DeepSeek Harness Plugin Marketplace

Explore the DeepSeek Harness plugin ecosystem powered by the Cordis microkernel. Discover official and community extensions for agent orchestration, custom tools, isolated execution sandboxes, vector memory, and private LLM adapters. Built with declarative YAML configuration, live HMR, and end-to-end Trajectory tracing for enterprise-ready AI agents.

110+

Total Plugins

Official Core

Community Verified

2.5M+

Monthly Downloads

DeepSeek Harness Core Infrastructure

@deepseek-ai/deepseek-harness

Production-grade AI agent execution kernel for DeepSeek Harness, decoupling models, sandboxes, and tools via Cordis.

#Core#Agent#Cordis

PTC Code-Mode Runner

@deepseek-ai/dsh-plugin-ptc-runner

Programmatic Tool Calling engine exposing tools as typed TypeScript APIs for complex control flow scripting by LLMs.

#PTC#TypeScript#CodeMode

Subagent Dynamic Orchestrator

@deepseek-ai/dsh-plugin-subagent

Dynamic subagent orchestration center enabling isolated context spawning, concurrent execution, and inter-agent bus messaging.

#MultiAgent#Subagents#Swarm

Minimal Dual-Tool Benchmark Preset

@deepseek-ai/dsh-plugin-minimal-runner

Minimal dual-tool evaluation preset featuring Bash and Str-Replace, designed for pure agent benchmarking without noise.

#Minimal#Benchmark#SWE-bench

Creative Mode In-Memory Sandbox

@deepseek-ai/dsh-plugin-creative-repl

In-memory interactive plugin playground for hot-reloading, debugging Cordis extensions, and inspecting live event buses.

#Creative#HotReload#Developer

Goal-Oriented Task Planner

@deepseek-ai/dsh-plugin-task-planner

Goal-oriented task planner breaking complex full-stack requirements into structured milestones, checklists, and verifications.

#Planner#Milestones#TaskPlanning

Docker Container Sandbox

@deepseek-ai/dsh-plugin-docker-sandbox

Physical Docker container sandbox enforcing strict resource quotas and isolation for shell commands and compilation.

#Docker#Sandbox#Security

Wasm Microsecond Lightweight Sandbox

@deepseek-ai/dsh-plugin-wasm-isolator

WebAssembly-powered millisecond sandbox delivering zero cold-start overhead and strict memory boundaries for user scripts.

#WASM#Sandbox#Lightweight

Trajectory Append-Only Logger

@deepseek-ai/dsh-plugin-trajectory-logger

Append-only Trajectory audit log and event streaming engine supporting full reasoning chain recording, forks, and replays.

#Trajectory#EventStream#Replay

Persistent Bash Terminal Engine

@deepseek-ai/dsh-plugin-bash

Stateful persistent Shell terminal executor with session retention, ANSI cleansing, and background daemon management.

#Bash#Shell#Terminal

Multi-chunk Precise Code Editor

@deepseek-ai/dsh-plugin-str-replace

Context-aware Multi-chunk non-contiguous code patch editor preventing token bloat and hallucination from full-file overwrites.

#Editor#Diff#Refactor

Ripgrep High-Speed Code Search

@deepseek-ai/dsh-plugin-ripgrep

Embedded native Ripgrep regex search engine providing millisecond codebase scanning, symbol matching, and path filtering.

#Ripgrep#Search#Regex

DeepSeek Native Provider & KV Cache

@deepseek-ai/dsh-provider-deepseek

Dedicated LLM provider adapter for DeepSeek-V3, R1, and V4, optimized for KV prefix caching and streaming tokens.

#DeepSeek#PrefixCaching#LLM

Visual Web UI & Trajectory Inspector

@deepseek-ai/dsh-ui-web

Out-of-the-box modern Web UI dashboard offering collapsible thinking trees, visual diffs, and event timeline tracking.

#WebUI#Dashboard#Visualizer

Model Context Protocol (MCP) Bridge

@deepseek-ai/dsh-plugin-mcp

Model Context Protocol (MCP) bridge seamlessly connecting hundreds of community MCP servers, tools, and data sources.

#MCP#Standard#Integration

AI Design & Prototype Engine for DSH

github:nexu-io/open-design

Empowers agents with 'AI Design & Prototype Engine for DSH' capabilities for automated Design and Prototype operations, static analysis, and code processing.

#Design#Prototype#Figma

AI-Powered Privacy-First Resume Builder

github:amruthpillai/reactive-resume

Empowers agents with 'AI-Powered Privacy-First Resume Builder' capabilities for automated Resume and Career operations, static analysis, and code processing.

#Resume#Career#Generator

Expert Experience Distillation & Persona Skill

github:titanwings/colleague-skill

Enhances agent lifecycle orchestration and scheduling for Knowledge and SkillGenerator, ensuring deterministic task execution.

#Knowledge#SkillGenerator#Persona

Multi-Model Enhancement & Prompt Hub

github:Nagi-ovo/voyager

Extends communication protocols for Extension and PromptManager, supporting multi-platform notifications, remote controls, and UI review.

#Extension#PromptManager#MultiModel

Verifiable Architecture & Sequence Diagrams as Code

github:tt-a1i/archify

Empowers agents with 'Verifiable Architecture & Sequence Diagrams as Code' capabilities for automated Architecture and Diagram operations, static analysis, and code processing.

#Architecture#Diagram#SVG

Industrial Image Generation Prompt Engine

github:freestylefly/awesome-gpt-image-2

Empowers agents with 'Industrial Image Generation Prompt Engine' capabilities for automated ImageGen and PromptTemplate operations, static analysis, and code processing.

#ImageGen#PromptTemplate#Design

AI-Native Multi-Device Workspace & Dashboard

github:YaoApp/yao

Extends communication protocols for Golang and Workspace, supporting multi-platform notifications, remote controls, and UI review.

#Golang#Workspace#Distributed

Modern Cross-Platform Desktop Client for DSH

github:anywhere-labs/deepseek-harness-desktop

Extends communication protocols for Desktop and Electron, supporting multi-platform notifications, remote controls, and UI review.

#Desktop#Electron#CrossPlatform

Self-Evolving Agent OS with Stage Evaluation

github:Q00/ouroboros

Enhances agent lifecycle orchestration and scheduling for AgentOS and SelfEvolving, ensuring deterministic task execution.

#AgentOS#SelfEvolving#Evaluation

Showing 24 of 110 plugins

Cordis Plugins & Ecosystem FAQ

Frequently asked questions about DeepSeek Harness plugin architecture, installation, sandbox safety, and custom development.

What is the DeepSeek Harness plugin system, and how does it differ from traditional agent tools?

Built on the Cordis microkernel, Harness plugins decouple tools, sandboxes, and state machines, enabling models to invoke tools via the bus or write programmatic TypeScript scripts (PTC) to batch-orchestrate workflows.

How do I install and activate a specific Cordis plugin in my project?

You can install plugins with 'npx @deepseek-ai/dsh add [package-name]' or declare the package and configuration directly in 'cordis.yml'. Harness supports live Hot Module Reloading (HMR), applying new plugins without restarting the host agent daemon.

How is runtime safety guaranteed when executing agent plugins?

Harness enforces multi-layered sandbox isolation. High-risk shell commands and builds run inside disposable Docker containers or WebAssembly environments with strict CPU, RAM, and read-only filesystem limits. AST static code scanning and security interceptors are also provided.

How can I integrate existing community MCP tools into Harness?

Harness includes a built-in MCP bridge plugin. Simply configure the MCP server launch command or SSE endpoint in 'cordis.yml'. Harness automatically performs JSON-RPC handshakes, discovers tools, and mounts them as native Cordis services.

How can developers build and publish custom Cordis plugins?

Run 'npx create-cordis-plugin my-plugin' to scaffold a TypeScript project. Implement the standard 'apply(ctx, config)' lifecycle function, publish to npm or GitHub, and leverage TypeScript declaration merging for full intellisense.

How does Harness handle plugin errors, exceptions, or timeouts?

Harness relies on a deterministic state machine and Fiber lifecycle controller. If a plugin throws an exception or times out, error interceptors trap the failure, execute self-healing rollbacks, and inject sanitized error logs into the next reasoning step without crashing the agent process.