If Voyager helps you, feel free to share it on X, Reddit, YouTube, Threads, etc.
dsh-stall-guard
A watchdog plugin for DeepSeek Harness sessions. It tracks each session's last activity and in-flight operations, and only nudges or terminates when the agent is truly silent — never while work is in progress. Every event is logged, and recovery is guided in steps.
Plugin facts
Screenshots
About this plugin
DeepSeek Harness watchdog plugin: detects truly stalled agent turns (never killing in-progress tasks — in-flight operations are exempt), nudges/terminates only on real silence, records every event to JSONL with a loopback status route
Install this DSH plugin
Copy the command below and run it in your DeepSeek Harness terminal.
dsh plugin add github:akira399/dsh-stall-guard ⚠️ Plugins run with the permissions of your dsh process and may execute code at install time. Review the source repository and its license before installing. Pin the commit hash for reproducible installs. Plugin safety →
Related guides
Learn how to choose, install, and use sessions & messages plugins.
About this plugin: dsh-stall-guard
What is dsh-stall-guard?
dsh-stall-guard is a sessions & messages plugin maintained by akira399. A watchdog plugin for DeepSeek Harness sessions. It tracks each session's last activity and in-flight operations, and only nudges or terminates when the agent is truly silent — never while work is in progress. Every event is logged, and recovery is guided in steps.
How do I install dsh-stall-guard?
Run dsh plugin add github:akira399/dsh-stall-guard in your DeepSeek Harness terminal. The dsh CLI resolves the plugin from GitHub and installs it into your active profile. For reproducible installs, pin a commit hash: dsh plugin add github:akira399/dsh-stall-guard#commit.
Is dsh-stall-guard free?
dsh-stall-guard is a community open-source project released under the MIT license. You can inspect its source and install it for free.
Related plugins
A task-aware reasoning-mode router for DeepSeek Harness. Each task is placed on one continuous react-to-spec axis, the router picks from three measured behavior bands before the first model request, and persona plus first-turn tool injection are supported with agent-visible tuning.
A context-insight dashboard — see what the model's context window is made of and how it evolves.