Every DeepSeek Harness plugin

What it is — The dsh-memory plugin injects long-term memory across sessions.

Who it is for — When you need to remember user preferences, project facts, session summaries or reusable knowledge across sessions. If you prefer to independently handle everything within each single session you can skip it.

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

The verdict — I would install it because it supports half-automatic memory writing.

What it is — Integrates git CLI into DeepSeek Harness for Conventional Commits enforcement, changelog generation, PR summaries and branch pushes.

Who it is for — When developing code with DeepSeek Harness and needing to enforce Conventional Commits along with auto PR summaries. Not suitable for users who rely on platform APIs or tokens.

Watch out — Installed from GitHub source code. Successfully tested in a fresh profile and registered by harness. No obvious issues found.

The verdict — I would install it because it enables git workflow without needing platform APIs or tokens.

What it is — dsh-prometheus plugin provides Prometheus metrics monitoring for DeepSeek Harness with a Grafana dashboard.

Who it is for — When developing AI agent workflows with DeepSeek Harness, installing dsh-prometheus allows tracking LLM request rates and tool call outcomes. Users not requiring Grafana visualization can skip installation, as the plugin does not change DeepSeek Harness core behavior.

Watch out — Sandbox testing completed successfully with installation in a fresh profile and registration in Harness. Defaults to loopback-only endpoint configuration. No obvious issues detected.

The verdict — I would not install it, as deploying the Grafana stack adds unnecessary setup if Prometheus metrics are not needed for monitoring.

What it is — dsh-kubernetes provides DeepSeek Harness with a safe Kubernetes capability including workload inspection, bounded logs and events, and failure diagnosis.

Who it is for — When debugging Kubernetes deployments using DeepSeek Harness Agent lacking workload management and bounded log capabilities, install it. If you only handle local files without cluster operations, don't need to install.

Watch out — Sandbox test passed: installed in new profile and registered by harness. Requires shell command execution and API key or token. No obvious issues found.

The verdict — I would install it because it converts Kubernetes into structured Agent-friendly tools with read-first security and approval gates.

What it is — Jira Cloud issue-tracker integration exposing eight model tools for DeepSeek Harness.

Who it is for — When managing issue trackers with DeepSeek Harness, this plugin exposes eight model tools. If you do not use the Jira Cloud REST API v3 provider, you do not need this plugin.

Watch out — Sandbox test passed. Registered successfully in a new profile. No obvious issues found.

The verdict — I would install it because it supports dry-run for writes and writes are opt-in.

What it is — dsh-ci is a generic CI/CD plugin that lets DeepSeek Harness agents inspect GitHub Actions pipelines, extract failure evidence, and safely rerun or cancel them.

Who it is for — When DeepSeek Harness agents need to check the status of GitHub Actions runs, extract source error evidence, and perform safe reruns or cancels within their loops, this plugin is suitable. If a project does not involve external CI/CD pipelines or does not require failure evidence extraction from logs, this plugin is not needed.

Watch out — Manual approval is required to run the build script from source. Sandbox testing showed it installs as an ordinary dependency without becoming a profile layer in harness. It executes shell commands. No other issues were observed.

The verdict — I would install it because it supports safe reruns and cancels gated behind write approval, but only if the work requires managing GitHub Actions pipelines inside DeepSeek Harness rather than purely local tasks.

What it is — Bridge image attachments to view_image tool for text-only models

Who it is for — Developers using text-only models to review code with uploaded screenshots can install it. If the model already declares support for image input then no need for this plugin.

Watch out — Sandbox test passed in new profile with successful registration. Configuration requires specifying vision backend baseURL, apiKey, and model. No obvious issues found.

The verdict — I would install it because it passed sandbox testing in a new profile and enables text models to handle images via the view_image tool.

What it is — Integrate Weixin bots into DeepSeek Harness through QR code authorization.

Who it is for — Suitable for users who want to send messages to DeepSeek Harness via Weixin private chat. Devices without access to *.weixin.qq.com do not need to install this plugin.

Watch out — Sandbox test passed: installed in a new profile and registered into profile by harness. The plugin will execute shell commands. No obvious pitfalls found.

The verdict — I would install it because it runs in the DeepSeek Harness Host process.

What it is — lazeword is a DeepSeek Harness plugin for offline vocabulary training. It supports spaced repetition, six quiz types, and games. It can open as a sidebar panel and run as a standalone HTML file.

Who it is for — When you are using a DeepSeek Harness plugin and need a tool to assist with English vocabulary learning, especially combining spaced repetition to reinforce memory, it can provide support.

It is not suitable for learners who do not need game assistance because it includes multiple interactive games such as space matching and memory flip.

Watch out — It was successfully tested in a fresh profile and registered into the profile by harness. It will execute shell commands. No other obvious issues were found.

The verdict — I would install it because it supports lazy large-screen mode, making it convenient to use on large screens while maintaining focus.

What it is — mimo-vision registers describe_image tool in DeepSeek Harness to send images to mimo-v2.5 and return text descriptions for text-only models.

Who it is for — Users of text-only models like deepseek-v4-flash who need image description capabilities should install it. Users with native vision models do not require this bridge.

Watch out — Manual build permission is needed when installing from source. No obvious issues were found.

The verdict — I will install it because it provides seamless image-to-text bridging for text-only agents without additional vision infrastructure.

What it is — Integrates 44 LiuHe MCP tools and Rust tree-sitter parse daemon into DeepSeek Harness.

Who it is for — Agents using DeepSeek Harness for code symbol search and batch editing tasks should install this plugin. Users not requiring MCP tool integration can skip it.

Watch out — Successfully tested in a fresh profile with harness registration. May execute shell commands.

The verdict — I would install it because it supports transactional edits with undo journal.

What it is — This plugin replicates DeepSeek Chat interface attachments in DeepSeek Harness, including paperclip button, drag-and-drop, DS-style file cards, and text extraction for model reading.

Who it is for — When using DeepSeek models for document analysis or summary tasks, install this plugin to enable file cards and full-text injection. Users who only do pure text chats or visual image processing do not need it.

Watch out — Tested successfully registered in fresh profile under sandbox. Depends on pdfjs-dist, mammoth, xlsx pure JS libraries with no native compilation. PDF without text layer requires OCR environment variables configured.

The verdict — I would install it because it enables the model to read full uploaded document content, which is necessary for full-text analysis tasks.

What it is — Provides the workspace_snapshot tool to DeepSeek Harness, reporting bounded repository-relative Git status.

Who it is for — Users who need the model to report Git branch and path status are suitable for installing this plugin with DeepSeek Harness. If the agent needs to manage branches or perform commits, users should use dedicated Git workflow plugins.

Watch out — Sandbox test shows it can be installed, but the build script is blocked by pnpm for github:txy-ucas/dsh-workspace-snapshot, causing harness to fail completing registration. Users need to manually allow the build to install properly. No other obvious issues found.

The verdict — I would install it because it provides a read-only structured Git status report that lets the model check before any changes, but source install requires manually allowing the build script first.

What it is — Add session tags in DeepSeek Harness web settings and support searching sessions by tag.

Who it is for — Users who need to add and manage session tags would find this plugin useful. If you primarily use DeepSeek Harness via the command line interface, you would not need to install this plugin.

Watch out — Installation from source requires manually allowing the build script to run. Sandbox testing passed with no obvious issues.

The verdict — I would install this plugin because it provides storage independent from the session logs and query pipeline.

What it is — It loads client plugins into DeepSeek Harness Web UI in real time via a hot directory without restarts.

Who it is for — Users developing real-time updating UI panels for DeepSeek Harness Web are suitable for installing this plugin. Users who only use pre-published plugins without needing dynamic updates are not suitable because this plugin targets the hot update scenario.

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

The verdict — I would install it because it allows updating plugins without restarts, but only in local trusted environments.

What it is — Deploy dsh-egress-guard in DeepSeek Harness as a runtime security gate for tool calls. It denies calls to hosts outside allowlist, redacts credentials from canonical values, and appends decisions to JSONL audit log.

Who it is for — You would install it when developing agent workflows that require egress control for tool calls and audit logging. You do not need it for users whose agents never perform network operations outside a controlled environment, as the plugin operates in monitor mode by default.

Watch out — Sandbox testing passed with successful registration in a new profile. No other runtime issues were observed. Configuration can be adjusted via cordis.patch.yml to switch modes.

The verdict — I would not install it unless runtime security and auditing for tool calls are required, because it ships in monitor mode without breaking existing setups and provides full decision logs.

What it is — Converts turns, tool calls, and errors into wild encounters for catching.

Who it is for — If you encounter errors or use tools in your work, these events will generate wild encounters that you can catch to build a team of six. If you run multiple DSH processes simultaneously, the state file will be overwritten by the last writer.

Watch out — Installation from source requires manually allowing the build script. Sandbox test passed with registration in a new profile. The dex is shared across all profiles in one DSH home.

The verdict — I would install it because it has zero token cost and reacts to work events, but under the premise of accepting shared dex and manual encounter checks.

What it is — Converts cordis_define dynamic packages built at runtime into permanent plugins that survive restarts.

Who it is for — Users who want agent-built plugins to survive restarts. Users who prefer managing plugins through the regular development workflow can skip it.

Watch out — Granted wishes are ordinary code that runs at every boot with the full DeepSeek Harness context. They are parsed with node --check without execution for validation. All grants are listable and reversible.

The verdict — I would install it because it requires no pnpm, no network, and no build authorization, and granted code is just readable files.

What it is — awesome-dsh is a DeepSeek Harness plugin that registers a skill for querying dsh-plugin topic plugins on demand.

Who it is for — It is suitable when you need to find and add new plugins using DeepSeek Harness. Users who have already mastered the plugins in the dsh-plugin topic can expand their capabilities from another angle.

Watch out — Manual build allowance is required when installing from GitHub source. It passed sandbox testing in a fresh profile and was registered with harness.

The verdict — I would install it because it provides a DSH plugin finding skill, but the premise is using the source installation method.

What it is — dsh-tool-bandit-search plugin provides a Thompson sampling based search tool that learns to choose between fast and thorough strategies and improves from usage.

Who it is for — When using DeepSeek Harness for web search, if you want the search tool to learn the best strategy from actual calls, it is suitable to install this plugin. If you prefer to use other search tools, you do not need to install this plugin because it is a replacement for the standard web_search tool.

Watch out — Sandbox test passed after installation and registration in a new profile. No obvious issues found. The bandit state is in-memory and resets on every restart.

The verdict — I would install it because it improves from real usage, which fixed heuristics cannot achieve.

What it is — Integrates long-task orchestration loop supporting Codex planning, DSH execution and Kimi supplementation.

Who it is for — When using DeepSeek Harness to handle long tasks requiring multiple rounds of planning and verification, this plugin is suitable. Not suitable for users who only use local scripts to complete tasks.

Watch out — Sandbox test passed in new profile with successful registration. No obvious issues found.

The verdict — I would install it because it provides a multi-model collaborative long-task framework and the test passed.

What it is — dsh-swarm-router provides sub-agent matrix routing to dispatch heterogeneous tasks to the most suitable models.

Who it is for — Users handling heterogeneous tasks that require model optimization are suitable for installing this plugin. If users do not care about token consumption statistics or feedback ranking, they do not need this plugin.

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

The verdict — I would install it because it enables feedback-driven routing adjustment, but only if task types are diverse.