English | 中文

dsh-kernel-kimi

DSH runs on one simple idea: everything is a plugin. Models, tools, subagents — plug them together however you like.

So we did exactly that: we turned Kimi Code into a DSH plugin. The kimi-cli tool surface you already know — ReadFile, WriteFile, StrReplaceFile, Glob, Grep, Shell, ReadMediaFile, SearchWeb, FetchURL, TaskList, TaskOutput, TaskStop, SetTodoList, AskUserQuestion, Agent, ExitPlanMode, EnterPlanMode — is now a set of native DSH tools. Same names, same schemas, same behavior.

The payoff is simple: use the kimi CLI natively inside DSH — no different from opening Kimi Code itself. Every model stays in the environment it knows best — main agent or subagent, it feels like coming home.

SearchWeb / FetchURL talk to the same Moonshot endpoints the Kimi CLI uses (api.kimi.com/coding/v1/search and /fetch) with the shared OAuth token. Distilled from kimi-cli 1.49.0.

We differentially verified against the real kimi CLI: the same task on the same directory produces identical results, item by item.

System prompt & subagents

lib/system-prompt.js carries the upstream Kimi Code CLI system.md (runtime placeholders adapted to DSH); apply() registers it as the agent's sole system-prompt section (complete: true + suppressRuntimeContext()).

lib/subagents.js ships the kernel's own subagent recipes — kimi-agent (coder), kimi-explore, kimi-plan — each = the full system prompt with the upstream roleAdditional block inserted. The mesh loads them and mounts this plugin on each child with a config.tools whitelist.

Install

Copy the package into your DSH profiles:

cp -r dsh-kernel-kimi ~/.dsh/profiles/node_modules/dsh-kernel-kimi

Then add a row to the kimi-kernel agent preset (inside the planning group, so the plan tools can reach planMode):

- id: kimi-surface
  name: dsh-kernel-kimi

Usage

Start a session on the kimi-kernel preset and pick the model kimi-kernel/k3-256k. Your main agent runs on the Kimi kernel with the full kimi tool surface.

License

MIT — see LICENSE.