dsh-doctor

DeepSeek Harness ships without a built-in doctor command. When DSH crashes, fails to boot, or a plugin corrupts the profile, your only real option has been killing the process and restarting, hoping it sticks this time. dsh-doctor turns that trial-and-error ritual into a predictable, fully logged, and completely reversible engineering workflow.

Its pipeline covers four stages. Diagnosis runs a read-only sweep across nine categories, pinpointing exactly what broke and why. Repair works in three escalating safety tiers (safe, deps, full), progressing from creating missing directories and fixing config placeholders, to rebuilding dependencies with pnpm, to terminating leftover processes, while recording an undo step for every reversible mutation. Rollback replays those steps in strict LIFO order with a single command. On top of that, dsh-doctor registers as a native Cordis service, so it can perform dynamic diagnostics, react to plugin lifecycle events, and raise alerts even while DSH is still running.

If DSH crashes keep derailing your workflow, or you manage a multi-plugin stack and are afraid to touch configs, rebuild dependencies, or hand-edit files on a production profile, dsh-doctor is built for you. You do not need to decode every stack trace, and you do not need to take DSH offline. Every repair is traceable, every rollback is one command, and the tool works out of the box on Windows, macOS, Linux, and fish.