dsh-login-gateway: Add TOTP Login to the DeepSeek Harness Web UI

`dsh-login-gateway` is a security authentication plugin for DeepSeek Harness (DSH) that adds username, password, and TOTP two-factor authentication in front of the Web UI. It listens on a dedicated port, performs server-side token exchange with Harness using an HMAC cookie, and supports traffic forwarding, rate limiting, and initial setup. It must be placed behind a reverse proxy (such as Caddy), and accounts should be managed using scripts. Installation requires specific environment prerequisites; if build issues occur during installation, configure `allowBuilds`. This plugin addresses the security risks associated with exposing DSH's default authentication mechanisms to the public internet.

Read More
dsh-sidebar-pins: Codex-Style Sidebar Pinning and Full-Row Context Menu

dsh-sidebar-pins is a plugin for DeepSeek Harness designed to address fragmented sidebar management. It integrates pin and menu capabilities, renders its own pinned bar while parsing official rows, and ensures every row has access to the full menu. The plugin supports pinning, renaming, archiving, and forking sessions and workspaces, taking over the official “⋯” button. Data is stored locally, with migration support from previous plugin versions. The project requires attention to changes in DSH’s underlying DOM structure and does not include a session deletion feature.

Read More
dsh-zcode-scribe: Read-Only Memory Room and Path Security Mechanisms of DeepSeek Harness

`dsh-zcode-scribe` is a security memory plugin for DeepSeek Harness (DSH), built on the ZCode memory model. It provides agents with read-only Markdown memory rooms and the `scribe_recall` retrieval tool, while including 52 strict path security rules to effectively prevent directory traversal, Unicode spoofing, and NTFS alternate data stream attacks. The plugin is currently in an early stage, implementing only read-only access and retrieval; write and extraction capabilities are not yet supported.

Read More
dsh-zcode-git: A Structured Git Toolset for DeepSeek Harness

The DeepSeek Harness plugin dsh-zcode-git provides a structured set of Git tools to replace unstable shell invocations. It includes six tools covering status viewing, diffs, logs, branch management, commits, and staging operations. The plugin enforces permission approval for read and write operations and ensures fixed output formats to improve security. Installation requires Node.js >=20 and a specific Harness version, but a defect may occur when the current working directory is located in a subdirectory of the repository.

Read More
dsh-zcode-breaker: Fast-Fill Circuit Breaker for Automated Compression in DeepSeek Harness

The dsh-zcode-breaker plugin is designed to address the "fast-fill" loop issue in the automated compression mechanism of DeepSeek Harness (DSH), which causes significant resource waste. By rejecting fast-fill operations before summarization calls are executed and tracking tool turns via persistent session logs, it prevents redundant compression. It features session-level state management, configurable thresholds, error handling, and administrative commands, and requires specific version dependencies for installation and use.

Read More
dsh-tui-session-manager: dsh-TUI Session Management Plugin

dsh-tui-session-manager provides a UI entry point for session management in dsh-TUI, addressing the gap where the adapter layer already has deletion capability but lacks an interface. The plugin uses a `/sessions` full-screen panel to support session browsing, filtering (including tri-state empty-shell identification), and batch deletion, while synchronously cleaning up residual state in the host deletion path. It features zero dependencies, automatic backup, and fault-tolerant handling, and is compatible with Node.js ^22.19 or >=24.

Read More
DeepSeek Harness Plugin: dsh-plugin-ponytail

`dsh-plugin-ponytail` is a plugin that integrates Ponytail's "laid-back senior developer mode" into DeepSeek Harness (DSH). It injects Always-on rules through `systemPrompt.context`, provides skills and a `/ponytail` command interface to control the mode intensity, and uses the `.ponytail/` directory and `state` file to enable project-level activation and state persistence. Unlike similar plugins that take effect globally or provide only skills, this plugin takes effect only when the project root contains that directory, and is installed as a mutually exclusive option.

Read More
dsh-openrouter-live: OpenRouter Real-Time Model Provider for DeepSeek Harness

The `dsh-openrouter-live` plugin enables DeepSeek Harness to fetch live model lists from OpenRouter, supporting image input, reasoning continuity, prompt caching, and precise cost tracking. Configuration is flexible, allowing model filtering and cost log export. Installation requires meeting specific version requirements, and the configuration file is located at `cordis.patch.yml`.

Read More
dsh-peakvalley: DeepSeek Harness Peak/Off-Peak Period and Price Monitoring Plugin

dsh-peakvalley is a plugin for DeepSeek Harness designed to help users monitor inference costs. By displaying real-time peak/off-peak indicators and price multipliers next to the input field, it helps users adjust their usage strategy based on pricing. The plugin supports online synchronization of Chinese holiday data, offers seven switchable display modes, mouse hover details, and local preference memory. Installation requires cloning the repository and specifying a local path. It automatically recognizes DeepSeek models, is compatible with browser settings, and provides intuitive cost monitoring through local calculations.

Read More
DeepSeek Harness Plugin: OpenCode Go Quota Monitoring

This plugin displays a pill under the input field in the DeepSeek Harness Web UI to monitor OpenCode Go's rolling 5-hour, weekly, and monthly quotas in real time. It supports color-based status indicators (gray/yellow/red) and a detail overlay, and uses a decoupled front-end and back-end architecture to manage API key security. After installation, users simply need to enter their API key in the settings to start using it, enabling usage monitoring without leaving the current page.

Read More
dsh-delivery-library: DSH Session Delivery and Output Files Library

`dsh-delivery-library` is a plugin for the DeepSeek Harness (DSH) client, designed to address the difficulty of retrieving files in long conversations. It adds a "Delivery Files" view tab to the conversation window and automatically aggregates the final deliverables explicitly declared by the model in the current conversation (`present`) as well as files actually written (`write`/`edit`). Core features include displaying items by "Delivery" and "Output" categories, searching by file name or path, and clicking to preview or copy paths. The list is generated solely from events already loaded in the conversation and does not scan the disk. The plugin must be installed via a GitHub specifier and supports loading historical records when tail data is missing.

Read More
DeepSeek Harness Plugin: Real-Time Sidebar Balance + Cost per Chat Turn

`dsh-deepseek-usage` is a plugin for DeepSeek Harness used to monitor account balance and per-conversation costs in real time. By default, it refreshes the sidebar balance every 60 seconds and displays a cost breakdown for each turn at the bottom of the conversation, including token price calculations. The plugin supports Windows/macOS/Linux and requires Node.js >= 22.19 as well as a web interface. After installation, configure `cordis.patch.yml` and set your API key; note that the old package must be removed when upgrading to avoid duplicate display.

Read More
dsh-locale-fr: French UI Localization Pack for DeepSeek Harness

dsh-locale-fr is a DeepSeek Harness (DSH) plugin developed by BlackAngel242, designed to provide complete French interface support. By injecting a French dictionary, the plugin covers 58 namespaces and 2,614 UI keys, with automatic fallback to English to ensure the interface remains available. After installation, users can enable it by selecting "Français" under Settings > General > Language. The plugin is suitable for developers who require a French environment, but note that some external text, such as command descriptions, and new keys added after a new Harness release may not be automatically translated.

Read More
dsh-proof: Evidence-Driven Completion Proof and Regression Attribution Plugin

`dsh-proof` is an open-source plugin designed to convert self-reports from agents in DeepSeek Harness (DSH) into verifiable evidence chains. Through content addressing, signed checkpoints, and offline anchors, it builds a tamper-resistant chain of trust, providing evidence-driven proofs of completion, regression attribution, and real-time verification. The plugin adopts a layered architecture with a core layer decoupled from the host, and includes a DSH plugin, an MCP server, and various adapters. It is applicable to safety-critical development, stringent regression testing, and third-party auditing, shifting agent development from "blind execution" to "verifiable execution."

Read More
DeepSeek Harness Community Plugin: `dsh-agent-observability`

The DeepSeek Harness community plugin `dsh-agent-observability` provides event-based, privacy-first agent observability and reliability analysis. It supports trajectory reconstruction, reliability metric calculation, and failed-case export, addressing challenges such as distinguishing retries from errors in logs. Users can enable it via a command and run `/observe` in the web interface to automatically export a report containing metrics, traces, and evaluation data. By default, the plugin collects metadata only and enforces resource limits on content capture, session count, and event count, making it suitable for developers who need structured analysis of agent execution traces.

Read More
DeepSeek Harness Plugin: A Free Multimodal Trio (Zero-Key Text-to-Image, Video Understanding, and Text-to-Video)

`dsh-free-media` is a free multimodal plugin for DeepSeek Harness, maintained by `bdydgz114514`. It provides three tools and associated Skills: text-to-image, video understanding, and text/image-to-video. Installation requires Node.js >= 18 and a configured FFmpeg. Video generation requires configuring an API key (Zhipu and Agnes are supported). The plugin uses third-party free services, with limitations such as reduced image quality, no audio track, no SLA guarantee, and rate limiting. It is suitable for low-cost trials and validation of multimodal capabilities, but users should be aware of data privacy risks.

Read More
DeepSeek Harness: Tang San Beautification Theme Plugin

TangSan is a visual enhancement plugin for the DeepSeek Harness (DSH) web interface, maintained by users and released under the MIT License. By overriding styles and injecting resources, it sets a Tang San illustration as the full-screen background and replaces square avatars with a brand logo. The plugin provides a dedicated settings panel for adjusting background opacity, blur, dark overlay, and position. It also replaces the browser tab icon. Installation is available via CLI and the built-in manager in the desktop app; the configuration file is stored locally, but note that Manifest syntax errors can cause crashes.

Read More
simple-plan: DeepSeek Harness Plan-Execute-Verify Workflow Plugin

The `simple-plan` plugin for DeepSeek Harness is designed to address issues in multi-agent collaboration, such as coarse task decomposition, lack of independent verification of deliverables, and an uncontrollable repair process. It enforces a Plan-Execute-Verify-Fix loop to transform complex tasks into controllable streams of subtasks. The plugin consists of five core stages: breaking the request down into 2–7 steps, executing the steps with worker subagents, verifying the results with stateless subagents, fixing defects within a maximum of three loops, and finally aggregating the results. It requires Node.js >= 22 and is suitable for complex, multi-step tasks that demand strict quality control.

Read More
plan-guardian-skill: A DeepSeek Harness Plugin Enforcing 7-Step Verification

`plan-guardian-skill` is a DeepSeek Harness (DSH) plugin designed to address the challenge of ensuring agents strictly follow a plan and deliver verifiable outcomes. It enforces a 7-step planning workflow with binary acceptance criteria, executes steps through worker subagents, and conducts strict verification by an independent, memory-based verifier subagent, producing evidence-backed reports with a PASS or FAIL result. Installation requires Node.js >= 22, Python >= 3.10, and a specific DSH version.

Read More
DeepSeek Harness Plugin: Generating and Saving Images to Disk via TokenRhythm

`opencode-tokenrhythm-image-mcp` is a DeepSeek Harness plugin for generating images by calling the TokenRhythm API. It saves generated images locally and returns their paths, supporting the `qwen-image-2.0` and `wan2.7-image` models. The plugin is stdio-based, has no third-party dependencies, and requires an API key and output directory to be configured. Users can enable it on Desktop by adding a repository link, or install it via CLI using the `dsh plugin` command.

Read More
opencode-ppocr-mcp: Local PP-OCRv6 OCR Recognition Plugin

This plugin, `opencode-ppocr-mcp`, is a local OCR tool that runs in DeepSeek Harness. It is based on PaddleOCR 3.7.0 and the PP-OCRv6 Medium model, and performs inference on CPU using ONNX Runtime. It supports text extraction from single images and multi-page PDFs, ensuring all data is processed locally for privacy and security. Installation requires specifying a profile, and the first use downloads an approximately 170 MB model. The plugin is suitable for privacy-sensitive or network-limited scenarios, but does not support GPU or screenshot recognition.

Read More
opencode-eyes-nvidia: Integrating NVIDIA NIM Vision Capabilities into Text Models

This plugin provides visual reasoning capabilities for DeepSeek Harness, enabling text models to process images. Built on NVIDIA NIM, it integrates seven visual models, including MiniMax-M3, and supports automatic key rotation, MCP server packaging, and tools for image description and model queries. Installation requires Python ≥3.8, the Pillow library, and an NVIDIA API Key. Configuration is flexible, supporting multi-key rotation and ThinkingMode tuning, and it is loaded via cordis.patch.yml, making it suitable for scenarios that require cost-effective integration of multimodal capabilities.

Read More
opencode-eyes: Giving Text Models a Pair of Eyes (StepFun step-3.7-flash)

The DeepSeek Harness plugin, opencode-eyes, provides visual capabilities for pure text large language models by mounting StepFun’s step-3.7-flash model to enable multimodal completion. The plugin packages an MCP server for a ready-to-use setup. To use it, users need to configure the `STEP_API_KEY`, with support for local image paths and custom prompts. The plugin allows adjusting parameters such as timeout, image size, and quality, while users should be mindful of runtime directory and permission limitations.

Read More
Nai (Milk Frog): DeepSeek Harness Web UI Appearance Layer Plugin

The Nai plugin for the DeepSeek Harness (DSH) Web interface is a UI customization plugin that integrates the "Nai Frog" mascot into the interface. It supports chat backgrounds, brand logo replacement, and real-time adjustment of opacity and blur, and can be configured from the settings page. The plugin hosts assets via a local server and requires no external resources. Installation requires Node.js 22 and DSH >= 0.1.5-rc.1, with installation path instructions provided for both the Web and desktop versions.

Read More
MDSM: DeepSeek Harness Male DeepSeek Mascot Appearance Layer

MDSM (Male DeepSeek Mascot) is a UI layer plugin for DeepSeek Harness (DSH), used to add custom wallpapers, branding, and settings controls to the Web GUI. It requires no build steps, hosts resources via the DSH web server, and supports parameter configuration such as transparency, blur, and position. It does not access the file system or network, ensuring security, and can be installed and enabled via the command line or a desktop application.

Read More