dsh-neutrino

DeepSeek Harness (dsh) plugin, published as @dsh-external/dsh-neutrino, that lets an AI assistant drive STUDIO NEUTRINO, described in the README as a free Japanese AI singing-voice synthesis engine: given a character, lyrics and melody, it writes a score, drives the engine and returns a rendered WAV file. The free NEUTRINO engine and voicebanks must be obtained separately; the plugin provides official download guidance and can still be used for score tools without an engine installed.

• Registers seven `_dsh_external_dsh_neutrino_*` tools: status, list_voicebanks, install_voicebank, create_score, render_score, synthesize_song, list_outputs.
• Turns lyrics and melody in SMN note format (pitch:duration:lyrics) into MusicXML scores; supports dotted durations, exact ticks, rests and ties.
• synthesize_song is one-shot: it writes the score and renders it, returning a WAV path.
• install_voicebank imports a voicebank from an official zip and auto-detects the directory layout.
• Engine parameters include model, supportModel, styleShift (-55..+55), transpose, numThreads, samplingRate, bitDepth, useGPU, phrase and skipTiming.
• Config options: neutrinoDir (auto-detect), defaultModel (first voicebank), scoreDir (default ~/.dsh-neutrino/scores).
• Without an engine or voicebank, status/list/install tools return official download guidance; only actual rendering requires the engine.
• npm test runs 28 node:test cases covering SMN parsing, MusicXML, voicebanks and the render pipeline; built lib/ is committed.