dsh-chime-sound
用于 dsh-tui 的通知音：为对话完成和用户提问提供不同提示音，支持终端聚焦与未聚焦状态，9 组可选音效，主音量可按 5% 步进在 0-150% 之间调节，设置部分，支持 zh/en
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add dsh-chime-sound@0.3.3Chime Sound 以 dsh-chime-sound 发布，当前版本为 0.3.3。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
点击下方 Badge 复制 Markdown，粘贴到 README 即可。
验证 package.json 声明的 GitHub 仓库，即可管理这个公开页面。认领后，Hub 会优先安排当前版本的安全扫描，并在通过后公开展示结果。
认领这个 Plugin →English · 中文
Built for dsh-TUI.
Notification sounds for dsh-tui (originally named dsh-notice-sound, renamed on 2026-09-07).
ask_user_question (session/event tool/call — a nudge to come back and answer) and a reply finishing (session/event turn/end; agent-loop emits this once per turn, while assistant/message fires on every tool step and is deliberately not used). Both distinguish a focused from an unfocused terminal.←/→ steps 5%; holding the key repeats, because dsh-tui's input layer expands the auto-repeat into per-step events. The new level is auditioned immediately. Above 100% the samples are 16-bit saturated (peak ≈ 105%, clipping only for very short stretches). The gain is applied to the WAV samples inside the plugin and never touches system volume.scripts/gen-sounds.mjs (WAV, 16-bit / 48 kHz); g5–g9 are Freesound CC0 1.0 material, decoded and then trimmed, faded and loudness-normalised to the same format. Sources and conversion parameters are recorded in assets/sounds/CREDITS.md (CC0 needs no attribution and may be redistributed with this plugin)./lang selected.Sound is played through Win32 APIs (koffi → winmm.dll PlaySoundW, with focus detection via user32.dll), so this plugin is Windows-only (Windows Terminal and classic conhost). On Linux and macOS it degrades silently to a no-op — no sound, no boot failure — and logs a Win32-only warning. Do not expect notification sounds elsewhere.
Option 1 — dsh CLI (recommended):
dsh plugin --profile <profile> add dsh-chime-sound
(Or from the dsh plugin market / GitHub: dsh plugin add github:VviLliAm-qwq/dsh-chime.)
Option 2 — manual:
~/.dsh/profiles/dsh-tui/node_modules/dsh-chime-sound/"dsh-chime-sound" to dsh.profile.bundles in ~/.dsh/profiles/dsh-tui/package.json (the package declares dsh.bundle.patch, so it mounts itself at boot)/restart)Compatibility: dsh-tui 0.10.x (the ctx.tuiSettingsSections and session/event seams, both soft-probed); dsh 0.1.2-rc.1+; Node ^22.19 || >=24; pure ESM. Every configuration key has a default, and a missing configuration degrades to “nothing happens”.
The settings live in dsh-tui's “Notice Sounds” section (or the chime: namespace of ~/.dsh/settings.yaml):
chime:
doneFocus: g2 # reply finished, terminal focused: g1-g9 / off
doneBlur: g2 # reply finished, terminal unfocused
askFocus: g2 # question asked, terminal focused
askBlur: g2 # question asked, terminal unfocused
volume: "50" # 0-150, 5% steps, default "100"
session/event and not the message observerdsh-tui's mediated tuiMessageObserver.subscribe() is unusable here: the API requires the activation to hold a verified Component identity through host admission, and a plugin mounted by a bundle patch never goes through admission — its subscriptions are refused (COMPONENT_NOT_ADMITTED) and no sound ever plays.
This plugin therefore listens on the session/event cordis bus published by @deepseek-ai/dsh-session (the same channel the official dsh-working-activity extension uses), which needs no Component identity:
tool/call with name === 'ask_user_question' → the question sound (askFocus / askBlur)turn/end → the completion sound (doneFocus / doneBlur); subagent sessions are ignoredGetForegroundWindow handle equals GetConsoleWindow, or the window class matches (CASCADIA_HOSTING_WINDOW_CLASS = Windows Terminal, ConsoleWindowClass = classic conhost). Under ConPTY GetConsoleWindow returns a hidden PseudoConsoleWindow that is never foreground, so the earlier handle-only and class-only readings were both flawed; the check is now a pair of criteria with a corrected class read.dsh-plugin.json no longer declares the unused messages.observe capability.Restart dsh-tui after installing.
v* tags drive .github/workflows/release.yml, which publishes to npm through trusted publishing (OIDC) — no token is stored in the repository.MIT — see LICENSE. Sound assets are licensed and attributed in assets/sounds/CREDITS.md.