← ClaudeAtlas

setuplisted

Fetch the speech engine and model, so the installed plugin can actually talk
TraxData313/claude-voice · ★ 0 · AI & Automation · score 70
Install: claude install-skill TraxData313/claude-voice
Install the parts a plugin download cannot carry: the speech engine, the model, and Python if this machine has none. Run once, after installing the plugin. **Before running it, tell the user what it will do and wait for them to agree.** This fetches about 3 GB. Nobody should discover that from their connection. ```bash powershell -NoProfile -ExecutionPolicy Bypass -File "${CLAUDE_PLUGIN_ROOT}/setup.ps1" $ARGUMENTS ``` `--whatif` maps to the script's own `-WhatIf`: it prints every step it would take and writes nothing. Offer that first to anyone who hesitates. ## What to say while it runs Ten minutes, nearly all of it download — 0.66 GB of engine and 2.4 GB of model, both resuming where they stopped if the connection drops. No administrator rights at any point. It needs **Windows and an NVIDIA card**: both engine builds are CUDA builds and there is no CPU one, so on other hardware this fails at model load rather than at the door. If the user is on AMD, Intel graphics or a Mac, say so before it downloads anything rather than after. ## Afterwards The user ends up with a second, unnamespaced `/voice` command as well as this plugin's `/claude-voice:voice`. That is expected, not a fault: `setup.ps1` performs the full install, and the plugin's own copy of the command is separate. Both drive the same engine, and neither interferes with the other. Say so if they ask, rather than suggesting something went wrong. Then hand over: - `/claude-voice:voice on` starts the engine. The