Every DeepSeek Harness plugin

What it is — Provides native browser tools for DeepSeek Harness, enabling per-agent isolated sessions, bounded snapshots, ref-based interactions, screenshots, and safe routing.

Who it is for — When an Agent needs to perform web navigation and data extraction via model calls, this plugin is suitable; pure-text Agents without web interaction needs do not require it.

Watch out — Tested successfully in a fresh profile where the plugin registers correctly. Agents complete Wikipedia browsing, search, and screenshot in about 15 seconds. It executes shell commands and may download a 200 MB Chromium cache on first use.

The verdict — I would install it because it allows Agents to safely route tasks to a browser using native tools.

What it is — PaperLab is a LaTeX paper workbench that supports annotating any text in PDF and lets a dsh agent revise sources with compile checks and git history.

Who it is for — Users who write and iterate on LaTeX papers in the DeepSeek Harness environment and require dsh agent assistance for revisions, compile verification, and git management would find PaperLab suitable. Users who do not rely on DeepSeek Harness plugin system or git version control for their projects would have no need for this tool.

Watch out — Sandbox testing passed with installation in a fresh profile and registration into harness profile. No obvious defects were found. The plugin executes shell commands, so manual permission may be needed for build scripts.

The verdict — I would install it to integrate PDF annotation-based revisions with DeepSeek Harness agent loops, as it provides compile checks and git history management for LaTeX papers.

What it is — Provides per-call cwd support for subagents in DeepSeek Harness by extending dsh-subagent-tools and including the two required provider patches.

Who it is for — When you need to control cwd per subagent call, this plugin is suitable. If you need subagent tools but not cwd control, you can use dsh-subagent-tools.

Watch out — Sandbox test passed: installed successfully in a new profile and registered by harness. Requires manual application of the two provider patches for foreground and background paths. Web sessions also require running the preset adapter.

The verdict — I would install it because it fixes cwd being silently ignored in the background in-process path, but you must manually apply the patches.

What it is — dsh-messager tracks session states in the DSH web app and sends alerts for interactions, task completions, and errors via system toasts, browser notifications, and services including Feishu, WeCom, Discord, DingTalk, and Telegram.

Who it is for — When you are handling interactive tasks or multi-turn dialogues in the DeepSeek Harness web interface and require alerts for pending interactions or completed tasks, this plugin fits your needs. If you only interact with DeepSeek Harness through the command line without the web UI, you do not need this plugin.

Watch out — Browser notifications require manual permission on first load, and the channel will degrade silently if refused. It passed sandbox testing by installing into a new profile and registering with Harness. No obvious issues found.

The verdict — I would install it because it provides notifications without needing to monitor the session list dots, but only if you set up at least one third-party messaging service.

What it is — Converts coding activities into RPG by awarding XP for turns, tool calls, and todos.

Who it is for — When using DeepSeek Harness to handle agent tasks, this plugin automatically scores your turns, tool calls, and todos. If you are already using another progress tracking tool, you can skip it because it turns your work into a game with XP.

Watch out — No obvious issues were found. Sandbox test confirmed successful registration into the profile. No runtime dependencies.

The verdict — I would install it because it treats coding as the game itself through event-driven scoring, but only if you want automated progress tracking.

What it is — The forge-tcm plugin provides tcm_diagnose and tcm_herb_pair tools for eight-axes TCM diagnosis and herb-pair retrieval.

Who it is for — When using DeepSeek Harness to analyze traditional Chinese medicine texts, you can use the tcm_diagnose tool for eight-axes diagnosis. If you lack a formula_db.json dataset, it is not suitable for installation because the herb-pair tool requires it.

Watch out — Herb-pair retrieval requires a private formula_db.json dataset not distributed with the repo, so FORGE_DATA must be manually configured. Sandbox testing was not performed. No obvious issues found.

The verdict — If I needed TCM diagnosis capabilities, I would install it because it is a subdirectory plugin with zero npm dependencies.

What it is — BM25 keyword-based memory recall plugin, supporting memory query, list folding and deep preview.

Who it is for — When your DeepSeek Harness agent needs to recall experiences using BM25 algorithm in specific tasks. If you prefer to implement custom memory management from scratch, you can build it yourself.

Watch out — Requires preparing your own memory data files to use. No obvious issues found.

The verdict — I would install it because it offers folded memory indexing.

What it is — Provide verification gates for symbolic math simplification, logic proofs, regex validation, E-prover FOL theorem proving, state-machine checks, and code repair, backed by Go-compiled binaries with prebuilt Windows executables.

Who it is for — If you use DeepSeek Harness agent for tasks requiring symbolic math simplification or logic proofs, this plugin provides verified results. If your work only involves text generation without external tool execution, this plugin is irrelevant.

Watch out — This plugin executes shell commands and requires manually allowing the build script when installing from source. Static checks indicate it will execute shell commands. No obvious issues found.

The verdict — I would install it because it provides code repair suggestions based on compiled binaries, using the prebuilt Windows version.

What it is — The evidence-first plugin provides evidence_audit tool that warns the agent when it claims success without a matching tool-execution record in recent context.

Who it is for — When DeepSeek Harness agents complete multi-step tasks, this plugin can verify that each success claim is backed by actual tool executions. Users who rely entirely on direct text responses do not need it, since their workflows do not require tool record verification.

Watch out — No sandbox testing has been conducted. Static analysis indicates shell command execution. No obvious issues detected.

The verdict — I would install it if the task requires verifiable tool usage steps for reliability, as it adds an accountability mechanism; the key is whether the workflow demands such records.

What it is — dsh-skill-pack plugin provides workflow skills support for DeepSeek Harness.

Who it is for — This pack is suitable when performing triage tasks. If you only use the minimal preset for simple conversations, you need not install this plugin.

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

The verdict — I would install it because it declares no @deepseek-ai dependency and the profile injects the official packages at boot time.

What it is — Provides a desktop launcher for DeepSeek Harness Web UI on Windows, including shortcuts, system tray, and PWA app window support.

Who it is for — If you use DeepSeek Harness on Windows and want desktop shortcuts and system tray support for the Web UI. If your main environment is not Windows or you prefer browser-native workflows without extra system integration, you can skip this plugin.

Watch out — The plugin executes shell commands for launch and tray functionality. Sandbox test passed: installed in new profile and registered by harness. No obvious issues found.

The verdict — I would install it because it achieves desktop-like experience using only official mechanisms with no extra dependencies.

What it is — A resident plugin that displays real-time DeepSeek API account balance in the DeepSeek Harness composer dock statistics bar, with 5-minute auto-refresh and manual refresh button.

Who it is for — Users performing chat or code generation tasks with DeepSeek Harness who need to monitor API balance are suitable to install this plugin. Those who do not involve DeepSeek API balance monitoring can choose other plugins.

Watch out — Sandbox test passed: successfully installed in a new profile and registered by harness. No obvious issues found.

The verdict — I would install it because it provides real-time API balance visibility, but only if you have set up the DEEPSEEK_API_KEY credential and are willing to install from source code.

What it is — Scans the skills visible to the current session, ranks them by relevance to the recent conversation, and recommends the next skill to load.

Who it is for — Agents working across multiple skill domains who need fast skill recommendation based on conversation. Users who only focus on a single skill domain won't benefit from it.

Watch out — Sandbox test passed: installed in new profile and registered by harness. The plugin only reads session text and skill catalog with no network or write operations. No obvious issues found.

The verdict — I would install it because it provides a deterministic cross-domain skill matching signal, but the premise is using a supported DeepSeek Harness version.

What it is — dsh-qrcode plugin provides offline QR code and barcode generation.

Who it is for — In DeepSeek Harness offline sessions where you need to encode links or text into scannable QR or barcodes, install it. If your project uses networked models and doesn't need local rendering, another angle is that the plugin's pure local advantage isn't prominent in online scenarios.

Watch out — Sandbox test passed: the plugin installed in a fresh profile and registered by harness. Static check indicates it may execute shell commands, but no other obvious issues were found.

The verdict — I will install it because the sandbox test passed and it supports multiple output formats, but the premise is that tool injection is configured and offline code generation is required.

What it is — Provides remote access to DeepSeek Harness from phone or anywhere.

Who it is for — Users who need to access DeepSeek Harness from their phone or remote locations. Users who only use it locally on the same device can skip it.

Watch out — Sandbox test passed after installation in a fresh profile and registration. First launch may auto-download cloudflared requiring internet; self-signed certificate prompts browser to proceed. No obvious pitfalls found.

The verdict — I wouldn't install it because it executes shell commands, though the premise is that the sandbox test passed.

What it is — SGME bridges multi-agent shared long-term memory via HTTP.

Who it is for — It suits users who need to share long-term memory across multiple AI agents. If you want to avoid the plugin executing shell commands, then you do not need to install it.

Watch out — Sandbox test passed: it installs in a fresh profile and is registered by Harness into the profile. The plugin executes shell commands.

The verdict — I would install it because it supports proactive care signals.

What it is — dsh-file-drop supports attaching non-image files to the composer via drag or click, with the desktop shell resolving the real Finder path instantly.

Who it is for — When using the DeepSeek Harness desktop shell, users needing to insert non-image file paths directly into agent drafts are suitable to install this plugin. Users who can rely on browser upload without needing desktop shell instant path resolution don't need to install it.

Watch out — Sandbox testing passed: installed in a fresh profile and registered successfully. No obvious issues found.

The verdict — I would install it because it resolves the real Finder path instantly in the desktop shell.

What it is — For the DeepSeek Harness plugin list page, add a toggle switch: support enabling/disabling plugins, take effect immediately without restart, and provide grouping, filtering, bulk operations, backup, and undo.

Who it is for — If you use DSH web profile and need to quickly switch individual plugin status from the list without restarting the service each time, this plugin fits. If you rely only on predefined plugin lists or do not need toggle controls, use the native features directly.

Watch out — Manual approval of the build script is required for source installation. Sandbox test passed in a fresh profile with harness registration. No obvious runtime issues found.

The verdict — If dynamic plugin list switching is needed in web profile, this plugin offers convenience; otherwise, no additional burden is required.

What it is — dsh-plugin-updater checks for updates on installed plugins, backs up configuration files, and supports rollback when updates fail.

Who it is for — If you manage plugins in a DeepSeek Harness web profile, you can install it to check for updates and support rollback. If you do not need update management for plugins, you can choose not to install it.

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

The verdict — I would not install it because there are no sandbox test results.

What it is — dsh-plugin-center is a DeepSeek Harness plugin center that allows discovery and management of community plugins from the Settings panel.

Who it is for — If you need to browse and manage DSH plugins from the Settings sidebar in DeepSeek Harness web profile, use dsh-plugin-center. Users who only add plugins via command line can skip it.

Watch out — Installation requires manual repository cloning and link-based integration to the profile with no one-click command. No risks were found in static checks. Sandbox testing has not been performed.

The verdict — I would not install it because it depends on manual source linking and has not been released to npm, requiring users to maintain the repository themselves.

What it is — A plugin that provides LLM request/response inspection, reasoning-effort tuning and external-think export.

Who it is for — When needing to adjust reasoningEffort in DeepSeek Harness for complex tasks, this plugin applies. Users not requiring reasoning block capture won't install.

Watch out — Installation from source requires manually allowing the build script. Sandbox test passed, installed in new profile and registered by Harness.

The verdict — I would install it because it supports traffic and bundle analysis, but only if the user understands that it will execute shell commands.

What it is — Provides a configurable Pomodoro timer with mini panel and notifications for DeepSeek Harness Web.

Who it is for — If you need to manage focus and break cycles in the DeepSeek Harness web profile, this plugin is suitable. If you are using a headless profile, you don't need it because it doesn't provide an interface.

Watch out — Sandbox test passed: installed in a fresh profile and registered into DeepSeek Harness. No obvious issues found.

The verdict — I would install it because it handles cross-session recovery.

What it is — The dsh-session-export plugin enables exporting the current session to portable schema-versioned Markdown and JSON files via the session_export tool and slash commands.

Who it is for — When using DeepSeek Harness in the Web UI to perform tasks, I would install it because it supports slash commands and cross-platform safe filenames. Not suitable for users in headless profile because they do not use slash commands.

Watch out — Sandbox testing passed on fresh profile. The plugin will execute shell commands but no other runtime issues found. Exported files may contain sensitive data and should be reviewed before sharing.

The verdict — I would install it if session export for archiving or diff is needed, because it decouples the export format from DSH internal implementation and has zero runtime dependencies.

What it is — dsh-webbridge bridges Kimi WebBridge to provide eleven webbridge_* model tools for real browser control.

Who it is for — When a model task needs to preserve login sessions and open tabs, this plugin enables agents to operate the user's real browser. For model tasks that only handle pure text or code simulation, this plugin is unnecessary.

Watch out — You must install the Kimi WebBridge daemon and browser extension separately. Sandbox testing passed with registration in a fresh profile. No obvious issues found.

The verdict — I would install it because it allows models to control real browsers with preserved state, but only if Kimi WebBridge is properly configured.