Foreword

When developing agents with DSH, a common practice is to break tasks into fixed stages: first capturing ideas, then filtering, asking questions to refine, writing specifications, breaking down task tickets, and finally implementing and reviewing each ticket one by one. If each step corresponds to a skill, the user has to remember skill names and type /skill one by one in the input box—completing a full cycle involves remembering seven or eight commands, and if interrupted, one has to recall where they left off.

dsh-matt-pipeline solves this problem: it registers Matt Pocock’s set of skills workflow into DSH and provides a visual pipeline panel in the sidebar. A single click dispatches the corresponding skill to the current session.

What is this

dsh-matt-pipeline (Version 0.2.0, MIT License) is a DSH plugin maintained by sa998aaron, positioned as a “Matt Pocock Skills Workflow Pipeline.” It does two things:

  1. Registers the built-in matt skill set as DSH skills callable by both the model and the user;
  2. Adds a new “Pipeline” tab in dsh-better-sidebar, displaying the workflow stages as panels and supporting one-click dispatch.

It is worth noting that the built-in skills come from Matt Pocock’s mattpocock/skills repository. The copyright of the skill content belongs to him; this plugin is only responsible for packaging for DSH and adding the visual panel.

Core Features

The plugin consists of two parts:

Server-side part: Registers the built-in skill set as DSH skills, including matt-triage, matt-grill-me, matt-to-spec, matt-to-tickets, matt-tdd, matt-code-review, matt-research, matt-diagnosing-bugs, etc. Once registered, both the model and the user can call them by name.

Browser part: Adds a 🔀 Pipeline tab in dsh-better-sidebar, displaying the seven stages of the workflow:

💡 Idea → 🔍 Triage → 🔥 Grill → 📝 Spec → 🎫 Tickets → ⚡ Implement → ✅ Review

There is also a set of helper tools at the bottom of the panel: 🐛 Debug / 📚 Research / 🎓 Teach / 📦 Handoff / ❓ Ask Matt, which can be triggered at any time.

Clicking the Start button for a specific stage dispatches the official /<skill> command directly to the current session via the panel; if the API connection is unavailable, it falls back to the clipboard method. Skills can also be used directly without the panel: type / in the input box to find matt-* entries, or call them by name in the conversation.

Additionally, the build artifacts dist/ have been committed to the repository, so no local build step is required during installation.

Installation and Usage

The web profile uses the following command:

dsh plugin --profile web add https://github.com/sa998aaron/deepseek-harness-matt-plugin.git

The DSH Desktop GUI uses the desktop profile:

dsh plugin --profile desktop add https://github.com/sa998aaron/deepseek-harness-matt-plugin.git

After installation, restart DSH (or toggle the plugin switch in Settings → Plugins) and perform a hard refresh (Ctrl+F5).

If you want to install manually, first clone the repository to the node_modules directory of the corresponding profile (e.g., ~/.dsh/profiles/desktop/node_modules/), then add dependencies and bundle entries in <profile>/package.json:

"dependencies": { "dsh-matt-pipeline": "file:./node_modules/dsh-matt-pipeline" },
"dsh": { "profile": { "bundles": [ "...", "dsh-matt-pipeline" ] } }

Then restart DSH and perform a hard refresh.

One prerequisite to note: the pipeline panel is registered into dsh-better-sidebar, so to display the panel, the corresponding DSH web or desktop profile must first install dsh-better-sidebar. The skills themselves do not depend on it and can be called normally even without the sidebar plugin.

Typical Usage

  1. Open the sidebar’s + menu → 🔀 Pipeline.
  2. Click on a stage to view its description and expected outputs.
  3. Click Start , the panel sends /<skill> to the current session, and the agent executes according to that skill’s workflow.
  4. The helper tools at the bottom (debugging, research, handoff, etc.) can be triggered at any time.

It can also be done without the panel: type / in the input box, find the entries starting with matt-* from the completion list, or directly call the skill by name in the conversation.

Development

If you want to participate in development or build it yourself, the repository uses tsdown for building. The server-side artifact is ESM (dist/index.mjs), and the client-side artifact is CJS (dist/client.cjs):

pnpm install
pnpm build      # tsdown: server ESM (dist/index.mjs) + client CJS (dist/client.cjs)
pnpm dev        # watch mode

There is one point worth noting for plugin developers: ctx.betterSidebar only exists in the browser cordis context, so this kind of UI code must be placed in the client/ directory and cannot be placed in the server side. From package.json, we can see that the client injects dependencies dsh-better-sidebar and @deepseek-ai/dsh-client-connection, with the platform as web; runtime dependencies are yaml ^2.5.0, and peer dependencies are react ^18.2.0.

Suitable Scenarios and Notes

Suitable for DSH users who are already using or planning to adopt Matt Pocock’s phased workflow: if you have to manually input a string of skill commands every time, this panel can converge the process into a few clicks; if you only occasionally use one or two skills, installing the plugin and calling matt-* directly in the input box is sufficient.

A few notes:

  • The pipeline panel depends on dsh-better-sidebar; without it, the skills are still usable, but there is no visual panel.
  • Built-in skill content comes from the mattpocock/skills repository and is copyrighted by Matt Pocock.
  • Like all DSH plugins, the plugin runs with the permissions of the current dsh process. It is recommended to check the source code and license (MIT) before installing to ensure it is correct, then load it into your own profile.

Conclusion

The value of dsh-matt-pipeline lies in integrating a ready-made multi-stage skills workflow into DSH’s interface layer: skill registration is handled by the server, workflow display by the sidebar, and dispatching only requires a single click. See the directory page at https://www.skillhub.cn/plugins/sa998aaron/deepseek-harness-matt-plugin, and source code and documentation at https://github.com/sa998aaron/deepseek-harness-matt-plugin.