Plugins

Turns new mail from every dsh-email account into dated to-dos: drops codes, alerts and newsletters without a model call, extracts actions and deadlines with the DSH default model (rules fallback), skips CC-only mail, dedupes across mailboxes and flags time clashes. Items go to a built-in to-do panel with a month calendar (adapted from dsh-sidebar-todo, same data file), optionally mirrored to dsh-calendar; a refresh icon, /inbox-sync and an inbox_planner tool trigger a check.

Turns a video, article, PDF, or text into a narrated video, with an in-app workbench for settings and project history, per-scene images, TTS, metadata, and ordered API fallback.

Primary and multi-level backup model failover router for DSH, with cross-provider reasoning-effort clamping, exponential cooldown, per-turn recovery, and a settings panel.

Configurable A-share, Hong Kong and US quote strip under the DSH composer: mini K-line per card, intraday, 5-minute, daily, weekly and monthly charts, five-level order book, A-share turnover and limit prices, session-aware polling and a Sina fallback source. Quotes come from public Tencent and Eastmoney endpoints; its HTTP routes have no DSH access check.

Video analysis panel in the DeepSeek Harness sidebar. Paste a YouTube or Bilibili link to fetch its subtitles and generate a summary, chapters, a mind map and citation-backed Q&A over the transcript, with an embedded player whose timestamps jump to that moment and keep playing. Videos without subtitles fall back to local Whisper / SenseVoice transcription (audio stays on the machine); silent videos fall back to frame OCR with macOS Vision, or to vision-model frame descriptions. Per-video analysis caching, a 20-item history, and two agent tools: video_digest and video. A Bilibili SESSDATA cookie, if you enter one, is stored in ~/.dsh/video-credentials.json (mode 0600), transcripts are sent to your current model, and its HTTP routes have no DSH access check.

Remote-control your local DeepSeek Harness from a phone through your own zero-knowledge relay: phone-first page, QR pairing, end-to-end encryption, approvals, full desktop UI fallback.

Lightweight cross-session reflection: at the end of each turn it writes 3-5 reusable facts (tool chain, file paths, config items, pitfalls) as a card in .dsh/reflections/{date}.md, and on a new session it injects the last 3 cards at the end of the system prompt. Extraction uses the real ctx.llm.stream; when no model route is available it falls back to a heuristic built only from the session log and labels the source honestly on the card. No timers, no quarantine, no approval flow; cards are plain Markdown you can edit or delete.

Defines named roles that each pair a persona with a bound model and fallback chain, and exposes one delegation tool per role, so subagents are dispatched by role name instead of a provider/model string.

Native DeepSeek Harness surfaces for markets and finance: a full page with 11 live quotes (SSE, SZSE, ChiNext, Hang Seng, Nasdaq, S&P 500, Dow, COMEX gold, dollar index, BTC, ETH) and sparklines, a market-cap heatmap of up to 96 A-share industry boards with a constituent drawer, movers, an eastmoney news feed, 25 plain-language articles and a 293-term glossary; the complete OpenFinLens terminal (MIT, vendored, served by the Host) embedded as the default tab; and a floating quote card with a searchable instrument picker and a boss key that also hides the optional always-on-top desktop window. The Host fetches upstream server-side with a 30 s snapshot cache, per-part timeouts and stale fallback: no API key, no browser CORS workaround.

Research-style problem solving in four gated stages: record sources by retrieval tier with local before web, turn each shortfall of existing tools into a gap that needs evidence, score candidates against weighted metrics with a deterministic engine, fuse or improve the weak ones, then iterate on measured deltas. Also organizes a directory into human-facing and machine-facing trees.

Port of MiMo Code's infinite-context memory: as the window fills it asks the model in the background (the current provider route unless writerProvider/writerModel is set) to write a structured checkpoint and a project MEMORY.md, then, ahead of DSH's native compaction, replaces the visible window with a rebuilt dump of tasks, checkpoint, recent user messages, memory and notes, falling back to a MiMo-style compaction. Memory files live under $DSH_HOME/mimo-context (per session, per project, plus a read-only global one); it adds mimo_* memory, history, checkpoint and rebuild tools and /mimo-* commands, with dream/distill off by default. Needs DSH 0.2.2-rc.1 or a later 0.2.x.

Translates Chinese text rendered by other DSH plugins into English as it appears - plugin cards, settings pages, dropdown menus, plugin view tabs and same-origin iframe panels - while leaving the chat transcript, tool output and anything you type untouched. Uses the shipped translator service (Bing or Google, no key needed) and falls back to the default model.

Mobile gateway plugin that pairs DSH Mobile App with a self-hosted DeepSeek Harness host over an authenticated WebSocket. Fork of Clarklevis1995/dsh-plugin-mobile-gateway (same package name, install only one) that adds an optional encrypted address book deployed to your own Cloudflare account as a Worker with wrangler, an experimental WebRTC P2P transport, and a public-DoH fallback for its endpoint check.

Adds one button to the sidebar foot that opens the desktop DSH Web GUI — in the system browser on a phone, in a new tab on a desktop. The target address is read from plugin config, DSH_DESKTOP_URL, or ~/.dsh/desktop-url.txt, and falls back to the relay origin published by dsh-remote-web-ui; the button renders inert with a hint when nothing is configured. It also adds a small gear button that saves the address into ~/.dsh/desktop-url.txt; on a phone with no separate launch address set, a click asks the local pairing service (/api/pair/issue) for a one-time pairing link and opens it; its HTTP routes (/api/desktop-link/…) have no DSH access check.

Adds a “why does this need escalation?” button to every DSH approval card: it opens a panel with the permission requested, the risk, what refusing means, the exact tool arguments and whether the named paths fall outside the session workspace; the card also keeps a one-line risk and refusal consequence visible under it, and the model’s original reason is split into labelled paragraphs instead of a run-on block.

Download only a time range of a video (yt-dlp segments; long videos fall back to HLS when the DASH URL cannot be seeked), verify the cut with ffprobe and sampled frame contact sheets, and serve the clip over a temporary LAN HTTP link with byte-range support. Exposes video_clip_download, video_clip_verify and file_share tools plus a video-clip skill; requires ffmpeg/ffprobe. Installs yt-dlp with pip into $DSH_HOME/cache/video-clip when it is missing; file_share serves a file or directory over unauthenticated HTTP bound to 0.0.0.0 (visible on the LAN) for 30 minutes by default.

Daily chronicle for DeepSeek Harness: collects session events, closes the previous calendar day after midnight into a Markdown week file (LLM scoring adds a daily summary; rule scoring does not), and extracts durable facts — decisions, conventions, pitfalls — that merge across days and can be superseded. 10 tools, a read-only sidebar panel, rule export, and a standalone Obsidian vault; zero third-party dependencies.

Folds older oversized tool results into spill files and replaces them with short, deterministic stubs at step boundaries, with no LLM call: the stub keeps the first 200 and last 80 characters plus the sequence number, tool name, line/char/size counts and the original file path, and each fold is a real surface replacement accompanied by compaction/prune shadow pricing so token counts stay honest. Triggers on context pressure at 25% of the model window (falling back to 32768 tokens when the window cannot be resolved) or on five accumulated candidates; the last three turns, error results and rich content are never folded. The original is recovered from the session log rather than the spill file, so /folds, /unfold <seq> and /unfold all keep working after spill cleanup; session_fold_recall lets the model fetch a folded result by sequence number or by keyword, and session_fold_diagnose reports the effective settings. /spill-prune removes orphaned spill directories, preview only unless --apply. A Settings page exposes 12 of the 13 options as draft-then-save fields; they are volatile, so saved changes apply without restarting the plugin.

Puts a domain bar under the session header — a domain groups skills and tools and toggles as one block, so one click loads or unloads every member. Hovering a chip lists its members (skills, then tools) in a height-capped panel that scrolls. The right end of the bar is a compaction-threshold counter (− 5 +, 100k tokens per step) that rewrites the live compaction engine policies, so the compaction trigger line changes on the next turn without restarting dsh; a degraded fallback turns the number amber and is reported in the snapshot. Settings maintains domains and members with search, skill/tool tabs, all/selected/unselected filters and a copyable AI grouping prompt. Tool removal uses tools.restrict plus a guard; hiding a skill re-registers it with modelInvocable false, so it leaves the model catalog while /name still works. Sessions of another preset are left untouched.

Export a DSH session tree (lead plus subagent fleet) to two Markdown files from one frozen snapshot: a readable narrative summary and a complete raw archive. Target directories are checked against the session file policy, and a blocked location falls back to the workspace exports folder.

Adds a live system dashboard with per-core CPU, memory, GPU, processes, disk usage and a STOP action for ending processes, plus per-process energy on macOS, exposed as a browser page, a GUI panel and a `systemInfo` service other plugins can inject. macOS is the main platform and Linux is best-effort. The dashboard page and its read-only JSON (process list with users, disks, logged-in users) are served with no login check; publicDashboard: false only refuses non-loopback peers. STOP accepts loopback requests only and sends SIGTERM, then SIGKILL after a grace period, to any process your account may signal except pid 1, the DSH host and its parent.

Draws an icon in front of every session row in the DSH sidebar, decided from the session itself: keyword rules on the title first, then the session content (title, first user message and tool names), then an optional one-shot model verdict, then a workspace fallback. 445 Ant Design outlined icons are inlined for offline use, and the rules, colours, keywords and row indent are editable in a settings page.

Read-only behavior audit for DSH sessions: flags unsafe actions in 8 categories (A–H) with rules for 14 subclasses (the 41-subclass taxonomy file is not shipped, so a built-in 14-subclass fallback is used), anchors every finding to a real event seq, and gives deep links plus the full tool-call / data-access chain of any assistant reply (parameters, results, files read and written, domains, sandbox and approval context). Every finding is drill-downable.

One-click restart for the DSH desktop app that keeps the window position, size, maximized state and display, with an optional resume of the sessions the restart interrupted (not only the one you had open). The entry is an icon at the right of the sidebar footer account row plus a /restart-desktop slash command; when the account row cannot be recognised it falls back to the sidebar.footer.action seat, so the button does not disappear. Windows desktop only.