dsh-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 MoreIntroducing the DeepSeek Harness Plugin: dsh-perm-gate
`dsh-perm-gate` is the permission gate plugin for DeepSeek Harness, combining approval, classification, and gating into a single package. It uses a fixed priority chain and a fail-closed mechanism to ensure that high-risk operations cannot be bypassed. Core features include command blacklists, session authorization, LLM-based risk-tier adjudication, and adjudication learning. The plugin supports a pure-function rule engine, an audit log stream, and an automatic review mode, resolving cross-plugin version coupling issues while providing a unified audit trail and fine-grained control.
Read MoreIntroduction to the dsh-llm-openai-completions Plugin
`dsh-llm-openai-completions` is an adapter in DeepSeek Harness for connecting to custom gateways, primarily resolving compatibility issues related to role definitions, thinking mode, and visual input. Its core features include enforcing the use of the system role, separating thinking content, supporting image uploads, and adapting to various Qwen model formats. Users can install it via the DSH command line and enable it in their configuration file. Please note that maintenance for this project was discontinued in October 2026, remains at version 0.1.0, and the 0.2.0 version with visual features was never released.
Read Moredsh-context-compression-improved: Context Compression and Code Skeleton Compression Plugin for DeepSeek Harness
The `dsh-context-compression-improved` plugin in DeepSeek Harness addresses context bloat in agent tasks through mechanisms such as pre-compression, aggregate replacement, and history compaction. Its core new feature, the Code Skeleton Gate, substantially reduces oversized source-code tool results by retaining only declarations and key lines, significantly improving inference efficiency. The plugin supports multiple strategies, including Fresh and Aggregate, is compatible only with DeepSeek models, and requires manual enabling of code skeleton compression in the settings.
Read Moredsh-openspec: OpenSpec Workflow Skills and Command Alias Plugin
The `dsh-openspec` plugin is designed to address the integration of DeepSeek Harness (DSH) with the OpenSpec workflow. It registers OpenSpec’s 12 core skills as `openspec-*` specification skills and provides `/opsx-*` command aliases, supporting invocation by the model and users. The plugin depends on Cordis, is easy to install, manages state through a standard skills registry, and requires no additional runtime configuration.
Read MoreDeepSeek Harness Bidirectional Sync Plugin (dsh-sync-plugin)
The DeepSeek Harness (DSH) bidirectional sync plugin uses a private GitHub repository to synchronize sessions, attachments, and workspace configurations across multiple devices. It supports manual or automatic synchronization and features grouped session migration, conflict handling, and patch hosting. Users must install the plugin and authorize GitHub; it is suitable for technical users seeking a consistent workflow across multiple devices, though privacy risks should be considered and the application must remain running.
Read Moredsh-plugin-updater: DeepSeek Harness Plugin Secure One-Click Updater
dsh-plugin-updater is a plugin update and uninstallation tool for DeepSeek Harness (dsh). It securely manages pnpm dependencies through a backup, verification, launch validation, and failure rollback process, preventing configuration corruption caused by manual operations. It provides features such as manifest viewing, one-click updates, granular uninstallation, smart restarts, and rollbacks, supporting both settings-page and command-line operation modes to effectively reduce environment maintenance risks.
Read Moredsh-v4flash-tiler: An Automatic Large-Image Tiling Plugin for DeepSeek Vision Models
`dsh-v4flash-tiler` is an automatic image tiling plugin for the DeepSeek v4Flash vision model, designed to mitigate detail loss caused by large images exceeding 800×800 pixels. It automatically slices images larger than 1024 pixels into grid tiles with row and column coordinates and overlap regions, ensuring the model can reconstruct the original image according to the coordinates. It supports automatic tiling in the chat interface, manual command-line analysis, and Python API calls, making it suitable for processing large screenshots and dense-text scenarios.
Read Moredsh-memory: A Long-Term Memory Plugin for DeepSeek Harness
dsh-memory is a long-term memory plugin for DeepSeek Harness (dsh), designed to address the context window limitations of agents and the fragmentation issue in conversational memory. It uses persistent JSON storage, automatically guides the model to call `memory_recall` before responding to retrieve relevant background information, and calls `memory_save` after responding to store new facts. The plugin provides five core tools (create, read, update, delete, and search), supports a 1-5 star importance rating, and includes a visual web-based management interface, intelligent search, and import/export capabilities. It can inject key memories into the context, effectively improving the memory coherence and interaction quality of agents.
Read Moredsh-ui-skin-switcher: DeepSeek Harness Model and Reasoning Effort Switcher
`dsh-ui-skin-switcher` is a plugin built for the DeepSeek Harness Web interface. It provides a compact Claude-style switcher next to the input box, supports selecting models grouped by vendor, and offers an animated thinking-intensity slider for models that support reasoning. The plugin includes per-model memory and automatic theme adaptation. It is ready to use after installation and restart, but note that it only supports DSH 0.1.5+, and the thinking intensity must be configured manually—an incorrect value will silently fail.
Read Moredsh-rollback: Conversation and File Rollback Plugin for DeepSeek Harness
`dsh-rollback` is a conversational and file-state synchronized rollback plugin for DeepSeek Harness (DSH), designed to address the issues of inefficient manual rollback and state inconsistency in agent development. The plugin supports creating checkpoints before each turn and managing historical states using a 10-turn sliding window, enabling restoration of file contents, deletion of files, and handling of newly created files. It provides both Web-based visual buttons and command-line triggering, supports change previewing and in-place truncation. The plugin requires DSH version 0.1.1-rc.2 or higher and includes a comprehensive error-handling mechanism.
Read MoreIntroducing the dnalec/dsh-auto-approve Plugin
`dnalec/dsh-auto-approve` is a DeepSeek Harness auto-approval plugin designed to balance efficiency and security in automated workflows. It automatically handles tool calls by using keyword matching, review model decisions, and risk-leveling strategies. The plugin supports configurable deny, manual review, and allow policies, and provides parameter interception, request limits, and custom tool extension capabilities. Users must install the plugin and restart the service, then configure keywords and decision tables to define the handling rules for different risk levels, enabling secure and efficient automated workflows.
Read Moredsh-plugin-gemini-theme: Give the DeepSeek Harness Web Client a Gemini Skin
The dsh-plugin-gemini-theme plugin for DeepSeek Harness (DSH) overrides the `--dsw-*` design tokens via a single stylesheet, allowing the Web client interface to be customized into a Google Gemini-like style without modifying the source code. It mainly resets styles for the chat area, sidebar, input box, animations, and glow effects. Its styles have extremely high specificity and do not interfere with the DOM structure. Developers must use specific commands to install and remove it, and should be aware of the risk of unintended style matches caused by class-name matching. Additionally, the session header, native carousel, and similar elements are not covered.
Read Moredsh-persona: Adding a Collaborative Style Layer to DeepSeek Harness
`dsh-persona` is a dual-end plugin for DeepSeek Harness, designed to add a configurable layer of collaborative style to model outputs. It appends system prompts to requests to optimize communication tone and progress reporting, without changing DSH's core logic or security policies. Users can select built-in styles such as “Friendly Collaboration” or “Pragmatic and Direct” in the web interface, or customize them using “Shared Prompt Blocks.” Once installed and enabled, the plugin takes effect at runtime, helping developers standardize the collaborative behavior of models.
Read MoreIntroduction to the DeepSeek Harness Voice Input Plugin
`dsh-voice-input` is a voice input plugin for DeepSeek Harness. It uses the browser's native Web Speech API to provide real-time speech-to-text without requiring a self-hosted service. After installation, users can click the microphone button on the left side of the input box to start or stop recording. The recognized text is appended to the input draft in real time and can coexist with manually typed content. The plugin is compatible with the DSH theme, but it is not available in Firefox and requires an internet connection to use cloud-based recognition services.
Read Moredsh-jenkins-panel: AI-Invocable Jenkins Tools and Sidebar Panel for DeepSeek Harness
The dsh-jenkins-panel plugin aims to seamlessly integrate Jenkins capabilities into DeepSeek Harness (DSH), addressing the fragmentation between AI and CI/CD tools. Through 22 `jenkins_*` tools and a right-side visualization panel, it enables build triggering, monitoring, log viewing, and automated failure analysis. Developers can manage Jenkins jobs directly in DSH, use AI-assisted investigation to troubleshoot issues, and achieve closed-loop management across the entire build workflow. Installation requires dsh version 0.2.0 or later; explicitly specify the version to avoid compatibility issues.
Read Moredsh-plugin-auto-review: Provider-Based Automated Approval Review for DeepSeek Harness
`dsh-plugin-auto-review` is a DeepSeek Harness (DSH) plugin aimed at reducing manual intervention in automated workflows by having models replicate the approval-review behavior of specific providers. It supports Codex Guardian and Grok escalation strategies, enabling users to configure custom review logic and LLM routing. It handles native approval requests and Bash invocation retries by inheriting machine reviewers rather than triggering manual prompts. Installation requires specifying a Profile and restarting the process; its configuration file is stored in `plugins/auto-review/auto-review`. The plugin requires Node.js >= 22.13.0, is high-privilege, and is responsible for model traffic security and
Read Moredsh-living-memory: A Localized Self-Maintaining Memory Plugin for DeepSeek Harness
dsh-living-memory is a localized agent memory plugin designed specifically for DeepSeek Harness. It uses SQLite for data storage, simulates hippocampal functions through a self-maintenance mechanism (deduplication, consolidation, and decay), and supports hybrid recall with Chinese-language optimization. The plugin provides read/write tools and a Web telemetry panel, runs with zero configuration by default, ensuring data privacy and efficient retrieval.
Read MoreDSH Maestro Supervisor: DSH Web Recovery Daemon
DSH Maestro Supervisor is a daemon and plugin combination designed to address state loss caused by DSH Web process crashes or network interruptions. It uses a daemon independent of the Web process tree to check service status every 3 seconds, automatically rolling back to the last known good (LKG) snapshot when a crash occurs and restoring interrupted sessions. Core features include: rollback using SHA256-verified LKG snapshots, automatic scanning and recovery of sessions marked as interrupted within the past 5 minutes, and health report generation. During installation, the plugin must first be built and added to the configuration, and it is recommended to run the daemon via Systemd. The system controls automatic recovery behavior through `~/.dsh/.supervisor/config.json`. Additionally, it supports ... via the RPC interface or command
Read Moredsh-maestro-review: A Pluggable MR Review Pipeline for DeepSeek Harness
`dsh-maestro-review` is a pluggable MR review plugin for DeepSeek Harness, designed to address the automation and standardization of code merge review. The plugin receives GitLab events via Webhook, runs review or audit agents in isolated worktrees, and uses the `review-findings` tool to output structured results. Its core features include a unified `ReviewProvider` interface, support for disposable worktrees, headless execution in CI environments, and event deduplication mechanisms. It provides multiple workflows, such as quick review and deep review, supports reusing images and templates in CI, and enables automated pipelines. During deployment, attention must be paid to version synchronization (between npm and Docker images) and environment dependencies, and it integrates well into existing systems.
Read Moredsh-maestro-remote: DSH Remote Access and Tunnel Management Plugin
`dsh-maestro-remote` is a plugin that provides remote access capabilities for DeepSeek Harness. It uses Cloudflare Tunnel to expose local services to the public internet, supports PIN-based authentication to protect the proxy entry point, and allows receiving Webhooks from services such as GitLab. The plugin offers features including tunnel lifecycle management, automatic download of `cloudflared`, traffic distribution, and configuration reloading. Configuration is managed through a shared namespace, and the plugin depends on a Node.js environment. It is designed to address the limitation of local sessions being inaccessible from external systems or unable to receive externally triggered requests.
Read Moredsh-maestro-observe: Local-First Observability for the Maestro Plugin Stack
`dsh-maestro-observe` is a local, lightweight observability tool designed specifically for the DeepSeek Harness and Maestro plugin stacks. It uses SQLite for local storage of traces, costs, latency, and health status, and provides CLI and RPC endpoints for querying. The plugin supports analysis grouped by tool or session, and includes features such as budget control and error monitoring. Developers should be aware of its Node.js version dependency and the fact that sensitive information is redacted before data is persisted.
Read Moredsh-maestro-notifier: A Pluggable Notification Service for DeepSeek Harness
`dsh-maestro-notifier` is a notification plugin in the DeepSeek Harness suite, designed to provide a pluggable notification service layer. It integrates the Telegram transport by default, supports sending messages via the `maestroNotifier` service, and follows default target resolution and Never-Throw design principles. It requires Node.js 22.19.0+ or 24.0.0+ and Cordis 4.0.4+; as a pure transport-layer implementation, developers can call it directly to reduce integration costs.
Read Moredsh-maestro-mobile: Mobile Adaptation for the DeepSeek Harness Web UI
`dsh-maestro-mobile` is a mobile-device adaptation plugin for DeepSeek Harness. When a touch-enabled device with a screen width below 1024px is detected, the plugin converts the sidebar into a drawer, changes dialogs into bottom sheets, and optimizes safe-area handling and layout to prevent element compression. It supports tablet compatibility, notch-screen adaptation, and diagnostic mode. Installation is performed via a DSH plugin command and requires Node.js ≥ 22, pnpm ≥ 11, and Chromium 105+. By default, the plugin is active only in mobile touch mode and acts as a no-op on desktop.
Read Moreddtcorex/dsh-maestro-meta
`ddtcorex/dsh-maestro-meta` is a lightweight v2 aggregator designed to simplify plugin installation and version management for complex Maestro Harness workflows. It re-exports eight core components via `cordis.patch.yml`, including Remote Agent, MR Review, Govard Bridge, Persistent Memory, Mobile UI, Notification Service, Configuration Store, and Developer Dashboard. Developers can install it with a single command and without writing any runtime code. It is well suited for scenarios requiring rapid initialization or centralized management of the full DSH-side infrastructure, though note that Govard must be installed separately.
Read More