dsh-kimi-bridge

In a multi-agent DeepSeek Harness workflow, letting Moonshot Kimi act as a second coding pass should be a one-liner, but in practice it means spawning kimi -p by hand, parsing the JSON stream, polling for status, juggling parallel calls, and resuming settled sessions. dsh-kimi-bridge collapses all of that scaffolding into four first-class tools (call_kimi, kimi_status, kimi_abort, kimi_steer), making Kimi a native citizen of the dsh session with async and blocking modes, status polling, abort, and same-directory session continuation available out of the box.

The plugin ships a dedicated Kimi tab in the conversation pane that renders the full Agent Loop as a live waterfall of messages, tool invocations with arguments and outputs, and turn separators. Toggle between Activity and Text views to switch between the structured loop and the raw streamed transcript. All state changes flow through dsh session projection events, so the tab updates in real time and survives a page refresh with complete history replay. The default reviewOnly mode locks the tool allowlist to read-only operations, while operators can explicitly opt into unrestricted execution when needed.

If you already orchestrate multi-agent collaboration in dsh and want Moonshot Kimi to serve as a parallel coder, reviewer, or second-brain pass, this plugin is the cleanest one-step integration—eliminating hand-written process management and stream-parsing code entirely.