AI Agent Hub
Back to plugins
⚙️

dsh-resume

Workflow Updated 2026.08.15

Run the following command in DeepSeek Harness:

dsh plugin install xiaohj233/dsh-resume

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

Run dsh plugin install xiaohj233/dsh-resume in the DeepSeek Harness terminal to install the plugin; source code is available at https://github.com/xiaohj233/dsh-resume.

About this plugin

Upstream DeepSeek Harness already provides session persistence and torn-log repair, yet two practical gaps persist: an interrupted turn cannot be continued with a plain empty submit, and the exact set of running sessions and subagents at process shutdown is never automatically restored on the next boot. dsh-resume exists to fill precisely those gaps.

When a turn is interrupted, the plugin folds the partial text and reasoning already received into session history, so pressing submit on an empty composer resumes the conversation rather than starting a fresh one. Before shutdown, every still-active top-level session and subagent is atomically recorded to ~/.dsh/resume-state.json; a crash or kill mid-restoration never loses remaining entries, and failed resumes are automatically queued for the next boot. All compatibility patches are idempotent, reversible, and version-adaptive, ensuring an upstream upgrade can never brick the agent loop.

This plugin is built for heavy DSH users who routinely hit network blips, rolling OS updates, or manual process kills — situations where manually re-pasting context to continue a long reasoning chain is tedious and error-prone. After a restart, unfinished inferences pick up exactly where they left off.

Use Cases

  • Auto-recover long reasoning chains after network blips or process kills
  • System updates restart DSH; unfinished sessions resume without re-pasting context
  • After a manual kill, atomic state writes ensure remaining sessions survive the restart

Best For

  • Heavy DSH users who routinely hit network blips or process crashes
  • Developers running long, complex multi-agent workflows
  • Production users who need high-availability session continuity