Introduction

When setting up a coding session in DSH, common demands are quite specific: you want a Codex-style preset, but you don’t want prompts, tools, context, and request tiers all hardcoded. DSH’s plugin architecture provides room for expansion, and the community plugin directory can also accommodate such scenarios.

shuind/dsh-codex-harness is a DSH plugin designed for this direction: it provides a coding mode for DSH with customizable prompts and Codex tools. Below is an introduction to its features, installation methods, and common configurations.

Positioning

@shuind/dsh-codex-harness is maintained by shuind and is licensed under MIT. Its core positioning is: to provide a coding mode in DSH that features editable prompts, configurable model behavior, and a set of Codex tools.

It is not an independent coding tool, but rather a preset and tool extension within the DSH plugin system. The prerequisite for using it is that you are already using a Web-based Agent session in DSH and wish to integrate a Codex-style workflow into the current session.

Core Features

Below are the verified main capabilities:

  • Uses a streamlined Codex prompt with minor constraints added to GPT.
  • View, edit, or restore the full prompt in Settings → Plugins → Plugin Config → Codex Harness.
  • Provides exec_command, write_stdin, apply_patch, and update_plan.
  • Supports Fast requests, GPT images, thinking intensity, and context capacity.
  • Displays the requested model, model response, and context compression status along with elapsed time in real-time.
  • Prioritizes native apply_patch, hosted web_search, and remote compression for Responses requests, falling back to the DSH implementation on failure.
  • Once Fast is enabled in the model menu for Codex mode, it sends service_tier: "priority" to Responses requests.
  • Clicking the context usage indicator next to the input bar allows setting a range of 1K–1000K tokens; this value is used for the next request and automatic compression.
  • Can be used with ChatGPT subscription models in conjunction with dsh-codex-connect.

Installation and Activation

First, install the plugin, then restart the Web interface, and finally switch the session to Codex mode.

  1. Install the plugin.
dsh plugin --profile web add @shuind/dsh-codex-harness@0.2.1

  1. Restart the Web interface.

  2. Select Codex Mode in Agent Preset and create a session.

Typical Usage

Configure the Model

First, set up the model connection: configure the provider, endpoint, API key, and GPT model in DSH’s Models that support OpenAI Responses.

The configuration example is as follows:

api: openai-responses
baseURL: https://your-responses-endpoint.example.com
apiKeyEnv: OPENAI_API_KEY

If you wish to use ChatGPT subscription models, you can also use it in conjunction with dsh-codex-connect.

Edit Prompts

Open:

Settings → Plugins → Plugin Config → Codex Harness

The editor will display the full prompt actually injected by this plugin. You can view, edit, or restore the default template; it will take effect starting from the next request after saving.

It is worth noting that identity, model, working directory, tool descriptions, and DSH runtime context are dynamically provided by the respective plugins and are not hardcoded into the editable template.

Enable Fast

Enable Fast in the model menu for Codex mode.

Once enabled, the plugin will send:

"service_tier": "priority"

Set Context Capacity

Clicking the context usage indicator next to the input bar allows you to set the context capacity, ranging from 1K to 1000K tokens.

The next request and automatic compression will use the value you set.

Applicable Scenarios and Notes

This plugin is suitable for the following scenarios:

  • You are using an Agent session in the DSH Web interface.
  • You want to use a Codex-style coding preset.
  • You need to view and modify the full prompt actually injected.
  • You need to use apply_patch, hosted web_search, remote compression, Fast requests, or context capacity control in Responses requests.
  • You wish to extend the coding toolchain through the DSH plugin mechanism.

Precautions before use: The plugin runs with the permissions of the current dsh process; you should check the source code and license before installing. The current license for this plugin is MIT.

Links

  • Directory page: https://www.skillhub.cn/plugins/shuind/dsh-codex-harness
  • GitHub: https://github.com/shuind/dsh-codex-harness