dsh-tool-monitor

You start a background service inside DeepSeek Harness and everything works for a while—until the log quietly fills with errors the Agent will only learn about when the process finally dies. Native DSH Jobs notify on exit, which is simply too late for long-running daemons. dsh-tool-monitor closes that gap: it attaches to an already-running Bash or PowerShell Job, scans the output stream line by line against a regex you supply, and the moment a match fires, a session event is delivered to the owning Agent. The original service keeps running as a single instance; nothing gets re-spawned.

The plugin extends the official LocalJobRegistry while reusing DSH's existing ownership isolation, capacity limits, and cleanup semantics. The job_list, job_output, and job_kill tools behave exactly as before. The Monitor is itself a regular Job with an independent lifecycle—killing it leaves the target service untouched, and when the target ends, the Monitor auto-terminates and flushes any trailing partial line. Jobs without a Monitor stay on the stock read path with zero extra polling overhead.

If you routinely launch API gateways, databases, build daemons, or any long-lived service with run_in_background inside DSH and want the Agent to react the instant a keyword, stack trace, or timeout marker appears in the log—rather than discovering a wall of errors after the process has already exited—this plugin is built for that workflow.