AI Agent Hub
Back to plugins
⚙️

dsh-launcher

Workflow Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install Nasbaye/dsh-launcher

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

Install via dsh plugin install Nasbaye/dsh-launcher in the DeepSeek Harness terminal; source code is available at https://github.com/Nasbaye/dsh-launcher

About this plugin

In the DeepSeek Harness web profile, every service restart kills every in-flight agent session outright, and a model output truncated at the token limit still requires you to type a follow-up by hand. dsh-launcher collapses those high-frequency pain points into a single click.

Four layers of capability work together: a real-time status dot beside the session header (green, yellow, or red, polled every 5 seconds) keeps service health at a glance; on confirmation a background worker runs dependency pre-checks, kills the old listener, waits for the port to free, relaunches node, and performs up to 90 seconds of health polling all logged and console-popup-free on Windows; before restart it detects unfinished sessions by status, inbox, and phase signals, then after the new process boots it sends a resume prompt to each one so agents pick up exactly where they left off; and whenever any top-level session is cut off by max-tokens, the plugin auto-sends a continuation with loop guards including a max of 3 consecutive attempts, a 5-second cooldown, and a counter reset on any non max-tokens turn.

It is built for developers and operators who debug the web frontend frequently, run long tasks that routinely hit token ceilings, and would rather not manually resume every chat after each restart. Once installed, restarting is one confirmation, resumption and continuation are handled for you, and your attention stays on actual development.

Use Cases

  • Restart the web service quickly during frequent frontend debugging while in-flight multi-agent sessions resume automatically
  • Auto-continue model output that was truncated at the token ceiling in long-running tasks without manual prompting
  • Batch-resume multiple agent conversations after a service restart without manually re-prompting each one

Best For

  • Developers who debug the DeepSeek Harness web frontend on a daily basis
  • Power users running long tasks, parallel agent workflows, and frequently hitting token ceilings
  • Ops engineers who need reliable service restarts without losing workflow context