Плагины

Менеджер навыков для веб-интерфейса dsh: пункт в сайдбаре и панель со списком навыков, видимых текущей сессии (разделены на глобальные / локальные группы), с переключателями вкл/выкл для каждого навыка, созданием / импортом / редактированием / удалением. Отключение навыка убирает его из контекста модели.

Плагин DSH: автоматически открывает браузер по умолчанию, как только `dsh web` готов обслуживать запросы.

Веб-плагин DeepSeek Harness (dsh): виджет баланса API и расходов за сегодня в нижней части сайдбара. Маршрут /api/dsh-usage + клиентский React-бандл, сборка не требуется.

Лёгкая аналитика использования для DeepSeek Harness — тренды токенов, тепловые карты активности, разбивка по моделям и экспорт.

Набор операционных инструментов с приоритетом доказательств: шесть инструментов только для чтения (каталог возможностей, поэтапные планы workflow, ридер упакованных skills, ограниченный поиск по локальной памяти, аудит релизов репозитория, чек-лист релиза) и пять упакованных skills — только планирование и аудит, никогда удалённая запись.

Врата проверки на реальных вычислениях: упрощение математических выражений, логические доказательства, проверка regex, FOL через E-prover, проверки конечных автоматов и исправление кода на основе бинарников, скомпилированных на Go, с готовыми исполняемыми файлами для Windows.

Операции ADB с устройствами и тестовым стендом для DSH: обнаружение устройств, структурированный logcat (фоновая потоковая передача), установка apk, передача файлов pull/push и снимки производительности dumpsys.

Сборка проверяемых пакетов базы знаний (SQLite FTS5) из md/txt/docx/pdf с детерминированным поиском и чтением исходного текста.

Изображения, прикреплённые в composer, распознаются в текст vision-моделью, совместимой с OpenAI, прежде чем попасть к текстовым моделям DeepSeek.

Память в рамках проекта между сессиями: снимок PROJECT.md внедряется в первое сообщение пользователя, инструмент memory_remember и авто-рефлексия после задач ReAct; у каждого проекта свой файл памяти.

Desktop-оболочка в трее Windows для DSH: автоматически устанавливает exe из GitHub Releases, создаёт ярлык на рабочем столе и добавляет инструмент desktop_launch для запуска из чата.

A conversation-style layer that rephrases trade-off warnings: when a shortcut is cheap but has a catch, the reply may say the fast road is the rival shop and the slow one keeps its rules. Ships a rewrite engine that turns any text into a fixed-format dialogue with five independent knobs (act count, catchphrase density, ending turn, seed, preset), a slider for four on/off intensity levels, and a /bushaoxin command that reports a joke "heartburn index" from the session tool-call log. Every rewrite ends with a real technical conclusion that cannot be disabled. Zero dependencies, no build step, engine is pure functions.

A no-limit Texas holdem table where you are the only human. Eight seats: you plus seven LLM agents written on AI-founder personas (Altman, Dario Amodei, Musk, Liang Wenfeng, Jensen Huang, Lei Jun, Zhang Yiming). Each agent bets through the currently selected host model and falls back to its own heuristic style when no model is available; every agent only ever sees its own hole cards, because the prompt is built from that seat alone. A hand ends by revealing the winner hole cards, with a flip animation replayed on every community card, and a rail keeps a live action-and-table-talk timeline. LAN mode (off by default, single-machine play is unaffected) lets another install on the same network join this very table: the host shows a six-digit room code, UDP broadcast discovers nearby tables, and a joiner randomly takes over one bot seat, with the seat handed back to its character on leave or after twenty seconds of silence. The room code is never put on the wire and every snapshot is generated for one viewer seat, so nobody, not even the host, can read another player hole cards. Fork of BubblePtr/dsh-holdem (MIT), credited in the README and in the UI.

A governed registry for MCP connectors (stdio and streamable-http): four-level fail-closed authorization, mcpServers JSON import that rejects malformed args, credential redaction at every output boundary, an agent bridge that returns an honest review verdict instead of a fake approval, and a /mcp-reg operator console plus a settings page.

Windows toast notifications when a DeepSeek Harness task finishes: the session title, the outcome, and a summary of the final reply, sent once when the agent really stops rather than once per round of a running goal.

Turn-handoff reminders in the DSH Web UI with a modal or top banner, six browser-synthesized chimes, configurable auto-dismiss, and a sidebar speaker toggle with an unread session list.

A Multitask agent preset where the main conversation only plans, briefs subagents and summarizes results, while file reads, commands, search and builds run inside subagent contexts; four worker kinds, an approval gate for dangerous operations, and zero changes to DSH source.

Deep-reading workflow for books: import EPUB/Markdown/TXT/PDF into chapters, read one chapter per subagent so the source text never enters the main session, keep \[n]-anchored chapter memories for cross-session continuation, and export notes or an EPUB.

Builds a Markdown knowledge base from your course materials, then drills you until every knowledge point reaches your target accuracy; each answer must include your reasoning.

Converts Office and PDF files in the DSH workspace to Markdown locally — no network, no API quota — before the model reads them. Writes each .md next to its source so only a file path enters the context, and returns byte size, line count, token estimate and a heading outline, for a single file or a whole directory.

Give the agent a talking face: seven tools that drive a self-hosted Mynah realtime digital human — speak (queued or interrupting), play a gesture clip, interrupt, list live sessions and channels, check health, add text to its knowledge base.

Reads animated GIF and WebP as animations: a pasted one keeps playing in the UI while the model gets a labelled sheet of frames sampled across it, and the read_animation tool samples a file on demand.

Local art pipeline for DSH: drives the locally installed SD WebUI and ComfyUI backends through ten image_* tools and an Art Studio sidebar panel with an embedded ComfyUI canvas, post-processes with the Python shipped with those installs (rembg cutout, Real-ESRGAN upscale, seamless tiling), and writes Godot-ready assets with res:// paths once a project root is set.

Chat with the DeepSeek web app (chat.deepseek.com) in a native DSH panel — a real browser drives the site on your existing web login, with streaming replies, deep-thinking and smart-search toggles, and image or document attachments. Conversations can be recovered from the web side, exported as Markdown, and distilled or replayed into a harness session, and the agent reaches the same web chat through five tools (dschat_status, dschat_send, dschat_recover, dschat_import, dschat_transfer).