v0.2.0 — model management

Adds five model-management tools (rapid_mlx_serving/cached/pull/remove/health) and a /rapid-mlx overview command, so the agent can see and manage rapid-mlx models without leaving the session. Served facts come from HTTP /v1/models (deduped); disk/pull/remove go through the harness subprocess seam to the rapid-mlx CLI. New peer deps: @deepseek-ai/dsh-tools, dsh-commands, dsh-subprocess. Verified end to end against a live 0.12.15 server + real CLI; 40/40 unit tests.

v0.1.0 — first npm release

First published release: npm i @raullenchai/dsh-provider / dsh plugin add @raullenchai/dsh-provider. Native Rapid-MLX provider for DeepSeek Harness — dsh reads model facts (context window, reasoning parser, capabilities) from the server instead of a hand-written settings.yaml. Verified against dsh 0.1.0-rc.8 (LlmAdapter contract byte-identical to rc.7; 31/31 unit tests + runtime smoke).