AI Agent Hub
Back to plugins
⚙️

dsh-subagent-codex-plus

Workflow Updated 2026.09.03

Run the following command in DeepSeek Harness:

dsh plugin install february2015/dsh-subagent-codex-plus

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

Run dsh plugin install february2015/dsh-subagent-codex-plus inside DeepSeek Harness to install it (source: https://github.com/february2015/dsh-subagent-codex-plus).

About this plugin

The official dsh-subagent-codex plugin handles one-shot Codex delegation, but every interaction still routes through a dsh model in between, hiding execution internals and blocking multi-turn conversation. dsh-subagent-codex-plus layers a true-gateway direct-connection mode on top: once bound, your input goes straight to a persistent Codex thread with dsh acting as a pure byte relay, while the original delegation path remains available and switchable within the same session.

Key capabilities include binding a dsh conversation 1-to-1 to a durable Codex thread (auto-reconnect after restart), a visible message queue you can edit, reorder, or force-insert while Codex is busy, near real-time streaming of reasoning summaries and tool-call events with heartbeat timers for long-running commands, and image/attachment passthrough to Codex with vision fallback handled by the gateway system. Delegation and gateway modes coexist, so a single conversation can hold both model-triggered one-shot runs and a user-attached direct session.

Built for developers who already use Codex inside DeepSeek Harness and want a long-lived, inspectable working session rather than a fire-and-forget tool call. If you need to follow intermediate reasoning, iterate rapidly, or simply want dsh to step out of the way as a clean pipe, this plugin fills that gap.

Use Cases

  • Bind a dsh session to a persistent Codex thread and iterate across many turns
  • Watch reasoning summaries, tool calls, and status events stream in near real time
  • Edit, reorder, or force-insert follow-up messages while Codex is still working

Best For

  • Developers who invoke Codex heavily inside DeepSeek Harness and want it driving a long session
  • Engineers who need to inspect intermediate reasoning and tool execution, not just the final answer
  • Teams that want dsh to act as a transparent relay with Codex fully owning the conversation flow