Every DeepSeek Harness plugin

What it is — Git status drawer plugin for DSH Web UI that displays a commit DAG lane graph on the right edge with branch operations.

Who it is for — Users who need to view and manage Git branch status in DSH Web chat interfaces are suitable for installing this plugin. Users who do not rely on DSH Web UI for Git tasks can choose not to install this plugin.

Watch out — Sandbox testing passed in a fresh profile and was registered by harness. No obvious issues found.

The verdict — I would install it because it provides commit DAG visualization and branch operations directly in the DSH Web UI without additional build dependencies.

What it is — Provides a left-edge user-Turn navigation rail for long DSH conversations, with current position highlighting, hover previews, jump controls, and local search.

Who it is for — In long conversations using DeepSeek Harness where quick access to turn history, model answers, and metrics is needed, this plugin is suitable. For short conversation users, the plugin can automatically hide without extra loading.

Watch out — This plugin is compatible only with a specific version of DeepSeek Harness 0.1.0-rc.6 and a commit; install via npm to Web profile then restart the process. Sandbox test passed in a new profile, successfully registering it. No obvious issues found.

The verdict — I would install it only on its specific DeepSeek Harness version because it passed sandbox testing and supports local search.

What it is — Adds translucency to DeepSeek Harness shell with opacity slider and custom full-page background.

Who it is for — When needing to add custom background to DeepSeek Harness interface for better visual effects, install this plugin. If the user does not care about interface transparency and background settings, they can skip this plugin because they can use default settings.

Watch out — Sandbox test passed in new profile with successful registration to harness. No obvious issues found. The plugin will execute shell commands.

The verdict — I would install it because it achieved translucency via ThemeRuntime and the sandbox test passed.

What it is — It captures snapshots of AI file write/edit operations in VS Code to support inline review, including accept/reject hunks and per-file undo without using git.

Who it is for — When using DeepSeek Harness for AI file write/edit in a VS Code environment, users who need inline review capabilities are suitable for installing it. If you are not using Trae or VS Code as the editing interface, it is not suitable because it requires the VS Code / Trae extension.

Watch out — The sandbox test passed. The plugin was installed in a new profile and registered into the harness profile. Static checks found no issues.

The verdict — I would install it because it supports the review_revert tool for version-guarded file restoration, preventing overwrite of manual modifications.

What it is — Integrate Metaso search interface, providing page summaries, full Markdown reading, and multi-scope search functionality.

Who it is for — If you need to perform web research in a DeepSeek Harness agent and require page summaries and full content analysis, this plugin is suitable. If you already have other search tools, you can turn to different plugins.

Watch out — Sandbox testing has not been conducted. No obvious issues found.

The verdict — If I were you, I would install it because it provides Metaso search and multi-scope content analysis.

What it is — Registers a vision tool for agents to describe local images.

Who it is for — When an agent needs to describe local images in task scenarios, users missing image description capability in their model would benefit from installing this plugin. Agents that only handle text inputs without requiring a vision tool do not need to install it.

Watch out — No sandbox testing performed, no obvious pitfalls found. Requires configuration of an OpenAI compatible vision endpoint with no built-in keys. Optional multi-model cross-check available.

The verdict — If I needed to describe local images, I would install it because it has no built-in keys.

What it is — Adapts DeepSeek Harness web GUI for narrow screens.

Who it is for — Users conducting web interactions with DeepSeek Harness Web GUI on narrow screens are suitable for installing this plugin. Users not using the DeepSeek Harness Web GUI interface do not need to install this plugin because it is designed specifically for the web GUI.

Watch out — Sandbox testing has not been performed. No obvious issues found.

The verdict — I would not install it because static checks indicate shell command execution.

What it is — This plugin provides LAN remote access support for the DeepSeek Harness Web UI, including 0.0.0.0 binding and token gate.

Who it is for — If you need to access the DeepSeek Harness Web UI from a trusted local network, if your access is limited to local single-machine use without remote needs, you do not need this plugin.

Watch out — No obvious issues were found. The static check indicates it executes shell commands.

The verdict — I would not install it because it requires trusted networks only and the token gate is authorization rather than encryption, increasing the risk of any reachable device controlling the agent.

What it is — dsh-coyote is an Agent- and GUI-controlled DG-LAB Coyote e-stim plugin for DeepSeek Harness, offering eight coyote_* tools and a web panel within one safety envelope.

Who it is for — Users of DeepSeek Harness agent who require control of e-stim devices are suitable for installing this plugin. Users who prefer direct interaction with other e-stim frameworks may choose different tools.

Watch out — Sandbox testing passed in a new profile with successful registration to Harness. No obvious issues were found.

The verdict — I would install it because it supports agent and GUI control within a safety envelope.

What it is — Shows a live USD cost badge for the open conversation in the Web UI of DeepSeek Harness.

Who it is for — Users who need to monitor costs in conversations using DeepSeek Harness Web UI are suitable to install it. Users who prefer not to edit the cost pricing configuration file can skip it.

Watch out — Sandbox test passed: installed successfully in a fresh profile and registered into the profile by harness. No obvious issues found. It reads ~/.dsh/cost-pricing.json on boot and writes the default table if not present, requiring a restart of the web profile after changes.

The verdict — I would install it because it reuses the token-meter's sample replacement rule to avoid double-counting through the session-projection seam without extra wiring.

What it is — DeepSeek Harness plugin that supports agent-driven planning with kanban_* tools and integrates per-workspace project kanban board browser UI.

Who it is for — Users where the agent needs to add and update project cards during conversation. Not suitable for users who do not want additional tools to appear in the agent's toolset.

Watch out — Sandbox test passed: installed in a fresh profile and registered in the profile. Some installation methods may require manually allowing build scripts. No obvious issues found.

The verdict — I would install it because it supports undo and templates.

What it is — Displays real-time tokens-per-second in the "Deep diving…" status line.

Who it is for — Users debugging token streaming outputs with DeepSeek Harness web can install this plugin. Users who only need average values from built-in StatsLine can skip it.

Watch out — Installation requires manually allowing the build script. In real test it can be installed but pnpm blocks the build. No other obvious issues found.

The verdict — I would install it because it provides pure-frontend real-time TPS monitoring for web debugging tasks, though manual build allowance is required.

What it is — The dsh-frosted-window plugin creates a full-window frosted glass theme by uploading an image for sidebar, conversation, and details.

Who it is for — If you run the web profile in DeepSeek Harness and want to add custom image background with shared frosted glass effect to sidebar, conversation and details, it is suitable to install. Those who do not want the plugin to save images in IndexedDB and settings in localStorage may find it unsuitable as it uses these for storage.

Watch out — Sandbox test shows it installs after manually allowing the build script for github:SenryLee/dsh-frosted-window. No other obvious issues were found. Users must add allowBuilds to pnpm-workspace.yaml for source installs.

The verdict — I would install it because it allows saving custom images and sliders with Save/Delete preview, but only after manually allowing the build script from source, otherwise registration fails.

What it is — The dsh-sampling-sliders plugin for DeepSeek Harness adds temperature and maxTokens sliders to the composer interface. These override model request parameters for all providers through the agent/request hook, supporting hot and persist modes.

Who it is for — Developers who generate content with DeepSeek models and want to fine-tune sampling parameters for every API call can use this plugin.

Users who do not rely on model sampling adjustments or work with providers outside the hook system have no need for it.

Watch out — Building from GitHub source and adding to a profile is required for installation. Sandbox testing confirmed successful registration in a fresh profile. No obvious issues were found, though the interceptor deactivates after plugin removal, and the LlmCallConfig layer only exposes temperature and maxTokens.

The verdict — I would install it because it provides persistent control over sampling parameters across all providers, which offers value in iterative model tuning workflows.

What it is — DeepSeek Harness web composer provider-tabbed model picker.

Who it is for — When using DeepSeek Harness web composer to handle multi-provider model browsing, you can install this plugin to get independent scrollable lists and cross-provider search. If you use other interfaces or tools, you don't need this plugin.

Watch out — Sandbox test passed: successful installation and registration in a new profile. No obvious issues found.

The verdict — I would install it because it shares the model directory with the built-in without introducing extra state, but the premise is that your web profile includes the model selection component.

What it is — This plugin provides a neon digital arcade HUD skin for the DeepSeek Harness Web GUI, integrating pixel fonts, animated sprites, and a custom crosshair cursor.

Who it is for — When using the DeepSeek Harness Web GUI interface, this skin provides neon glows, pixel animations, and scan feedback for buttons, inputs, and HUD elements. Users who prefer the official default color scheme do not need this plugin.

Watch out — Sandbox testing passed; it was successfully registered and loaded in a fresh profile. It is compatible with the settings panel, ensuring complete display of masks and panels when opened. No obvious issues were found.

The verdict — I would install it because it offers system theme synchronization and unique interactive feedback, making web GUI operations more engaging.

What it is — In DeepSeek Harness Web the bottom right floating mini player streams Bilibili videos with native player including danmaku comments. It pauses when agent needs you and resumes precisely from original position.

Who it is for — When using DeepSeek Harness to handle tasks and wanting to watch Bilibili videos to relax, lacking a player that pauses on agent needs and resumes precisely. People who want complete danmaku and comments not suitable to install, because it uses lightweight self-drawn danmaku layer for precise resume, with simple style.

Watch out — Installation is plug and play with no configuration needed. Sandbox test passed, it was installed in new profile and registered by harness. No obvious issues found.

The verdict — I would install it because the client bundle is pure handwritten JavaScript with no build step required.

What it is — Integrates Unreal Engine .uasset inspection and Blueprint Python editing.

Who it is for — Users working with Unreal Engine projects who need .uasset parsing or Blueprint editing are suitable for this plugin. If not using DeepSeek Harness for UE asset management, users can choose alternative approaches.

Watch out — It passed sandbox testing in a new profile where it was registered into the profile. No obvious issues found.

The verdict — I would install it because the sandbox test confirmed successful registration.

What it is — Cooperative timeout guard for ask_user_question in DeepSeek Harness: lost or unanswered questions resolve as structured ASK_TIMEOUT instead of hanging the turn forever.

Who it is for — When using DeepSeek Harness to handle user questions that may be interrupted by browser backgrounding or network drops, this plugin is suitable for your profile. Users not employing DeepSeek Harness for interactive agent turns do not need it.

Watch out — Build script must be manually allowed when installing from source. Sandbox test passed on a fresh profile with automatic registration. No obvious issues found.

The verdict — If agent turns in your workflow may hang due to connection instability, this provides structured timeout resolution for graceful turn end, making it worthwhile.

What it is — Plugin for message review and quote chips in conversations.

Who it is for — Users conducting multi-model code reviews and long conversation analysis need review reports and history review. Users not relying on host-persisted preferences.

Watch out — Sandbox test passed with installation and registration in new profile. Limited to 30 reviews per minute and 16 concurrent tasks.

The verdict — I would install it because it supports review history.

What it is — This plugin serves as the Ollama Cloud chat adapter for DeepSeek Harness. It performs direct NDJSON translation of the Ollama /api/chat wire format, enables model discovery including context windows and capabilities, and registers Web search/fetch providers and a settings card under the ollama-cloud identifier.

Who it is for — When using DeepSeek Harness with local Ollama models for chatting and requiring model discovery along with web search and fetch tools, this adapter is suitable. It uses a separate protocol for capabilities from the chat endpoint. If your workflow involves only cloud-based models without needing Ollama-native independent features, another adapter may suffice.

Watch out — Sandbox testing confirmed successful registration in a new profile. No obvious issues were found during static checks. The implementation includes retry policies and timeout settings for web requests.

The verdict — I would install it if I need Ollama's native capabilities for model discovery and web tools, as it cleanly separates the chat protocol from the independent features.

What it is — DeepSeek Harness plugin providing persistent token usage statistics recording provider-reported session, model, and daily activity.

Who it is for — When using DeepSeek Harness to handle session tasks and need to track token usage, this plugin is suitable for users requiring session and model-level statistics. If you only need temporary token counts without persistence, you do not need to install this plugin.

Watch out — Sandbox test passed: installed in a fresh profile and registered by Harness. No issues found in static checks. No obvious problems detected.

The verdict — If I were you, I would install it because it works normally in the profile, but only if you need persistent token statistics.

What it is — Connects to a local Unsloth Desktop server to delegate repetitive text and vision tasks via tools.

Who it is for — When you need to delegate OCR or image comparison tasks to a local multimodal model via unsloth_vision while keeping the main model as the conversation interface. If you handle all vision work directly in the main model or lack a running Unsloth Desktop, this plugin does not apply because it requires the local server setup.

Watch out — Sandbox test confirmed successful registration in a new profile. Static analysis notes possible shell command execution, but no runtime issues observed.

The verdict — I would not install it because it depends on first running and configuring Unsloth Desktop separately to access the tools.