven dynamic security audit, applied only to entries the static scan already flags as needing review; it reads source context and covers that subset, not the whole registry. Merging works the other way round: when the dynamic audit has a conclusion, that conclusion is authoritative — it can raise the tier or lower it; only when there is no dynamic conclusion does the static scan tier (vet primary, dsh backstop) apply. A dynamic conclusion really moves the tier here: the tier follows the audit's own finding counts — any critical finding reads critical, otherwise any high finding reads high, otherwise any medium (warning-level) finding reads medium, otherwise low — with badges, list filters and registry-wide counts following, and a conclusion softer than the static scan lowering the site tier accordingly. The audit script never writes the suppression list — exemptions stay manual — and a conclusion a human review withdraws stops applying, dropping the tier back to the static scan level. 2,919 dynamic audit records exist so far (including runs that produced no conclusion), of which 2,341 carry a model conclusion and 2,340 set the site tier from the dynamic conclusion (a conclusion a human review withdraws stops moving it). Entries outside the scanning scope (for example not yet through L4 install testing, or with the discovery topic removed) follow the existing policy: they do not participate in the registry-wide risk statistics and get no published risk-badge value. The plugin detail page and the per-plugin JSON are the authoritative tier.

What is it?

Type, categories, use cases extracted from the repo — no guessing.

Is it safe?

Automated static scan plus a model-driven dynamic security audit — a dynamic conclusion is authoritative when there is one (it can raise or lower the tier, and the tier follows the finding counts of that audit: critical → Critical, high → High, medium → Medium, otherwise Low), the static scan tier (vet primary, dsh backstop) applies otherwise, and exemptions stay manual; out-of-scope entries stay outside the risk statistics and their badge reads unknown.

Can I install it?

Verification L1-L5 — from "found" to "installed and tested in an isolated sandbox".

Is it maintained?

Health: active / stable / inactive / archived / broken.

Discover useful dsh plugins

Registry trust in numbers

Add dsh.so as a DSH plugin index source

Editor picks: start with these

@deepseek-ai/deepseek-harness

@anywhere-labs/dsh-desktop

@liustack/modlens

@omdsh-dev/DSH-better-sidebar

@xiazhi88/dshgo

@Ebony-Vinyl/dsh-our-free-model

@Alterssss/dsh-bili-whale

@MeteorNOX/DeepSeek-Balance-Whale-Widget

What can developers do with dsh plugins?

Files & data

Memory & knowledge

Models & agents

Dev workflow

Connect & integrate

Look & feel

Office & docs

Runtime & terminal

Creation & media

Web & search

Interface add-ons

Security & audit

Finance & trading

Fun & life

Desktop & clients

Workbench

Ecosystem plugins

What was recently updated?

@KongFangXun/sofagent

@qishuilalala/dsh-voice-mode

@vshulcz/deja-vu

@better-er/dsh-peak-block

@better-er/dsh-write-rule-guard

@zhu1090093659/dsh-web

@harness-home/harness-ai-plugins

@zbc0315/dsh-synomega

@coeasy/dsh-go

What is newly listed?

@maverikhe-cpu/zhuangzai-dsh-workbench

@Tinxyoo/dsh-file-mention

@mznxbcvsdc/dsh-shot-send

@yuxiangwhn/dsh-chat-wallpaper

@retweve/dsh-plugin-pinky

@tyrantblue/dsh-trek-mcp

@HimLeoing/dsh-task-progress

@yan92626/dsh-task-butler

@bob868335-rgb/dsh-lightvela

Not sure what to install? Start with what you want to do

How does dsh.so verify plugins?

What do the security scans say?

AI News

Connect a Verified Plugin Market to DSH: Index Source Guide

DSH Quick Start: From Install to Your First Verified Plugin

FAQ

Built a DSH plugin?

Find dsh plugins, clients, and integrations in one trusted directory