Plugins

Paste a link; the agent reads it. Xiaohongshu, Douyin, Bilibili, WeChat, articles, PDFs and audio become citable Markdown for a wiki or knowledge base. On 60 turns with vs without: finishes reading 22% -> 92% (+70 points), times out 68% -> 0%, answer quotes a real source fact 23% -> 82% (the A/B is exploratory: different time caps, no human correctness review). On 123 live links 80% came back usable (93% of the pages that answered); an unreachable link reports in 26 s. Codex, Claude Code, DSH.

Evidence-first attention supervision for DeepSeek Harness with C0–C3 policy, deduplicated Inbox alerts, quiet hours, and redacted outcome records.

Desktop notifications for turn completions, with per-outcome controls and keyword rules.

Runs Midscene on the models already configured in DeepSeek Harness instead of separate third-party API keys: hosts an OpenAI-compatible bridge on 127.0.0.1 (default port 8787, automatically skipping occupied ports) that proxies chat completions, streaming and non-streaming, to the DSH host model layer, so DeepSeek official routing, CodeBuddy, Codex and custom OpenAI/Anthropic endpoints all become available to Midscene agents while credentials stay inside the provider plugins and the bridge never touches them. A `/v1/midscene-env` endpoint generates the four MIDSCENE_MODEL_* environment lines with the model family inferred, and seven built-in skills — browser, desktop, Android, iOS and HarmonyOS automation plus a Vitest E2E skill and a Midscene knowledge reference — appear in the session skill catalog without `npx skills add`. Install from npm with `dsh plugin --profile web add @tnnevol/dsh-midscene`.

dsh 浏览器扩展（全端）的桌面端一半：把浏览器扩展连到 dsh，提供 browser_* 工具，并在需要时替你识别页面上的图片。这是它的全部配置项。

Agent Teams for DeepSeek Harness: hand work to a team of agents that carries it end to end while you set the direction and hold the gates. Agents claim Tasks, hand work to each other, and have someone other than the builder verify it.

Word alternatives and underscore-gated fill-ins in the composer. End a line with _ and it is filled; misspellings are flagged as you type. Routes through the model dsh is already configured with, so it needs no API key.

Third-party adapter that registers the upstream REA (Reverse Engineer Anything) MCP server into DeepSeek Harness as mcp__rea__* tools. 2.x provisions its own runtime (Node, the REA CLI, a JDK, Ghidra) under <DSH_HOME>/plugins/rea-bundle and launches REA through a shim, so a machine with nothing pre-installed works after one setup run; 1.x (branch v1, npm 1.0.6) is configuration-only and expects REA and Ghidra to be installed already. Analysis is upstream's: native binaries (Hopper/Ghidra/IDA), JavaScript/Electron apps, .NET assemblies, websites, Android APKs and adb devices, firmware, and process captures.

IM webhook and local notifications on turn completion, errors, or approval (Feishu/WeCom/DingTalk/Slack/Discord/custom).

Shared multi-agent task board (create/claim/transition/query) over a Cordis service key.

In-app plugin store for the DSH Web UI: browse, search, one-click install, compat badges.

Export the append-only session log as human-readable Markdown or HTML, grouped by trajectory source.

Cross-session long-term memory as a local SQLite store: dual scopes (user + per-git-origin project), hybrid FTS5 + local-vector retrieval fused with RRF, an evidence gate that forces an explicit sufficiency judgement, provenance audit chains, SM-2 spaced-repetition review, consolidation distillation and decay forgetting, and a nine-view "Memory Palace" sidebar panel. 22 narrow-parameter tools, no external service, no API key, no Docker.

Daily RSS/Atom news digest for DSH: built-in channels plus a curated awesome-rsshub-routes catalog and custom feeds (validated on save), aggregated into one "今日值得读" Markdown digest each day and injected into the system prompt; Web GUI settings card and sidebar reader.

Rewrites outgoing /chat/completions requests with an evolved chain-of-thought scaffold (nine-dimension genome, elitist selection with fragment crossover, one deterministic re-evaluation pass), scores answers against a fixed 20-objective baseline, and appends one line per rewrite to live-requests.jsonl in the plugin data root (default <cwd>/.aecot, overridable with AECOT_ROOT).

Native DSH browser integration for the DSH Plugin Store

Controls your real Chrome from DSH over a local WebSocket bridge: tab management, navigation, DOM snapshots, page JS evaluation, clicks, typing, screenshots, waits and dialog handling behind a Settings toggle.

Bottom info bar for the DSH Web UI: provider/model, live balance, peak/off-peak pricing with countdown, and real persisted per-session spend in a single line.

Remote DeepSeek Harness with full server-side API access (`settings.*` / `credentials.*` / `host.listDirectory`). Token-gated reverse proxy, per-device sessions, phone invite QR, fence self-check, optional approval / CIDR / idle timeout / local TLS, WebSocket/SSE.

Puts the Markdown documents, source files, images and video that already exist anywhere in the session workspace into the DSH sidebar, ranked by relevance to the current conversation. The default lexical mode matches your recent messages against title, summary and body locally with IDF weighting (BM25) — no model calls, no network, nothing leaves the machine. The list comes from scanning the workspace rather than from a recent-files list, so restarting DSH or opening a new session does not empty it. Results are organised into a three-tier reading order (primary / supporting / related) with a verifiable reason on every row — which field matched, which document cites this one — and one click opens an in-place preview: Markdown as rendered text, relative-path images actually drawn, source code with the host's own syntax highlighting, video streamed over HTTP Range. The same Context Pack is exposed to the agent as a knit_docs tool. Later versions added read evidence and lifecycle tracking (attribution is frozen at the moment of reading), pin / exclude (exclusion applies before retrieval, and is undoable), relation evidence (references / imports / tests, with line numbers) and an incremental index reused across restarts. Since v0.23 there is also an optional hybrid mode: it is off by default and only runs once you configure an Ollama or OpenAI-compatible embedding provider yourself. When it runs, what leaves the machine is your query text plus each candidate file's title and summary — a Markdown summary is an excerpt of the body's first paragraph (at most 60 characters), a code file's summary is its workspace-relative directory, and an image or video contributes its file name; full bodies and absolute paths are not sent, and a loopback provider keeps everything on this machine. Since v0.24 it also associates uncommitted changes: a read-only git status (fixed argv, no shell, whitelisted child-process environment) combined with the same reference / import / test / mention evidence shows which files a change is likely to drag along, and it is off by default. The benefit of semantic retrieval still awaits a real-provider run, so no recall improvement is claimed.

Reference logged-in Chrome/Edge tabs, authorized local HTML and running Windows apps through a searchable native sidebar or @. Open browser candidates and the webpage sidebar refresh when the extension connects, disconnects or changes tabs, including an unchanged search query. Windows application lists skip protected processes with unreadable identity metadata and preserve other running apps. Read scoped snapshots, batch actions, capture screenshots and use available WebMCP tools; supports authorized uploads, drag/drop and opt-in network metadata. Browser actions preserve failure evidence, accept an explicit post-action observation scope and can check one final visible/text/value/URL condition without replaying effects. Expandable context preserves model content; dockable annotations support a session-bound staging queue with edit/delete/restore, and append to chat drafts without sending. App control prefers a matching installed MCP or approved installation, then Cua Driver with text, hotkeys and paged controls; desktop reads present the window once per connection or when recovery is needed. Background input depends on the app. Requires Chrome or Edge 125+ with the companion MV3 extension manually loaded and paired; installing the plugin does not install the extension.

Browses and manages user skills already on disk in DSH and supported agent home directories.

Paper reading workbench: library + PDF reader + tutor-style companion. Answers cite clickable page numbers that jump back and flash the exact passage; select-to-ask without leaving the paper; bundled strict-tutor preset (reading plan, section coaching, graded quizzes tracked across sessions). With MinerU configured (self-hosted, no signup / mineru.net cloud token): scanned PDFs become readable, formulas become click-citable (one click sends the LaTeX source into the chat) and searchable (QK^T hits directly), plus a per-page formula panel; optional embeddings endpoint adds semantic recall for plain-language questions. Rename/delete topics and papers — deletes go to a recoverable .trash with a batch-clear flow for non-empty topics; reader with scroll/single-page modes, rightbar-fullscreen safe, one-click path copy and translated-PDF export; in-place original ↔ Chinese toggle (babeldoc, zero preinstall). Transcription and search run locally by default — documents never leave your machine; only when cloud MinerU, an embeddings endpoint, or a translation endpoint is enabled does the corresponding text go to the service you configure.

Kimi no Na wa (Your Name) theme: cinematic wallpaper, comet-blue glassmorphism, movie logo swap and unified scrollbars for the DSH Web GUI.