Python Modern：按项目 Python 版本自动适配的现代编码规范
Python Modern 是 SkillHub 社区 Skill，坐标 @user_7b1db0fd/python-modern，v1.0.0，提供基于项目 Python 版本自动适配的现代 Python 编码规范，无需 API Key。文章说明其定位、可核实功能、通过 skillhub CLI 安装到 Cursor 等客户端 skills 目录的步骤，以及对话式开发中的典型用法与适用注意。官方页面：https://www.skillhub.cn/skills/user_7b1db0fd/python-modern
Read Moreevo-engineering：用固定回合协议让 agent 多轮改进任意产物
evo-engineering 是 DSH 工作流插件，让 agent 对文档、代码、skill 等产物按固定回合改进：标准写入 evals/criteria.md，子代理 propose→act→critique→select，冠军线选优并跑回归探针，每圈 git 打点可回滚。提供 /evo 命令与 evo_* 工具，依赖 workflow 引擎；随包含跨 harness 的 evo skill。安装：dsh plugin --profile web add "github:vlln/evo-engineering#main"。MIT，v0.1.0。
Read Moredsh-agent-governor: DeepSeek Harness 的两重拦截层
`dsh-agent-governor` 是 DeepSeek Harness (DSH) 的一个插件，旨在通过两层拦截层治理 Agent 行为。插件包含 **DOL 层（通信语义治理）** 和 **Sentinel 层（工具执行治理）**，分别管理 Agent 的输出承诺和工具调用安全。它拦截破坏性命令、凭据泄露、绝对化承诺等风险操作，并提供可视化的开关、统计及影子模式。插件通过 npm 安装，支持在配置文件中灵活覆盖规则，有效降低 Agent 的错误率和安全风险。
Read More文件自动整理助手：用 Agent 按类型归档指定文件夹
文件自动整理助手是 SkillHub 社区 Skill（@user_d0b67476/file-organizer-gjn，v1.0.0，office-efficiency），由 user_d0b67476 发布。它整理指定文件夹，按文档、图片、代码、压缩包、视频等类型归入子目录，并支持多种格式识别，无需 API Key。文章说明定位与核心能力、通过官方提示或 SkillHub CLI 安装到 Cursor/Claude Code 等 skills 目录的方法，以及指定路径后让 Agent 执行归类的典型协作步骤与备份注意。
Read Morenuwa-skill-qszf：把人名或主题蒸馏成可运行 Skill
nuwa-skill-qszf 是 SkillHub 社区 ai-agent 类 Skill（v1.0.0，@user_15292d5a/yjkj-nuwa-skill-qszf），简介为「女娲造人」：输入人名或主题，自动深度调研并蒸馏成符合 SKILL.md 规范的可运行技能包，元数据标注 requires_api_key=false。本文介绍其定位、已核实功能、通过 skillhub CLI 或官方安装提示词安装到 Cursor/Claude Code/Codex 等 skills 目录的方式，以及基于官方描述的典型使用流程与审阅注意点。
Read Morectf misc：跨类 CTF 杂项题的 Agent Skill
ctf misc（@user_257ad2a8/ctf-misc，v1.0.0）是 SkillHub 社区 Skill，面向难以归入 web/pwn/crypto 等主类的 CTF 杂项题，覆盖编码、pyjail、RF/DNS、unicode、QR/音频、约束求解等方向，并强调主类题应优先专用 Skill。本文说明其定位、能力范围，以及通过官方提示词或 skillhub CLI（含 Cursor 的 ~/.cursor/skills/）的安装方式与使用注意。requires_api_key=false。
Read More《Cursor Docs》-Projects
A Project takes on a larger body of work, such as a feature, a migration, or a full app. You direct it by chatting with its coordinator agent. The coordinator doesn't write code itself. It plans the work, delegates it to agents that write the code, and brings the finished work back to you to check. Projects live in the left hand nav of the Agents Window and run on Cloud Agents.
Read More法眼-全能法律专家顾问：在 Agent 中接入多领域法律辅助
法眼（@user_6c0dcd6a/law-expert）是 SkillHub 社区发布的专业类法律 Skill，面向个人与中小企业，覆盖合同、劳动、税务、知识产权、数字合规等 12 个领域，提供合同审查、纠纷诊断、文书起草、合规体检、案例检索等 13 类服务；无需 API Key。本文说明其定位与边界，并给出 skillhub CLI 与 Cursor 等客户端的安装路径与典型对话用法。
Read MoreDeepSeek Harness 插件：zseven-w/dsh-qa
zseven-w/dsh-qa 是 DeepSeek Harness (DSH) 的一个 QA 协调器插件，旨在通过 Agent 驱动的探索模式，让智能体以真实用户方式操作应用并捕获证据，最终将验证过的操作转化为可复现的回放场景。该插件支持 Web、桌面及移动端多平台测试，提供从观察到回放的完整工作流，允许将场景导出为声明式 JSON 文件。开发者需自行安装对应驱动包并配置入口，安装后可运行示例脚本验证。使用时需注意证据导向、无权限升级限制以及移动端和 macOS 计算机辅助的特殊要求。
Read MoreDeepSeek Harness 插件：dsh-actions
DSH Actions 插件旨在解决开发中反复操作割裂的问题。它允许开发者通过一份 `actions.json` 配置，既在 Web 右侧栏一键运行，也能让 Agent 直接调用。插件支持三层任务定义、实时流式输出及审批闸门机制，将脚本、CI/流水线操作统一固化，并提供了基于入口的细粒度权限控制，降低了 Agent 调试成本与操作门槛。
Read MoreDeepSeek Harness 模型分发插件
该插件为 DeepSeek Harness (DSH) 增强了模型管理能力，允许用户在统一界面中为主会话、子代理及计划模式分别配置独立的模型和推理强度。核心功能包括输入栏右侧的芯片控制、统一的模型搜索菜单、实时生效的推理滑块，以及计划模式下的自动切换与会话覆盖机制。该插件需配合 DSH 0.2.0-rc.2 使用，安装后需完全重启应用。
Read MoreDeepSeek Harness SearchOps 插件：Agent 原生搜索与日志调查
`dsh-searchops` 是专为 DeepSeek Harness 设计的 Agent 原生 OpenSearch 插件，旨在解决智能体直接查询原始端点导致的数据过载与 Token 消耗问题。它提供 9 个只读工具，负责处理日志去重、指纹识别和边界限制等确定性工作，并将结构化证据交给智能体进行推理。插件支持多源配置、提供者抽象及严格的安全边界（如只读模式、凭证隔离）。该工具严格基于 OpenSearch 构建，目前为 v0.1.0 预发布版本，适用于日志调查与集群健康检查。
Read MoreDispatch Gate
The Dispatch Gate is a DSH plugin designed to address the issue of AI directly executing tasks in multi-model collaboration. By intercepting tool calls such as `pwsh` and `write`, it enforces that a session must have a dispatch record before execution. The plugin supports evidence-based approval (v10) and behavioral ledger logging to ensure “delegate first, execute later.” Installation and configuration are implemented through `plugin_manager` and `cordis.patch.yml`, offering multiple self-service release methods, a rollback toggle, and audit log functionality.
Read MoreDeepSeek Harness Plugin: dsh-creator-mode-plus
`dsh-creator-mode-plus` is a fail-closed Creator Mode+ bridge plugin purpose-built for DeepSeek Harness. It enables users to directly develop, activate, and validate plugins through an Agent in web sessions, providing full lifecycle management from scaffolding to hot reloading. The plugin employs an externally supervised security design that disallows modifications to the host environment, aiming to simplify the development experience through automated workflows.
Read MoreDeepSeek Harness Plugin: dsh-sidebar-enhance
The `dsh-sidebar-enhance` plugin for DeepSeek Harness (DSH) uses the official API to enable AI access to the sidebar. Core features include: after an Agent completes a task, it calls a tool to automatically display files, a terminal, or web pages; the AI can also use tools to navigate, read, and interact with pages in the sidebar browser. After installation, the desktop client must be fully restarted for the changes to take effect. Browser navigation requires no authorization, while reading and interacting with pages requires user authorization. By connecting sidebar functionality, the plugin optimizes the Agent’s workflow loop.
Read Moredsh-stepwise-distill: Solidifying DSH Session History
`dsh-stepwise-distill` is a DSH plugin for addressing history bloat in long-conversation scenarios. It stepwise consolidates session history, rewriting tool results as the facts actually retained by the model, preventing repeated resending of completed content, and improving context density. The installation command is `dsh plugin --profile <name> add dsh-stepwise-distill`. Summary features (`stepSummary` and `turnSummary`) are enabled by default, and their status can be checked via the `/distill` endpoint. Note that this plugin roughly doubles token and time costs and requires a Node.js >= 20 environment.
Read Moredsh-decision-layer: The Structured Decision Layer of DeepSeek Harness
dsh-decision-layer is an open-source plugin that provides a structured decision layer for the DeepSeek Harness Agent loop. It addresses unreliable judgment from large models by leveraging arbitration backends (such as dev/laya) and integrating automatic mechanisms such as risk gating, output self-inspection, tool scope reduction, and loop prevention. It supports manual arbitration, backend configuration, and local deployment, ensuring a safety boundary that only restricts allowed behavior without expanding it, and provides comprehensive log persistence and analysis to help developers precisely control Agent behavior.
Read Moredsh-simple-codegraph: Production-Grade Single-Agent CodeGraph Integration for DeepSeek Harness
`dsh-simple-codegraph` is a plugin designed specifically for DeepSeek Harness (DSH), aiming to address the costly context consumption and inefficiency caused by frequent use of `grep`, `glob`, and `read_file` when agents refactor codebases. The plugin provides production-grade, agent-isolated CodeGraph integration, supports concurrency across multiple projects, and simplifies the interface by exposing a single `codegraph_explore` tool. Once installed, an agent can retrieve source code, call hierarchies, and impact scopes through a single centralized exploration call, significantly improving code comprehension efficiency.
Read MoreDSH Remote Control: Enabling Agents to Control Local DSH Instances
The `dsh-remote-control` plugin allows an Agent to directly control another independent DSH instance in a zero-intrusion manner, addressing the limitations of collaboration within a single instance. Its core capabilities include direct control of remote sessions, incremental reads, search, and reference management, and it provides six persistent tools, such as session creation and message sending. After installation, connection information must be configured in the conversation, enabling the Agent to use the plugin for testing, collaboration, and decomposing complex tasks.
Read MoreDynamic Thinking Chain: A DSH Plugin That Automatically Adjusts Reasoning Depth Based on Task Difficulty
`dynamic-thinking-chain` is a DSH plugin that lets DeepSeek models automatically adjust their reasoning level based on task difficulty. It monitors task texts and instructions in `todo_write`, calculates a difficulty score, and dynamically maps it to one of four levels: `off/low/high/max`. `max` is used for complex planning, while `low` is used for simple tasks, saving tokens without compromising quality. The plugin has zero dependencies and supports custom threshold configuration, but be aware of the limitation of disabling DeepSeek’s thinking mode mid-session.
Read Moredsh-peak-hours: DeepSeek Harness Execution Window Toggle Plugin
The DeepSeek Harness plugin `dsh-peak-hours` helps users reduce model invocation costs by automatically running or pausing tasks during off-peak hours to optimize resource usage. It supports two modes, "Always Run" and "Run During Off-Peak Hours Only," and determines peak and off-peak windows for Monday through Friday based on the Beijing time zone (UTC+8), with support for statutory holidays. Users can switch modes, query status, and simulate the clock through UI controls or the provided HTTP API. Implemented purely in JavaScript, the plugin should be installed carefully to avoid mixing installation methods. It controls task execution via an interception mechanism, making it suitable for scenarios that require running tasks only within specific time windows.
Read MorePreset Node Graph: Visualizing Agent Presets in ComfyUI Style
`agent-preset-graph-xww` is a visualization plugin for DeepSeek Harness that turns abstract agent preset manifests into ComfyUI-style node graphs. Maintained by laojingwei, it runs in the web frontend and does not modify host state. It displays an agent’s composition, real-time execution status, and various capability blocks, and supports features such as node search, detail viewing, automatic execution following, and image export. The plugin is compatible with DSH ≥ 0.1.7-rc.2, requires specific services, and helps developers monitor agent behavior and organize preset configurations.
Read Moredsh-memory-lite: Giving Your dsh Agent Cross-Session Long-Term Memory
The DSH plugin `alanzhao0128/dsh-memory-lite` provides dsh agents with zero-dependency, Markdown-based long-term memory across sessions. It persists preferences and decisions using a resident memory directory, five explicit memory tools (read, write, update, and forget), and a background implicit extraction mechanism. The plugin supports peer isolation and cross-peer sharing, with permission-based security controls and runtime status indicators. After installation, restart to use it; it is suitable for development scenarios requiring context retention and multi-session collaboration.
Read Moredsh-simulated-life: Providing Realistic Life Context for DeepSeek Harness Agent
dsh-simulated-life is a plugin for DeepSeek Harness (DSH) that provides Agents with a realistic life context. Each turn, it automatically reads and injects “life events” from the workspace within the past 24 hours, along with the current state of the world, while enabling Agents to write their thoughts and actions back to the log via the `life_react` tool. With automatic deduplication, zero intrusion, and time-window control, and in combination with external Skills, it supports continuous world evolution and long-term role-playing for Agents.
Read MoreDeepSeek Harness Skill Center Plugin
The DeepSeek Harness plugin `dsh-skill-center` adds a unified entry point in the sidebar to address challenges such as difficult skill discovery, fragmented management, and installation blind spots. Core features include search and installation with built-in health checks for naming validity, file integrity, and references; commit-level installation to avoid version confusion; reading local state and parsing zstd logs to track invocation counts; scanning and importing local skills; and safe deletion with backup. The plugin supports hot mounting, significantly lowering the barrier to using Agent skills.
Read Moredsh-j-space: J-Space Cognition Suite SV1 for DeepSeek Harness
dsh-j-space is the J-Space Cognition Suite SV1 integration plugin for DeepSeek Harness (DSH). It provides a plug-and-play cognitive framework featuring full-lifecycle Scope isolation, externalized state (persistent controllers and ledgers), and 13 cognitive engineering modules, including security and multi-agent orchestration. The plugin supports multiple base models, requires Bundle installation, and depends on Subagent model selection settings to deliver durable, controllable, and modular engineering and reasoning capabilities to DSH.
Read MoreDeepSeek Harness Thinking Loop Guard Plugin: dsh-loop-guard
`mrweicodes/dsh-loop-guard` is a plugin that resolves the “thinking-only, no-action” infinite loop in DeepSeek Harness agents. Through a five-layer detection mechanism, it accurately identifies pure reasoning loops or repeated spamming, directly interrupts calls within the stream, and injects corrective prompts. The plugin does not interrupt the session, supports offline analysis, has a very low false positive rate, and requires DSH version 0.1.2 or higher. After installation, automatic resumption can be configured for unattended operation.
Read Moredsh-todo-board: Cross-Session Task Management
dsh-todo-board is a cross-session task management plugin for DeepSeek Harness. It appears as a floating window in the top-right corner and supports grouping by working directory, drag-and-drop sorting, and switching between four UI styles. Each todo item can independently configure an execution mode (reminder, automatic continuation, automatic new session) and a deadline. Task state is derived from live sessions and includes a dual check-off mechanism for AI completion and user acceptance. The plugin supports task decomposition for long tasks, pasting screenshots, scheduled execution, and self-healing from failures. Users plan tasks using the `/todo` command. After the AI completes a task, it automatically marks it as done and identifies the next task; after user acceptance, the task is moved to the completed list. After installation, restarting the Profile is required, and data is stored locally.
Read Moreliudejua27-blip/fitmeet-dsh-plugin: Agent Social Networking
liudejua27-blip/fitmeet-dsh-plugin is a DeepSeek Harness plugin based on MCP 2.3, designed to address the pain point of agents connecting with strangers. It provides 23 tools that help an Agent discover resources, make plans, establish exchanges, and maintain connections, acting as a social connector. Users can issue natural-language commands to search the network, such as looking for a ball partner or a carpooling companion. Installation requires Node.js 22.19+ and OAuth 2.1 + PKCE authentication. The plugin transforms the Agent into a social agent, suitable for users who want to expand their social circles.
Read Moredsh-context-enhancement: Context Enhancement Plugin for DeepSeek Harness
`dsh-context-enhancement` is a context enhancement plugin for DeepSeek Harness, designed to address limited context windows and information dilution in long coding sessions. With core features such as key-point extraction for tools, forget-zone compression, independent session summarization, and main-thread segment compression, the plugin sets `contextual` as the default mode while preserving existing modes, enabling coherent management of context information and persistence of task state. Users must install Node.js 22.19.0 or later and DSH 0.1.2-rc.1, then select this mode through the Web interface to enable automatic context management and view status and summaries on the “Context Optimization” page. The plugin is currently in testing and is compatible only with the specified versions.
Read Moredsh-heartbeat: Autonomous Heartbeat Plugin for DeepSeek Harness
`dsh-heartbeat` is a DeepSeek Harness (DSH) plugin that gives an Agent an autonomous "heartbeat" and a sense of presence. Every 20 minutes, it runs a seven-phase cycle that includes maintenance, collection, wandering, gating, rumination, and delivery. A discretion gate checks conditions such as quiet hours and user status to decide whether the Agent should speak, while a material pool and `web_search` support perception and accumulation. The plugin supports profile maintenance, ledger accounting, and local encrypted storage, and is Windows-only.
Read Moredsh-evidence-gate: DeepSeek Harness Evidence Gate Plugin
`dsh-evidence-gate` is an audit plugin for DeepSeek Harness designed to enforce that agents reach conclusions through a structured evidence hierarchy and cross-checks, eliminating hallucinations based on unverified guesses. The plugin provides a three-layer evidence policy, a persistent session-level ledger, and a false-positive prevention ladder, with toggles available via command line and UI. After installation, it must be enabled in the configuration; it can generate detailed session reports and is suitable for scenarios that require high reliability and strict auditing.
Read Moredsh-acp-adapter: Using ACP Agent in DeepSeek Harness
The `dsh-acp-adapter` plugin enables DeepSeek Harness (DSH) to use ACP Agents directly in the UI. Users need to install the plugin and log in to the Agent CLI, then connect, verify, and switch to the Agent model in DSH to start a session. The plugin supports native session management, automatic tool approval, approval card display, and diagnostic exports. When using it, manage plugin updates manually and ensure that no sensitive information is leaked.
Read Moredsh-tauri-turnrewind: Automatic Snapshots and One-Click Rollback for Agent Changes
dsh-tauri-turnrewind is the local rollback plugin for DeepSeek Harness (DSH), designed to mitigate the risk of accidental operations caused by agents modifying files. It automatically creates private Git snapshots for each change and provides a two-stage undo workflow (reviewing a red/green diff before confirming execution), effectively reducing the risk of code contamination. The plugin features core capabilities such as automatic private snapshots, conflict detection, self-healing diagnostics, and an asynchronous model. When in use, users can specify a Turn ID with the `/undo` command to roll back changes, with support for skipping conflicts or force overwriting. Currently, this feature is experimental, supports only Git directory workspaces, and does not modify project history.
Read Moredsh-language-control: Preventing Language Drift in Agent Chain-of-Thought
The dsh-language-control plugin for the DSH workflow addresses the “language drift” issue in an Agent’s chain of thought caused by tools returning English content, by injecting language constraints into the System Prompt before each reasoning step. This plugin enforces Chinese throughout the Agent’s internal reasoning, planning, and tool calling, ensuring consistency between thought and output. It takes effect automatically after installation without requiring additional configuration, and users can also customize or restore the default constraints via the settings panel.
Read Moredsh-rich-context: Agent Instruction Manager for DeepSeek Harness
The `dsh-rich-context` plugin provides a visual Agent instruction management interface for DeepSeek Harness (DSH), addressing the issue of fragmented instruction files across multiple tools. Located in the Skill Center sidebar, the plugin allows users to edit instruction files under the **Global** and **Workspace** tabs and supports custom path management. Its core features include cross-tool directory scanning (supporting 10 tools) to locate instruction files and “Set as default,” which creates symbolic links for scanned files to simplify routing logic. Developed with a front-end/back-end separated architecture, the plugin requires restarting the `dsh web` process after installation to take effect.
Read MoreZ.AI / Zhipu Standalone Web Search API Plugin for DeepSeek Harness (DSH)
A Z.AI / Zhipu Web Search API plugin for DeepSeek Harness (DSH), designed to provide web access capabilities for the DSH ecosystem. Maintained by Riianfar, this plugin allows developers to use Z.AI or Zhipu search services. The current feature list and usage examples are available by referring to the source code implementation. Before installation, users should review the source code and license. This plugin is suitable for DSH users who need to integrate the above search APIs to enhance the information retrieval capabilities of their agents.
Read MoreKun Like Desktop Pet: DeepSeek Harness Desktop Pet Plugin
`dsh-kun-pet` is a fun dress-up plugin maintained by Practice019, designed to enhance the development experience of DeepSeek Harness (DSH). It introduces a desktop pet in the bottom-right corner of the web interface, mapping 9 Agent running states (such as idle, working, and thinking) to specific animated actions and providing voice feedback. It supports dragging to move and clicking to interact, and plays a voice message when a task is completed. Installation requires a Windows system, PowerShell, and the PresentationCore assembly; after installation, restart for it to take effect.
Read Moredsh-pure-chat: A Plugin That Makes DeepSeek Harness Agent Chat in Pure Text Mode Only
`dsh-pure-chat` is a DeepSeek Harness plugin designed to implement a pure text chat mode. By hiding global tools, intercepting execution, and modifying the system prompt, it ensures that the Agent only engages in text-based interactions and does not trigger filesystem or network requests. Users can dynamically enable or disable this mode for each session via the `/pure-chat` command. The plugin is maintained by community developers, is licensed under the MIT license, and runs on the `web` profile in the 0.1.x series.
Read More@kasenri/dsh-orbit: A Deterministic Long-Running Task Orchestrator for DeepSeek Harness
Orbit is a community plugin for DeepSeek Harness (DSH) designed to address the lack of determinism and traceability in long-running task execution. By decomposing tasks into roles such as Commander (planning), Executor (execution), and Heartbeat Watchdog (monitoring), and incorporating persistent state, loop limits, and session reuse mechanisms, it enables controlled workflow orchestration. Users can activate Orbit through commands or specific modes and use the `orbit_controller` tool to manage starting, resuming, and stopping. The plugin supports optional multi-candidate competing execution (MoA) and web operations, requires Node.js >= 22.19.0, and is suitable for complex engineering tasks that demand long-running, recoverable, and auditable workflows.
Read Moredsh-arxiv: arXiv Search and Abstract Retrieval Plugin
`dsh-arxiv` is a lightweight function plugin designed for DeepSeek Harness (DSH) to address the need for agents to access external academic information. The plugin provides "read-only" query functionality through the public arXiv API, supporting paper search and detail retrieval without requiring API keys or local storage. It is easy to install and supports Headless mode. The plugin strictly follows read-only and stateless principles, returning only metadata and abstracts, making it suitable for quickly retrieving the latest papers. Developers should ensure compliance with arXiv API usage guidelines and data ownership.
Read Moredsh-tool-failure-circuit-breaker: DeepSeek Harness Tool Failure Circuit Breaker
`dsh-tool-failure-circuit-breaker` is a security-related plugin for DeepSeek Harness that intercepts and blocks an Agent's repeated failed tool calls with identical parameters, helping prevent wasted resources and costs. By listening to the `tools/pre-execute` event, it directly rejects subsequent requests with the same parameters after the configured consecutive-failure threshold (`maxFailures`) is reached, and records the rejection as a normal result to preserve consistency in the interaction history. It supports flexible configuration of the protection scope through `include` and `exclude`, making it suitable for Agent scenarios that frequently call external APIs and involve retry logic.
Read Moredsh-kit: A Practical Tool to Fill the Gap in the DeepSeek Harness Plugin Ecosystem
DeepSeek Harness (DSH) provides a plugin package called `dsh-kit`, designed to address the “quiet failure” problem in agents, including 8 plugins. The package is written in pure ESM and requires Node.js >= 22. The plugin features include automatic code formatting, protection of generated lock files and build artifacts from unintended modification, detection of undocumented environment variables, code validation, workspace rollback, credential sanitization, Git context provision, and desktop notifications. All plugins support flexible configuration and disabling, helping to improve the stability and security of agent operations.
Read Moredsh-screen-reader: Giving Agents Eyes
The `dsh-screen-reader` plugin is designed to address Agent screen-perception automation and supports the Windows platform. It provides two core tools: `see_screen`, which directly captures the screen and sends the image to the model, and `see_diff`, which locates changed regions using local pixel differencing. The plugin enforces image normalization, decouples visual tasks, and allows the model to focus on "what happened" rather than "whether anything changed." Installation must be done via a profile bundle. This plugin is suitable for Agents to self-verify their operations, but it has limitations, including low measurement precision, no handling of temporal causality, Windows-only support, and an attachment library that can grow without bound.
Read Moreobsidian-dsh-acp: Bridging DeepSeek Harness to Obsidian via the ACP Protocol
`obsidian-dsh-acp` is an ACP server plugin that connects DeepSeek Harness (DSH) with Obsidian. It enables users to operate DSH within Obsidian’s native interface, with support for session management, model switching, tool-call frame display, and importing external sessions. The plugin depends on Node.js and a DSH environment. It can be installed by adding it to Obsidian’s configuration or DSH’s configuration file, and it supports session import and management via the command line or the built-in Web panel.
Read MoreDeepSeek Harness Plugin dsh-debate: Enabling Agents to Engage in Adversarial Debate
`dsh-debate` is a workflow plugin for DeepSeek Harness, designed to address the issue of agents providing single-perspective responses that lack adversarial challenge. It introduces independent pro and con subagents to enforce adversarial argumentation, with the main agent making the final decision. The plugin provides three core tools: startup, round execution, and adjudication, and supports codebase retrieval and web search. Its design includes a read-only mechanism, graceful degradation, and card-based Web UI display, requires a Node.js environment, and offers a zero-footprint installation.
Read Moredsh-opencli: Connecting Real Browsers and Site Adapters to DeepSeek Harness
The `dsh-opencli` plugin for DSH Harness provides “task execution” capabilities via OpenCLI. It drives actual browser windows through a browser extension and supports 176 sites with 200+ commands. Users can retrieve structured data and perform actions such as posting or approving, with all operations requiring approval by default. The plugin supports scheduled tasks, auditing, and concurrent multi-site execution. Installation requires deploying the OpenCLI environment first; ensure Windows paths are configured correctly, and note that headless servers are not supported.
Read Moredsh-web-ding: Implementing an Agent Turn-End Notification Sound in the Browser
`dsh-web-ding` is a DSH Web client plugin that synthesizes notification sounds using the Web Audio API to provide idle-state prompts when an Agent turn ends. The host side listens for idle transitions in `agent/status` and emits signals, while the browser side mirrors this namespace. It supports three modes: frontend-only question triggers, Agent turn-end triggers, and Windows 11-style Toast notifications. Messages are stored locally, and volume, frequency, and decay duration are configurable. When using the plugin, be aware of the browser’s autoplay policy and background-tab suspension mechanisms; the AudioContext must be warmed up on first use.
Read Moredsh-self-improvement
The DeepSeek Harness Agent plugin provides agents with a maintainable long-term memory pathway through tiered and weighted mechanisms. It categorizes memory into five types, supports retrieval based on importance and recency, and offers governance features such as deduplication, poisoning protection, and score-based eviction to manage capacity constraints. The plugin also supports automatic retrospection, skill accumulation, and cross-session experience reuse, ensuring that agents retain context after restarts or session switches. Users can manage memory using commands such as `/memory` and `/retro`, and can configure memory scope at either the workspace or global level. It is suitable for complex application scenarios that require long-running operation, automatic experience accumulation, and secure governance capabilities.
Read Moredsh-prime-memory: Hierarchical Distilled Memory Plugin for DeepSeek Harness
The `dsh-prime-memory` plugin for DeepSeek Harness is designed to address memory loss in long conversations by leveraging a hierarchical distillation mechanism (L0-L3) to automatically consolidate and inject historical information. It offers session tier switching, a cost dashboard, tool calls, and a “write-only, no-read” mode, all transparent to users. Installation requires Node ≥ 22.16, and data remains in a local directory after uninstallation, making it well suited for handling complex Agent tasks.
Read Moredsh-s2s — DeepSeek Harness Same-Host Session Interconnection Plugin
`dsh-s2s` is a lightweight plugin for DeepSeek Harness designed to address collaboration among multiple AI Agent sessions on the same host. It eliminates cross-host network communication and uses an in-process Broker for zero-port, zero-reconnection direct message delivery. The plugin supports in-process delivery, Name-based primary addressing, resuming dormant sessions, and loop-prevention rate limiting. It provides six core tools for session management, message sending, and history queries, and supports persistent message storage. During installation, configure `autoResume` and `budget` to enable automatic session resumption and rate limiting, making it suitable for low-complexity scenarios requiring only same-host collaboration.
Read Moredsh-preset-switch: In-Session Agent Preset Switching for DeepSeek Harness
The DeepSeek Harness (DSH) plugin `dsh-preset-switch` enables users to dynamically switch Agent Presets within the same session without restarting it. It uses extended commands and an HTTP interface to bypass the official blank-session lock restriction. The plugin provides `/preset` menu selection, keyboard shortcuts, and toggle controls, allowing developers to flexibly test different roles and toolchain configurations. Install it using `--profile web`, and note that switching may trigger tool-call compatibility warnings.
Read Moredsh-plugin-installer: A Plugin Store and Installation Assistant for DeepSeek Harness
`dsh-plugin-installer` is a plugin management tool for DeepSeek Harness (DSH), providing a visual plugin store and an automated installation assistant through a Web GUI. Users can browse plugins in a browser and install them with one click, while the Agent executes the actual commands and verifies the results. This plugin supports viewing plugin details, and installations are handled by the Agent in the current session, with the Host process having read-only permissions. Installation requires adding the plugin via a command and restarting the `dsh web` service.
Read MoreLume — DSH Desktop Enhancement Plugin: Task Execution Protocol and Persona System
DeepSeek Harness Desktop’s Lume plugin provides structured constraints and memory support for agentic development, using protocols to ensure that models remain “grounded in facts.” It consists of three layers: the Discipline Layer, responsible for request-type classification, fact-checking, and context warnings; the Method Layer, which ensures verifiable outputs through mechanisms such as task contracts, change ledgers, and assumption ledgers; and the Persona Layer, responsible for style control and memory management. Lume includes a dashboard and triggers to monitor runtime status, and supports defining contracts, querying metrics, and managing project knowledge. Rather than replacing model reasoning, it provides only mechanism-based safeguards. Open-sourced under the MIT license, the plugin can effectively address issues such as requirement drift and unauthorized operations.
Read Moredsh-kanban-flow: An Agent-Driven Kanban Board for DeepSeek Harness
`dsh-kanban-flow` is an agent-driven Kanban plugin for DeepSeek Harness, designed to enforce human–agent collaboration workflows through a structured Kanban interface. It maps each workspace to a board with fixed columns (from Backlog to Done) and strictly controls task flow—for example, only humans can add items to Backlog, while agents execute tasks up to Review. It supports local storage and a responsive UI. By providing agents with specific Kanban operation tools, it binds task state to agent sessions, making it suitable for scenarios requiring rigorous control over task delivery quality.
Read Moreagent-discord-presence: Sync Active Agent Workflows to Discord Rich Presence
`agent-discord-presence` is a local service tool designed to address the lack of clear status visibility for agent tools such as DeepSeek Harness on Discord. It synchronizes agent workflows in real time to Discord Rich Presence, displaying task summaries, model information, and session duration. The tool supports multiple platforms, including Codex and Claude Code, and features session management, custom configuration, and multi-agent support. Users need to install Node.js and run `npm install` to install it globally, then start the service using the `adp start` command. Note that the project is in an early stage of development, and the service must be started manually after restarting a computer.
Read Moredsh-swarm: Multi-Agent Task Pipeline Plugin
dsh-swarm is a plugin for DeepSeek Harness (DSH) designed to address role drift and deadlocks in multi-agent collaboration through a strict six-role coordination mechanism. The plugin converts requirements into a visual Kanban workflow with six dedicated roles—V, P, PT, W, D, and DT—responsible for orchestration, planning, review, knowledge synchronization, execution, and acceptance, respectively. The system operates in swarm mode, supporting automatic intent recognition, confirmation gating, and machine-verified evidence delivery. The main session is limited to read-only orchestration, while code writing and submission are performed by the executor (D) in an isolated environment. Users must confirm a checklist before starting the pipeline and can query progress in real time using natural language. This plugin requires Node.js ≥ 22.19 and a specific DSH runtime environment.
Read Moredsh-harvest: DSH Native Multi-Platform Research Pipeline Plugin
The DeepSeek Harness plugin `dsh-harvest` is a plugin-based tool for research and analysis scenarios, automating multi-platform research through a `scout → extract → verify → audit` pipeline. It supports 11 channels, including GitHub, the web, and social media, and features a fault-tolerant fallback mechanism. The plugin depends on multiple CLI tools and has specific compatibility requirements when running on Windows, macOS, and Linux, requiring configuration of the Python path, cookies, and API keys. Users can install it from the native DSH marketplace to perform broadcast discovery, in-depth research, and source auditing, aiming to reduce the complexity of agent research.
Read Moredsh-watcher: Read-Only Session Insights and Statistics Plugin for DeepSeek Harness
`dsh-watcher` is a read-only plugin purpose-built for DeepSeek Harness, designed to provide developers with session insights, work path observation, and cost estimation. It visualizes Agent execution steps, parallel branches, and tool results using collapsible views, supports session aggregation and cost statistics, and does not intrude into the conversational flow. It requires DSH 0.1.7-rc and Node.js 22/24+, and can be used by adding `github:aa2246740/dsh-watcher#v0.6.1` through the desktop app or Web CLI, helping developers clearly analyze workflows and monitor model invocation costs.
Read MoreDeepSeek Harness Plugin: Use Web-Based DeepSeek as a Tool for Programmable Agents
Developed by `wilianyichen`, the `dsh-deepseek-web` plugin aims to address the cumbersome task of directly handling data from the DeepSeek web interface. It reuses Python scripts to turn web sessions into a programmable, read-only interface. Core features include list viewing, full-text download, incremental synchronization, search indexing, and high-density summary generation (such as `deepseek_web_digest`). It supports daily activity analysis and topic clustering, and relies on a local token for secure access. The plugin is suitable for developers who need to process session history in bulk or create knowledge archives.
Read Moredsh-budget-split: Budget Allocation
The `dsh-budget-split` plugin for DeepSeek Harness (DSH) is designed to support household financial planning. By registering the `budget_split` tool, it generates budget plans based on input parameters. Its key benefit is a pure Node.js implementation with no network dependencies, making it easy for developers to integrate directly. Developers can install it using the `dsh plugin add` command and use it with the DSH toolkit, providing agents with a lightweight, standalone solution for basic financial planning.
Read MoreDSH Web Search Plugin
The DSH Web Search plugin is a host tool in DeepSeek Harness that replaces the official `tool-web`, designed to provide more controllable search and fetching logic. Installation requires Node.js 22+ and other prerequisites. After installation, the official tool of the same name must be disabled to avoid conflicts. The plugin supports Bing and SearXNG engines, exposes `web_search` and `web_fetch` interfaces, and enforces output size limits to prevent memory overflow, making it suitable for direct invocation in production environments.
Read MoreDeepSeek Harness Long-Term Memory Plugin: dsh-gitmemo
`dsh-gitmemo` is a plugin for DeepSeek Harness aimed at addressing the long-term memory storage problem for agents. It manages memory entries using a local Git repository, providing persistence and auditability while avoiding database dependencies. The plugin exposes five core tools (read, write, search, delete, and replace) through the Cordis mechanism and is registered only with the Root Agent. It supports configuring a search limit; the environment must have the Git CLI installed and the process restarted for changes to take effect, offering agents a lightweight, local long-term memory solution.
Read MoreDeepSeek Harness Plugin: dsh-prompt-customizer
`dsh-prompt-customizer` is a prompt and tool management plugin for DeepSeek Harness, designed to address prompt contamination and bloated tool catalogs. The plugin supports hierarchical management by session phase (onboarding, persistent, and compression-controlled phases) and agent presets, providing prompt blocking, replacement, injection, and reordering capabilities, as well as skill set management and MCP Server integration. It enables the `forceSections` mechanism by default to ensure fine-grained control over prompts, allowing developers to flexibly adjust configurations through drag-and-drop and real-time preview.
Read Moredsh-agent-preset-switcher: Hot-Swap DSH Agent Presets at Runtime
The DSH Agent Preset Switch Plugin enables dynamic switching of Agent modes during runtime without restarting the session. Through the /mode command, users can apply changes at step boundaries, leveraging the official AgentPresets.recompose mechanism under the hood. The plugin is suited for debugging and testing scenarios, supporting preset list viewing and working mode switching, though sub-agent sessions and underlying exceptions may restrict switching.
Read Moredsh-verify: Browser Acceptance Testing Gate for Agent-Built Applications
dsh-verify is a quality gating tool for web applications built by AI agents. Based on Playwright, it runs acceptance tests in real browsers to address the issue where code self-checks do not reflect actual behavior. Its core capabilities include real interaction testing, MCP server integration, GitHub Action CI support, visual regression detection, and multi-browser compatibility. It defines test logic using JSON specifications, ensuring reliable regression testing for AI agent applications through objective pass/fail results.
Read More@zhourenke/dsh-agent-rate-limit: Automatically Add TPM/RPM Rate Limits for DSH
The DSH Agent loop can easily exceed rate limits, triggering HTTP 429 errors. The `@zhourenke/dsh-agent-rate-limit` plugin inserts an adaptive delay layer before the LLM streaming pipeline, uses a sliding window algorithm to manage the request queue, and requests are queued before vendor rate limits are reached to prevent conversation interruptions. The plugin works with default configuration, supports TPM/RPM limits, safety factor adjustment, and hot configuration reloading, but shared window state is not supported and it does not guarantee that rate limit violations will never occur.
Read MoreIntroduction to the DeepSeek Harness Documentation Tool Plugin
`dsh-tool-doc` is a plugin for DeepSeek Harness (DSH) designed to let Agents process documents directly without invoking external commands via the Shell. It provides three core tools—read, create, and edit—and supports formats such as PDF, DOCX, XLSX, and CSV. Editing operations support text replacement, cell updates, and appending rows, with strict validation mechanisms. Key limitations include DOCX supporting only paragraph-level replacement, PDF/PPTX not supporting in-place editing, and binary format writes not adhering to the sandbox policy. By integrating document operations, the plugin significantly reduces the latency and complexity of document processing for Agents.
Read MoreDeepSeek Harness Plugin: Stone623/ai_skills
`ai_skills` is a plugin package for DeepSeek Harness, designed to address coherence issues during long-running Agent sessions. By invoking `$daydream-break` in task instructions, the plugin allows the Agent to “zone out” during work intervals and briefly reflect on its current state, thereby reducing monotony. Users can customize the number, duration, and style of these pauses. This mechanism retails context by leveraging session snapshots, without relying on background timers. It is suitable for enhancing natural interactions in work-paced scenarios, but not for automated workflows that require enforced interruptions.
Read MoreDeepSeek Harness System Information Tool `dsh-tool-sysinfo`
`dsh-tool-sysinfo` is a system information tool plugin for DeepSeek Harness, designed to help agents obtain snapshots of local environment configurations. Built exclusively with native Node.js modules, it has zero external dependencies, zero network requests, and zero write operations, and can read information such as the operating system, CPU, memory, hostname, and Node.js version. Users can install it via the official command, and after a restart, they can ask the agent in the chat interface to call the `system_info` tool to view the data. The plugin is implemented as a single file, and maintainers recommend using pnpm ≥ 11 to avoid a known bug on Windows.
Read MoreAgent Review Governance Plugin: Human Approval Windows and Signed Receipts
The DSH workflow plugin maintained by satan9394 is designed to provide audit and governance capabilities for Agents. Its core features include governing Agent execution, configuring human approval windows, and generating signed receipts, thereby enhancing process controllability and traceability. It is suitable for business scenarios that require strict auditing of Agent behavior or manual confirmation of critical operations. The plugin runs with the permissions of the current process, so review the source code and license before installation.
Read Moresatan9394/dsh-cli-anything: A CLI Wrapping Pipeline to Agentize Any Software
The DSH plugin dsh-cli-anything is designed to wrap any command-line tool into an agent with intelligent interaction capabilities. It provides a standardized 7-stage wrapping pipeline, supporting real backend integration, JSON output, REPL interactive mode, and refine incremental completion. Through an output validation mechanism, it ensures data validity and lowers the barrier to making CLI tools intelligent, making it suitable for developers who need to convert specific CLI tools into intelligent agents. When using the plugin, run it with the current process privileges and check the source code and license for security.
Read Moredsh-delayed-task: Letting Agents Truly Come Back Later
dsh-delayed-task is a DeepSeek Harness plugin designed to address delayed decision-making in agents. It stores tasks persistently, supports waking an agent after the page is closed, and re-evaluates based on current evidence rather than mechanically executing outdated plans. The plugin offers two authorization modes: a default re-evaluation-only mode restricted to observation tools, and a limited background execution mode that requires user confirmation. Its main tools include creating, viewing, and canceling tasks, as well as revoking authorization. It relies on the DSH background process remaining active and is suitable for scenarios such as waiting for asynchronous tasks or checking service status.
Read Morelbryany/dsh-herdr: DeepSeek Harness Herdr Status Integration Plugin
The `lbryany/dsh-herdr` plugin is used for DeepSeek Harness (DSH) status integration, aggregating root and sub-agent states into Herdr semantic states. It listens for events and maps DSH process states to Herdr states such as blocked, working, idle, or release-agent. The plugin communicates with Herdr preferably via Unix sockets, supporting persistent connections and session recovery. Installation requires a DSH and Node.js environment, making it suitable for developers who need to view DSH status in real time in the Herdr dashboard.
Read MoreDeepSeek Harness Plugin: dsh-build-diff
DeepSeek Harness (DSH) has released dsh-build-diff, an Agent loop change review plugin designed to visually review changes made by the Agent to files in a workspace. This official dual-sided plugin has zero dependencies and provides core features such as line-level diff review in the right sidebar, one-click change rollback, and message bubble summaries. Installation requires dsh-better-sidebar as a prerequisite. Notes include: it only tracks loops after activation, large files cannot display diffs, reverting old changes may overwrite newer modifications, and there is a limit on the number of files that can be scanned.
Read Moredsh-agent-teams-meta: A Standalone Agent Teams Plugin for DeepSeek Harness
`dsh-agent-teams-meta` is a DeepSeek Harness plugin designed to enable multi-agent parallel collaboration through the "everything is a plugin" philosophy. It supports persistent team management, dynamic routing, task graphs, private mailboxes, and web-based visualization, allowing a lead agent to decompose work among members using different models. The plugin provides a family of `teams_meta_*` tools for creating teams, assigning tasks, and sending instructions, and can track token usage by routing. After installation, a Node.js environment must be configured. All data is stored in a local directory, with no sensitive information saved.
Read Moredsh-telegram-bridge: A Bridge Plugin Between Telegram and DeepSeek Harness
`dsh-telegram-bridge` is a Telegram plugin for DeepSeek Harness (DSH) that enables private chat interaction, session management, and status monitoring with DSH Agent via Telegram. The plugin features message queue management, structured response rendering, multi-session switching, and stability safeguards such as queue persistence and rate-limit protection. After users install it via the CLI, they can use Telegram commands such as `/start` and `/interrupt` to control the Agent. The configuration file requires setting a Bot Token and Owner ID, and security settings can be modified through the Web UI. The plugin runs within the DSH process and does not require an additional server.
Read Moredsh-digital-life: Adding Configurable Digital Life Agents to DeepSeek Harness
The DSH plugin `dsh-digital-life` provides a solution for managing conversational digital life agents, supporting the persistent storage, import, and management of agent identities. Users can create agents through the settings interface and invoke them using tools in separate sessions or the main conversation. Core features include single consultation, concurrent consultation by category, and read-only sessions without a Workspace. When using the plugin, note the build dependencies and category limitations.
Read MoreDeepSeek Harness (DSH) Plugin: dsh-tools-meta
`dsh-tools-meta` is a plugin for DeepSeek Harness (DSH) designed to provide models with "self-evolution" capabilities, enabling agents to dynamically add and remove tools. By leveraging model metacognition, the plugin implements dynamic tool registration and persistence using simple TypeScript scripts, with zero additional dependencies, minimal code, and high readability. Models can autonomously author script files as writable assets, shifting from merely "using tools" to "creating tools."
Read Moredsh-evolve: DeepSeek Harness Local Execution Experience Plugin
dsh-evolve is a plugin that runs within the DeepSeek Harness process, aimed at addressing recurring Agent failures and the lack of locally retained experience. It uses deterministic signals to identify errors and non-progressing trajectories, and provides core capabilities such as record retention, local distillation, background mining, and retrieval filtering. Users can enable the plugin through simple installation and configuration to optimize Agent execution strategies. It is released under the MIT License and features robust data security and permission boundary controls.
Read Moredsh-open-deep-research: DeepSeek Harness Deep Research Plugin
dsh-open-deep-research is a TypeScript framework based on DeepSeek Harness that addresses the challenges of building research agents for web search, web page reading, and generating cited reports. It offers planning, execution, and multimodal retrieval capabilities, and supports access via CLI, DSH Tool, and APIs. After installing Node.js, pnpm, and the DSH environment, users can run it through the `research` or `research-jina` profile; the former supports search only, while the latter integrates Jina to enable web page reading. The project is currently in Alpha preview and has limitations such as long-document truncation, but it is well suited to quickly building complex research workflows.
Read Moredsh-turn-budget: Adding fail-closed steps, tools, and token budgets to DeepSeek Harness per turn
dsh-turn-budget is a fail-closed resource governance plugin for DeepSeek Harness (DSH), maintained by Nunchakus888 and following the MIT license. It independently sets three types of limits for each Agent turn: model step count, tool execution count (including nested dispatch), and token usage reported by the provider. After the quota is exhausted, subsequent requests are rejected before execution, and the turn is closed with a blocked status. The plugin only uses DSH public extension points and implements interception through agent/pre-step, tools/pre-execute, tool/result, and turn/end hooks, with results preserved in standard Session logs. token 上
Read MoreSpecsRelay for DeepSeek: Handing DeepSeek conversations to the DSH project
SpecsRelay for DeepSeek is an open-source plugin for DSH desktop clients, designed to bridge the gap between DeepSeek web discussions and local project implementation. It allows users to open and log in to the real DeepSeek web interface directly within the DSH client, without the need for manual conversation copying or context organization. Through built-in skills, the plugin automatically retrieves the current full multi-turn conversation, structures it into requirements, asks for supplementary information if necessary, and sends it to a selected DSH project to launch an Agent for immediate processing. The plugin reuses DSH-configured models and does not require additional API keys. It supports four DSH clients, including anywhere-labs and Pilot Harness. Installation is simple, allowing for automatic recognition and installation via the `npx` command. It functions as...
Read Moredsh-rigorquant: DSH Session-level Research Plugin for Experimental/Computational Mathematics
dsh-rigorquant is an MIT-licensed plugin for DeepSeek Harness (DSH) designed to optimize long-term research workflows in experimental/computational mathematics through context-isolated multi-agent collaboration. It addresses the conflation of hypothesis generation, exploration, verification, and implementation by organizing a single session into a research laboratory with boundaries, division of labor, and pre-checks. Its core mechanisms include parallel exploration, independent re-derivation of analysis forms and bounds, elimination of invalid paths via counterexamples, and evidence verification by the meta-validator `rq_check.py`, which rejects passing conclusions lacking evidence. The plugin introduces a "Pre-Implementation Four-Check Battery" to ensure sufficient evidence exists before numerical implementation. For stochastic tasks, it employs fixed seeds and a Law of Large Numbers (LLN) convention, and offers an optional Jacobian MCP upgrade path.
Read More《Cursor Docs》-Terminal
Cursor runs shell commands directly in your terminal. Your Run Mode controls when commands run, when Cursor asks, and when terminal commands enter the sandbox.
Read More《Cursor Docs》-Canvases
Canvases let Cursor create interactive artifacts that render next to the chat. Instead of scrolling through a long markdown table or code block, you get a standalone view, laid out with sections, stats, and tables, that you can reopen, edit, and iterate on.
Read More《Cursor Docs》-Browser
Agent can control a web browser to test applications, audit accessibility, convert designs into code, and more. With full access to console logs and network traffic, Agent can debug issues and automate comprehensive testing workflows.
Read More《Cursor Docs》-Run Modes
Run Modes control how the Cursor agent runs tool calls, and when Cursor interrupts you for approval.
Read More《Cursor Docs》-Agent Security
AI can behave unexpectedly due to prompt injection, hallucinations, and other issues. We protect users with guardrails that limit what agents can do. By default, sensitive actions require your manual approval. This document explains our guardrails and what they mean for you.
Read More《Cursor Docs》-Prompting agents
Direct Agent with text prompts in the chat input. You can attach context, images, and voice, and switch models at any point.
Read More《Cursor Docs》-Plan Mode
Plan Mode creates detailed implementation plans before writing any code. Agent researches your codebase, asks clarifying questions, and generates a reviewable plan you can edit before building.
Read More《Cursor Docs》-Cursor Agent
Agent is Cursor's assistant that can complete complex coding tasks independently, run terminal commands, and edit code. Access in sidepane with Cmd+I.
Read More《Cursor Docs》-Design Mode
Design Mode lets you direct agents with visual prompts. From the browser in the Agents Window, you can click an element, draw on the page, or describe a change by voice. Cursor captures the context it needs and edits the code while you move on to the next change.
Read More《Cursor Docs》-Debug Mode
Debug Mode helps you find root causes and fix tricky bugs that are hard to reproduce or understand. Instead of immediately writing code, the agent generates hypotheses, adds log statements, and uses runtime information to pinpoint the exact issue before making a targeted fix.
Read More《Cursor Docs》-Agents Window
The Agents Window is Cursor's agent first interface. It provides a unified workspace to build with agents across repos and environments, including local, cloud, remote SSH, and more. It combines the power of parallel agents with the depth and control of a development environment.
Read More《Cursor Docs》-Agent Review
Agent Review runs a dedicated code review on your local changes from inside Cursor.
Read MoreDSH Noema: Installing Viewable Long-Term Memory for DeepSeek Harness
dsh-noema is a Noema long-term memory plugin developed by ZSeven-W for DSH. It integrates local-first, non-vector memory into Agents, providing noema_* tools such as recall, retrieval, and import, along with a web-based settings page. Memories are stored in Markdown format in the ~/.agent-memory directory and can be migrated from tools like Cursor, Claude Code, and Codex, with support for service keep-alive. This article introduces its positioning, features, installation commands, and usage considerations.
Read MoreIntegrating AnySearch into DeepSeek Harness: Native Web Search and Vertical Retrieval in One Step
@anysearch/anysearch-dsh is a web tool plugin developed by the anysearch-team for DeepSeek Harness, integrating AnySearch into DSH's web profile. After installation, the Agent continues to use the native `web_search` and `web_fetch` functions to access real-time web search, URL content cleaning, as well as advanced capabilities such as vertical retrieval and batch concurrent processing. It supports anonymous usage without an API key, and credentials can be configured for higher quotas. This article introduces the plugin's positioning, environment requirements, installation commands, key configuration, and typical query examples, while also noting that the plugin operates with DSH process permissions and advising users to verify the source code and license before installation.
Read MoreTurn DeepSeek Harness "Everything is Plugin" into a Runnable Chinese Course Using dsh-harness-tutorial
dsh-harness-tutorial is an MIT open-source Chinese tutorial maintained by yanhua1010, which breaks down DeepSeek Harness's "everything is a plugin" concept into a VitePress site, 8 runnable demos locked at version 0.1.0-rc.6, and a teaching-oriented mini-harness (React + Node.js TypeScript). Although categorized under "Tools and Capabilities" in community directories, the repository itself serves as educational material rather than a runtime capability plugin. This article verifies the directory page and GitHub repository to explain what it is, how to read the four lessons, how to run the demos locally, and how to complete the final project.
Read MoreY Combinator Open-Sources QM: Company-Grade Multi-Person Collaborative Agent Workspace
On July 31, 2026, Y Combinator open-sourced its internal multi-agent framework QM (Quartermaster) under the MIT license, and it quickly gained over 10,000 stars on GitHub. QM is a work-agent infrastructure for entire companies, supporting both Slack and Web terminals. Each employee and channel has isolated memory, sandboxes and permissions, and can switch Harness among Pi, OpenCode, Codex and Claude Code. This article sorts out its architecture, Strict/Auto/Dangerous security policies and qm init self-hosted deployment method, for teams evaluating enterprise-level agent operating systems for reference.
Read MoreThe Two AI Coding Powerhouses in 2026: How Claude Code and Cursor Will Divide Labor
In 2026, a new consensus has formed in the developer community: Claude Code (terminal Agent + Claude Agent SDK) and Cursor (AI IDE) are often used in combination rather than being an either-or choice. Based on official documentation from Anthropic and Cursor, as well as public reviews from Faros.ai, FutureProofing.dev and other sources, this paper compares the differences between the two tools in terms of MCP integration, subagents orchestration, multi-model routing, Bugbot review and token efficiency, proposes a practical division of labor scheme such as "Cursor for daily coding, Claude Code for cross-file refactoring and CI automation", and explains that Claude Code can coexist as an extension of Cursor.
Read MoreMajor Version Update of MCP: Stateless Core, Official Extensions and Enterprise-level Certification
On July 28, 2026, the Model Context Protocol released its 2026-07-28 specification, marking the largest revision since the protocol's launch. The core changes include: shifting from a bidirectional stateful protocol to a request/response stateless model, removing the initialize handshake and Mcp-Session-Id; upgrading MCP Apps and Tasks to official extensions; and strengthening OAuth 2.0/OIDC authentication to integrate with enterprise IdPs. With monthly SDK downloads exceeding 400 million and over 950 Claude connectors, MCP is becoming the de facto standard for Agent interconnection.
Read MoreGPT-5.6 Luna's Price Cut Twice: How OpenAI Cut Agent Reasoning Cost to $0.2 per Million Tokens
On July 30, 2026, OpenAI announced an 80% cut in the input pricing of GPT-5.6 Luna API to $0.2 per million tokens, a 20% reduction for Terra, and no change for Sol. This move came less than three weeks after the general availability of the three-tier models on July 9, reflecting the intensifying AI inference cost war. This article sorts out the new pricing lists for Sol, Terra and Luna, the changes to Prompt Caching and Codex quotas, as well as selection suggestions for cost-sensitive Agent workloads.
Read MoreMCP's Largest-Ever Update: Stateless Core Shift on 2026-07-28, Agent Infrastructure Goes Enterprise-Grade
On July 28, 2026, the Model Context Protocol released its 2026-07-28 specification: the core of the protocol has shifted from a bidirectional stateful model to a stateless request/response architecture, the initialize handshake and Mcp-Session-Id have been deprecated, and it now supports any instance handling requests after load balancing. It simultaneously introduces MRTR multi-turn interaction, Mcp-Method/Mcp-Name header routing, and list response cache hints, and graduates MCP Apps, Tasks and enterprise managed authorization to official extensions. OAuth hardening includes RFC 9207 iss verification and DCR migration to CIMD. The monthly downloads of Tier 1 SDKs are nearly 500 million, and companies such as Anthropic and Google Cloud have stated that this is a key step toward the productionization of Agents.
Read MoreOrca: An ADE Running Five Coding Agents Simultaneously, A New Parallel Orchestration Paradigm with Over 20,000 Stars on GitHub
Orca, an open-source project by Stably AI, was a trending ADE (Agent Development Environment) on GitHub Trending in July 2026. Built on Git Worktree, it provides isolated parallel runtime environments for over 30 CLI Agents including Claude Code, Codex, and OpenCode. It supports one-click Worktree creation from GitHub/Linear tasks, UI context injection via Design Mode point-and-click, as well as SSH remote Worktree. Its GitHub stars exceeded 20,000 in July and continued to rise. This article sorts out the differences between ADE and traditional IDEs, core mechanisms, installation methods and applicable scenarios, to help evaluate multi-agent parallel development solutions.
Read MoreMoonshot Open-Source Kimi K3: An Open-Source Cutting-Edge Model with 2.8T Parameter MoE, 104B Activations and 1M Context Window
On July 27, 2026, Moonshot AI released the full weights of Kimi K3 on Hugging Face and GitHub: a 2.8T total parameter MoE model with 16 activated experts out of 896, 104B activated parameters, 1 million token context window, and native multimodal capabilities. Based on the official README and arXiv technical report, this paper sorts out the key points of the KDA and Stable LatentMoE architectures, interprets the differences between Agent evaluations such as Terminal-Bench and model harness tests, and introduces the API, deployment paths via vLLM/SGLang, and the usage boundaries of the Kimi K3 License.
Read MoreCopilot Connects to Grok 4.5 and Claude Opus 5 in July, IDE Becomes a Battlefield for Agent Model Selection
On July 24 and 28, 2026, GitHub's official Changelog consecutively announced that Claude Opus 5 and Grok 4.5 have been integrated into GitHub Copilot, and users can switch between them via the model selector in VS Code, Copilot CLI, JetBrains and other entry points. Opus 5 is designed for complex long-chain Agent tasks, while Grok 4.5 boasts a maximum context window of 500,000 tokens and parallel tool calling. For Business/Enterprise plans, the feature requires administrators to enable the relevant policy, and Grok 4.5 is disabled by default. This article sorts out the access scope, plan differences, billing methods and selection references based on the official announcements.
Read MoreOmniRoute Sees Over 9,000 Star Gains in a Week: Unifies 290+ Model Providers via Local AI Gateway
OmniRoute (diegosouzapw/OmniRoute) is one of the open source projects with the fastest growth in stars in the GitHub AI Agent category in July 2026, with an increase of approximately +9,200 in 7 days. This MIT-licensed local AI gateway aggregates over 290 model providers (including over 90 free tiers) through a single OpenAI-compatible endpoint, with built-in Combo automatic fallback, quota-aware routing, cost telemetry, and support for MCP/A2A. It can connect to programming agents such as Claude Code, Cursor and Codex. This article introduces the pain points of multi-model switching it solves, the core routing mechanism, quick start steps, as well as the security and trust boundaries during deployment.
Read More