Plugins

A terminal workbench for dsh with streaming chat, tool execution, session management, provider management, and image input.

Composer code-quote collapse for the DSH Web UI: pasting a `path:line` header followed by code folds the input into one compact inline quote chip, and at send time the `agent/pre-step` hook injects the full pasted snapshot to the model as a separate context message. 0.5.x also folds sidebar file-selections that insert only a `path:lines` line (large selections), fetching the code from disk by session cwd, and preempts Ctrl+V paste interception.

语音输入插件 for Deepseek Harness

Agent-agnostic skill that turns task input files into reusable structural briefs and a per-task searchable index (OpenAI Codex, Claude Code, DeepSeek Harness)

科研收敛驾驶舱 (developer preview): hard-gated research convergence loop for DSH agents - 7 conv_* tools + host session projection + turn-tail card (conversation.chat.turnTail) + cockpit strip (conversation.input.dock) + settings section. 零 npm 依赖 (node builtins

Screenshot-to-input for DeepSeek Harness: composer camera button + global hotkey (Alt+A) + listener bound to the app lifecycle, configurable in settings. 截图自动粘贴到 DSH 输入框：相机按钮 + 全局快捷键 + 生命周期绑定 + 设置页配置。

A microphone button for DeepSeek Harness that writes browser speech recognition into the composer draft.

Audio Copilot for DeepSeek Harness: transcribe audio (ASR) and synthesize speech (TTS) — gives text-only agents ears and a voice. Windows-local SAPI TTS out of the box; OpenAI-compatible ASR/TTS endpoints configurable. Includes an in-composer voice-input

OpenCode Go plan usage monitor for DeepSeek Harness: real-time 5h-rolling / weekly / monthly quota in a status bar under the chat input

Auto-detect the current model vendor (7 built-in) and show its balance or quota in the composer input row, with custom endpoints for unsupported vendors.

Voice input (Web Speech API) and read-aloud (speechSynthesis) for the DeepSeek Harness web GUI — DSH 语音输入 + 回复朗读套件

Type normally; about two seconds after you pause, it rewrites your draft into a clearer, first-person version more likely to get a good answer, leaving pasted code, links and long text untouched, marking missing details as placeholders instead of guessing, and keeping multi-part questions numbered.

Composer GitHub picker: click the GitHub icon at the bottom-right of the input box to search the workspace repository's issues and pull requests via the gh CLI, and insert the reference as a URL or @owner/repo#number mention.

DeepSeek Harness plugin: a `deepseek-vision` route that declares image input and transcribes pasted images via the free Zhipu GLM vision models before delegating to the DeepSeek adapter.

Minimal ↑/↓ input history for the DSH web composer: press ↑/↓ to walk back through sent messages. Nothing else.

Vision-augmented DeepSeek adapter: a vision-capable model describes image input, then a text-only DeepSeek model reasons over the description.

Multi-provider LLM API balances under the chat input, with presets for DeepSeek, OpenRouter and SiliconFlow.

Voice for DeepSeek Harness（dsh） — speech-to-text input + read-aloud TTS for text-only DeepSeek, zero API key.

Browser Web Speech API voice input: zero server, zero keys, zero model downloads (Edge=Azure, Chrome=Google speech).

Read-only Ascend NPU fleet monitor for DeepSeek Harness: environments are read from envs.json, npu-smi is collected locally or over SSH with one connection per environment, and the sidebar shows each environment's idle/partial/busy occupancy, running processes, per-card and per-chip details, five-minute background polling, manual refresh, and a field-by-field configuration modal. It never starts or stops services or containers and never writes to the remote host.

Per-workspace DSH conversation status in the Windows system tray, plus native DeepSeek Harness notifications when a conversation needs approval, a choice or free-text input, completes, or fails. Windows 10/11 and DSH Desktop 0.2.3-rc.1+ only. The tray menu and notifications work only after a one-time desktop bridge install (user-level, reversible): it repoints the DSH shortcut at a hidden launcher that injects a hook into DSH's Electron main process through a short-lived loopback Node inspector; without the bridge the plugin only writes state.json. Idle conversations stay out of the tray by default; nine live options are editable in a Tray follower settings page.

A local mirrored A/B benchmark for context compression, not another compressor. It runs one fixed read-only scenario twice with the same model and temperature — compression off, then on — and reports input tokens, turns, tool-call success, fact coverage, answer similarity and a per-line diff, with JSON, a self-contained SVG and raw warnings. 18 node --test tests, zero dependencies. The first measured run: 63,029 → 6,927 locally estimated input tokens (−89.0%), 3 → 3 turns, facts 3/3 → 3/3, answer similarity 1.00 on one sample.

Computer-use for DSH on Windows: a session-scoped toggle in the composer that registers nine tools for screenshots, mouse clicks and drags, keyboard typing and shortcuts, window switching, UI Automation element lookup, window-to-account mapping, a batch prompt queue and download collection. Desktop automation input is native Win32 SendInput and defaults to human-like pacing - bezier cursor paths with overshoot correction, real VK keystrokes, log-normal typing with occasional typo corrections and a pause before each click - with realistic, fast and custom (0.2-8x) speed presets. Windows only. Off by default and switched on per session by the user; screenshots are sent to the model, and clicks on send/pay/delete-like controls are refused unless confirmed.