AI Agent Hub
Back to plugins
🖥️

dsh-recall-unread

Client Updated 2026.08.27

Run the following command in DeepSeek Harness:

dsh plugin install hg1048596-pixel/dsh-recall-unread

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

Install the plugin in DeepSeek Harness by running `dsh plugin install hg1048596-pixel/dsh-recall-unread`, or get the source from https://github.com/hg1048596-pixel/dsh-recall-unread for installation.

About this plugin

In DeepSeek Harness, text messages sent via steering while the model is running appear as pending bubbles at the end of the conversation, but the official UI offers no way to recall them. If you accidentally send something you haven't fully thought through, the only option is to wait for the model to read it—no undo, no second chance. dsh-recall-unread fills exactly this gap.

The plugin adds an "unread messages" strip above the input box, listing all sent but not yet claimed steering messages in real time. You can recall a single message with one click, or use "recall all" when several are pending. The strip strictly reflects items still in the Agent inbox; as soon as the model starts reading a message, it leaves the strip automatically, preserving the "only unread can be recalled" semantics. Recalls call the official session RPC and behave just like the queue dock's delete action, with clear feedback on failure and no disruption to the session.

The plugin is zero-config and ready to use out of the box. Once installed as a static plugin, it loads with DSH on every startup and can be toggled from the plugin marketplace or management page. It is ideal for power users who frequently interject during long model tasks and need a safety net for accidental sends, as well as developers and researchers who want finer control over conversational input and wish to avoid polluting the model with unwanted messages.

Use Cases

  • Quickly recall accidentally sent steering messages during long model runs.
  • Bulk recall multiple pending messages to avoid interfering with the next response.
  • Manage unread messages via the strip above the input box before the model reads them.

Best For

  • DeepSeek Harness power users who frequently interject while the model is running.
  • Developers who need a safety net for accidental sends to keep conversations clean.
  • Researchers who want fine-grained input control and avoid polluting the model with unwanted messages.