Plugins

Feishu/Lark channel that pushes finished or failed turns to a phone and delivers group messages back into the workspace session. Group messages are delivered into the session as user messages, so anyone who can post in a chat listed under task.chats can instruct the agent with that session's tool permissions; no chat is listed by default. The app secret is stored in plain text in ~/.dsh/feishu/app.json.

Adds one system-prompt section that limits delegation to subagents and teammates to two cases: adversarial review of freshly produced output, and one-shot context reading that returns a summary.

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.

Turns the QQ groups you follow into a Chinese daily digest. Pulls chat history through a local NapCat/OneBot 11 service into SQLite, summarizes each group, and renders an HTML/Markdown report with an opportunity radar that surfaces repeated complaints, tool requests and unmet needs as product ideas; can push a card to Feishu on a schedule.

Writes engineering decisions down the way DSH documents its own — every non-trivial change leaves a Markdown note under .agents/notes/<lifecycle>/<class>/ holding the problem, the decision, the rejected alternatives and the consequences, committed to git and reviewable like code. It supplies the workflow for that: five tools (note_search, note_read, note_write, note_verify, note_skip), an end-of-turn hook that asks a code-changing turn to either write a note or record an explicit skip, the tree/format/archived checks, and the write-notes-like-deepseek skill bundled in the package and mounted by the plugin. A sidebar board that renders the notes into four views (architecture baseline, timeline, rejected proposals, decision list) comes with it; the notes themselves stay plain Markdown.

Blocks the first animated-image read in a session until the model has read a strategy document, returning the opening lines of that document in the denial and releasing the session once it is read; only reads by file path are covered.

One credential store for many API platforms: keeps keys for GitHub, Cloudflare, Vercel, npm, Docker Hub, Hugging Face and custom platforms in the DSH credential store rather than in configuration or the session log, and gives the model three tools. apikey_call takes a path rather than a URL, so a key only reaches the host its platform fixes; apikey_platforms lists platforms and accounts without values; apikey_key gets, sets or clears one. Supports five auth shapes (bearer, custom header, x-api-key, basic, query string), verifies a key against the identity the platform returns, and reports transport failures as readable results instead of ending the turn. By default (exposeKeys on) the apikey_key tool can return a stored key's value to the model, which puts it in the session log; turn exposeKeys off to keep keys inside the Host.

Homework helper for the AI Passport device: turns homework screenshots into a checklist with per-task countdown timers and push-to-talk, plus a parent dashboard. It runs its own HTTP server on 0.0.0.0:8787 by default so the device on the same Wi-Fi can reach it: the parent dashboard only accepts connections from this computer, and a device needs a token issued after you approve its pairing code.

Externalizes oversized assistant and tool messages in long sessions into project files while keeping a recent window, and archives reasoning from empty turns as a visible message so long tasks do not stall on context size.

Sidebar tab with every file this conversation produced — preview images, PDF, Markdown, CSV and code in place, and jump back to the turn that made it.

Drops text blocks that carry nothing but leaked DSML tool-call tags, then rewrites the resulting empty turn into a retryable error so the host's built-in retry re-issues the request; prose that merely contains tags passes through untouched.

Shows a Kobe iced-tea side popup and reports per-turn costs or token counts with bundled offline audio. Uses dsh-cost-meter first, or estimates each request from native usage and public prices. Includes twenty cost phrases, four token phrases, CNY/USD, three-decimal rounding and custom previews; no runtime TTS service or API key. The bundled voice clips are pre-synthesized offline and the portrait is a fan-made image; the package declares UNLICENSED. When it has to estimate a cost itself it fetches a public price table from raw.githubusercontent.com (Han-1413141/dsh-cost-meter) and the USD/CNY rate from open.er-api.com, at most once a day, and sends no session data.

Adds a Markdown outline and full-reply search to conversations, with section references, reversible turn hiding, and new sessions from selected turns.

Makes a DeepSeek Harness agent reason in Simplified Chinese: each turn it appends one runtime-context instruction at the end of the conversation, and it can keep a stable declaration in the system prompt as well. Settings pick Simplified Chinese or off, and whether that declaration is included.

Collects the failures recorded in DSH session logs — tool failures, turn failures and model retries — folds them into one table merged across sessions and sorted by frequency, showing category, source, error code, count, last-seen time and a summary per row. Each row can start a new session draft pre-filled for root-cause analysis or for prevention (sending still needs your confirmation), and ignoring a row only hides it in the panel without deleting logs.

Moves reasoning that a provider leaks into the visible text channel back into the reasoning channel, and emits a text block when a whole turn of answer text lands only in the reasoning channel. Catches three shapes: `<thinking>`-tagged text, text that continues a reasoning tail which ended mid-token, and an answer-only reasoning turn that stops. Healthy streams pass through unchanged, block order is preserved, and each of the four rules can be switched off in one JSON config.

Sound cues played by the host operating system, not the browser: three bundled WAV files fire through the machine's own audio (SoundPlayer on Windows, afplay on macOS, paplay/aplay on Linux) when an agent's turn completes, when the context is compacted and when the agent is waiting for your answer; subagent turns can be included, cues fired closer than a configurable interval are skipped, and each bundled sound can be replaced with your own file via plugin config.

Posts a macOS notification titled with the session name when that session finishes a turn, and clicking it returns to the session. An in-app notice is always shown alongside it (9 s in the foreground, 60 s in the background); the system banner needs notifications allowed for DeepSeek Harness in macOS settings, otherwise only the in-app notice appears.

Windows toast when a turn finishes, with one setting in Settings → General: Never, Only when unfocused, or Always.

Plays a distinct chime when a turn finishes, is interrupted, fails, or asks a question, and can show a Windows toast for questions.

LAN task delegation between DSH hosts on one network: the parent dispatches a task to a child host, the child works in its own session and workspace, results and files return to the parent's inbox, sensitive actions wait for the parent's approval, and a finished task can be continued with one follow-up inside the child's existing session. Since 0.4.0 a fresh host listens on loopback only and keeps its listener off until it is enabled, the six-digit pairing code is single-use and rotates with a lockout after repeated failures, tokens travel in an Authorization header and are stored hashed, and a dispatch with no grant or with a permission above the grant is refused before anything is sent. Transport is still plain HTTP with no TLS, so use it on a trusted LAN.

Turn useful conversations into a knowledge base you can revisit. Start a discussion with /cardex and end it with /cardex end; a model organizes the questions and answers into Markdown cards, a knowledge tree, and an offline graph. Related discussions can add to existing cards, with links back to the original questions and answers, so your knowledge keeps growing and stays easy to browse. Cards and the graph are written to ~/journal/cardex unless you change it with /cardex dir, and the recorded discussion is sent to the organizing model (the session's model unless you pin another one).

Continuous voice conversation for DSH: auto sentence splitting, a 2-second confirm window before sending, and spoken replies you can interrupt by speaking. Speech is recognised by DSH's own speech service by default (a local Whisper bridge or an OpenAI-compatible endpoint can be set instead), the recognised text is corrected by a DSH model call, and replies are read aloud through Microsoft Edge's online text-to-speech service, so reply text is sent to Microsoft. A daily check of PyPI and Hugging Face for faster-whisper updates is on by default; settings and any API keys are stored in plain text in ~/.dsh/voice-tutor-config.json. Its /voice-tutor routes require an authenticated DSH session (admission is checked per request), API keys are never returned, and applying an update needs a one-time confirmation token.

Turn Bilibili links into b23.tv short links with covers, with a tool, a /bili command, and a share-card panel.