Introduction¶

When advancing a project step-by-step using DSH, a common issue is that requirements are scattered across chat logs. The previous sentence might say “add an export,” while three hours later, another conversation round mentions “change the export to time-based.” No one merges these items, leaving the user to rely on scrolling through history messages to recall them.

The dsh-req-workbench introduced below is the plugin for this stage. It is maintained by wowswows and organizes explicit functional requirements raised in the conversation into a checklist. Subsequent discussions update existing items rather than appending new messages, categorized under “Memory.”

One-sentence positioning: This is a community plugin for DSH that maintains a “Requirements” page, allowing the AI to actively maintain a structured list of functional requirements during the conversation. It is refactored based on Songran241/dsh-req-workbench and is an unofficial adaptation; please refer to NOTICE.md and LICENSE in the repository for authorship attribution.

What problems does this plugin solve¶

Let’s first look at the two approaches it replaces.

• Relying on manual memory: Requirements are scattered across multiple rounds of conversation, making them hard to trace over time.
• Asking the AI to repeat the requirements every time: The same requirement appears repeatedly across different rounds, making the text grow longer and more fragmented.

dsh-req-workbench’s approach is to provide a unified “Requirements” page: The AI maintains it by referencing the same checklist, and users can also edit it manually without overwriting each other’s changes. Requirements are separated from the conversation, so checking requirements does not require scrolling through chat logs.

Core Features¶

The following capabilities are taken directly from the “What it can do” section of the plugin README without any extension.

• Automatically organize functional requirements that are explicitly stated by the user and accurately understood by the AI.
• Merge and update when discussing the same function repeatedly, writing short titles and descriptions that are easy for anyone to understand.
• Ideas that are uncertain are kept separately for confirmation rather than being assumed as user requirements.
• Details like original wording and subtasks are collapsed by default; requirements can be manually edited with concurrent modification protection.
• Isolated by session; the installation program does not bring in other people’s chat or requirement data, nor does it automatically sync across machines.
• It uses the AI currently conversing with you to call maintenance tools, without launching an additional model. The model still needs to actually invoke the tool to write, and it cannot guarantee that any model will correctly extract information every time.

The last two points are worth confirming before integration: the list will not automatically sync across sessions or machines; writing relies on the model actually invoking the tool, and the prompts thrown are just requests, so accuracy is not guaranteed every time.

Take a Look Before Installing¶

The README explicitly validates only DSH 0.2.1-rc.5 and its Web interface. Before installing, execute the command below to confirm the version first. If the version is different, confirm compatibility before proceeding; do not force an exemption.

dsh --version

Installation requires no downloading of source code, Node development dependencies, or compilation; the plugin code is already placed in the lib directory.

If the original plugin is already installed, the package with the same name will replace the original version, and both versions cannot be enabled simultaneously. The installation affects all sessions in the target profile, not just the window currently being chatted in.

Method 1: Send the Package to Another Computer¶

This method does not require the target machine to have Git or a compilation environment; simply copy the .tgz file over.

1. Copy dsh-community-dsh-req-workbench-0.2.1-dsh021.1.tgz to the target computer and place it in a location where it won’t be accidentally deleted later.
2. Open PowerShell or a terminal and execute the command below, changing the path to the actual location:

dsh plugin --profile web add 'C:\DSH-plugins\dsh-community-dsh-req-workbench-0.2.1-dsh021.1.tgz'

3. If the application prompts for a restart, restart the corresponding DSH service/application; then refresh the page, open the project session, and click “Requirements” at the top.

“web” must be the profile actually used by that DSH instance. If you are using a different configuration name, replace that parameter. Do not unzip the .tgz file, nor directly import the source code ZIP as a plugin installation. Keep the .tgz locally, as it may still be referenced when reinstalling dependencies.

Method 2: Install from GitHub¶

The repository address is https://github.com/wowswows/dsh-req-workbench. It is recommended to install the published version directly without compilation:

dsh plugin --profile web add github:wowswows/dsh-req-workbench#v0.2.1-dsh021.1

You can also install from a GitHub Release attachment. The command is as follows; note that this is the installation address for the Release of this repository:

dsh plugin --profile web add 'https://github.com/wowswows/dsh-req-workbench/releases/download/v0.2.1-dsh021.1/dsh-community-dsh-req-workbench-0.2.1-dsh021.1.tgz'

This project does not have a prepare/prepack installation script. GitHub installation uses the lib already committed in the repository and does not require the user to rebuild on another machine.

Uninstall¶

The uninstall command; note that it operates on the package name @dsh-community/dsh-req-workbench:

dsh plugin --profile web remove '@dsh-community/dsh-req-workbench'

Follow the prompts to refresh or restart the target DSH. Uninstalling does not equal deleting saved data; do not casually clean the DSH data directory.

Typical Usage¶

After installation and enabling, you do not need to open a specific page separately; simply proceed with the project session as usual. The process is as follows.

1. State requirements directly in the conversation, for example, “The export function needs to support filtering by time range.”
2. The AI attempts to call the maintenance tool to record this requirement in the list; if the write is successful, a corresponding item will appear in the “Requirements” page at the top of the session.
3. When discussing the same requirement later with colloquial additions, for example, “change the default time range for export to the last 7 days,” the AI updates the existing item rather than creating a new one.
4. When content is uncertain, the AI treats it as pending confirmation rather than writing it as a confirmed requirement.
5. Manually edit items directly in the “Requirements” page.

The “Requirements” page is at the same level as “Conversation / Trace / Context / Formation” and is located above the session.

Applicable Scenarios and Notes¶

Suitable for these types of people.

• Developers or project recorders using DSH to gradually finalize the scope of features in multi-round conversations, who need a checklist that does not rely on manual organization.
• Users who want to keep the conversation area clean by stripping requirements out of the chat logs.
• People who need to pass requirement lists between teams can copy and pass them manually after editing, as the plugin itself does not perform cross-machine synchronization.

A few limitations are explained in the README; please confirm before integration.

• Only validates the Web interface of DSH 0.2.1-rc.5; compatibility with other versions must be confirmed by the user.
• Runs with the permissions of the current DSH process; it is recommended to review the source code and license before installing.
• Writing relies on the model actually invoking the tool, and it cannot guarantee that any model will correctly extract information every time.
• There is no cross-session or cross-machine synchronization; for cross-machine use, you must pass the list manually or install it on each machine.
• This is an unofficial refactored version based on the original repository, retaining the original author’s Copyright (c) 2026 liangsongran, with the license being MIT; the Zod license used on the browser side can be found in THIRD_PARTY_NOTICES.md.

Conclusion¶

A one-sentence review: dsh-req-workbench does not aim for smarter conversation, but rather provides a unified destination for requirements. It compresses “who mentioned it and what changed” into a checklist that can be manually verified, reducing the cost of flipping through chat logs after the fact.

The division of responsibilities for organizing items and manual editing is clear: the AI is responsible for merging and maintaining, while the final judgment is still confirmed by humans. The plugin itself is an editable workbench, not an automatic decision-maker.

View entry points:

• Community Directory Page: https://www.skillhub.cn/plugins/wowswows/dsh-req-workbench
• Project Repository: https://github.com/wowswows/dsh-req-workbench