AI Agent Hub
Back to plugins
⚙️

dsh-subagent-cwd

Workflow Updated 2026.08.21

Run the following command in DeepSeek Harness:

dsh plugin install lynx-gt/dsh-subagent-cwd

Paste the following prompt into your AI chat to install this plugin:

To install this plugin in DeepSeek Harness, run the command: dsh plugin install lynx-gt/dsh-subagent-cwd. Source code: https://github.com/lynx-gt/dsh-subagent-cwd

About this plugin

In DeepSeek Harness (dsh), the default subagent delegation tools lack support for specifying a working directory (cwd) per call, which limits the flexibility of agents operating in different environments or directories. The dsh-subagent-cwd plugin addresses this issue by enhancing the subagent tools with per-call cwd control, implemented through two essential patches that ensure the working directory works correctly in both foreground (one-shot) and background (continuable) paths.

The core capability of this plugin is to allow users to specify an independent working directory for each subagent invocation, enabling environment isolation and resource access. It retains all the features from dsh-subagent-tools, including per-call overrides for model, provider, persona, and toolFilter, as well as @preset: references. This means you can enhance control over the subagent's runtime environment without disrupting existing workflows.

If you are a dsh user who frequently needs subagents to execute tasks in specific directories, or if you build complex workflows involving multi-agent collaboration, this plugin is ideal for you. It is particularly useful for scenarios such as testing different environments, accessing project-specific files, or performing multi-stage tasks. Be aware that under router-style presets, custom personas may not be preserved, so consider using message-layer injection or standard presets.

Overall, dsh-subagent-cwd fills a gap in the DeepSeek Harness ecosystem by providing a practical extension for working directory control, making agent workflows more flexible and reliable.

Use Cases

  • Assign isolated working directories for subagents in complex workflows.
  • Isolate resource access during multi-agent collaboration.
  • Switch agent working directories during testing or deployment.

Best For

  • DeepSeek Harness users seeking enhanced subagent working directory control.
  • Developers building multi-agent collaboration systems.
  • Testers who need to simulate agent behaviors in different directory environments.