dsh-migrate

Switching AI coding agents rarely fails because of the new tool—it fails because of everything the old tool left behind: months of conversation transcripts, memory files, tool-call logs scattered across a dozen config directories. dsh-migrate is built to close that gap. It reads the on-disk artifacts left by Claude Code, Codex CLI, Gemini CLI, Cursor, Kimi CLI, ChatGPT Web exports, Cline / Roo Code, OpenCode, Aider, Hermes Agent, and OpenClaw, then writes them as native DeepSeek Harness JSONL session logs. Imported sessions show up in the sidebar grouped by project, render their full event stream in the Trajectory view, and can be fork-style resumed as if DSH had recorded them from the start.

Two workflows are provided: a /migrate slash command that runs inside DSH itself, and a standalone CLI that scans the whole machine, previews everything with a dry run, and only writes after you confirm. Every adapter is intentionally tolerant—malformed lines and unknown row types are skipped rather than fatal, missing fields degrade gracefully (an unknown model is still imported, an unknown working directory lands in a _no-cwd project), and source agent directories are always opened read-only. Writes use append-only semantics with fresh session IDs, so an import can never shadow a live session, and original timestamps and provider / model metadata are preserved for provenance.

If you are moving to DeepSeek Harness and do not want to start from a blank context, dsh-migrate is the bridge. One scan collects your conversation history and memory notes from up to eleven upstream agents and turns them into a searchable, inspectable, resumable archive that plugs directly into the DSH session store.