Plugins

DeepSeek account balance above the sidebar account area, the running cost of the open conversation, the cost of each turn beside its token count, and a one-click AI breakdown for any turn above a token threshold.

Watchdog for quota exhaustion: on a 429 or quota error it creates a recurring schedule that re-sends a continue-the-task prompt until a turn closes normally, so a long task resumes without manual restarts; /keep-running list and /keep-running cancel <id> manage those schedules. Requires @deepseek-ai/dsh-schedule to be mounted first (it registers the schedules).

Sidebar workspace panel that switches between running tasks and all conversations, shown as one freely sortable card list with per-card pin, rename, copy ID, fork, stop, and archive actions.

Renders an Agent Teams swimlane canvas above the composer: lead centred with members below, tasks laid out in lanes by longest dependency path, and four SVG connector families drawn as dashed paths (task dependency as a cubic bezier from the upstream card edge to the downstream card edge, lead bus, peer bus, member-to-task tether). Cards are absolutely positioned DOM, connectors are one SVG layer sharing the same geometry constants. Layout constants and path algorithms are ported from cc-haha. Reads the live per-session agentTeam projection (roster, shared tasks, dependency edges, running state) and falls back to the session todo list when the session has no team; team messages are not part of the projection yet, so the communication panel shows an empty state. Body colours come from host theme tokens; the eight member identity colours are the deliberate exception — fixed hues mixed with the theme accent via color-mix, declared as CSS variables because the same values feed an SVG stroke attribute that rejects colour functions. Ships zh/en dictionaries with identical keys.

Live preview inside the sidebar session hover card: what the session is doing right now (running tool with elapsed time, thinking, last tool duration), an attachment thumbnail with click-to-enlarge, the latest user prompt and assistant reply with injected system scaffolding stripped out, a recent-tool icon strip, turn count plus token totals plus model name on one line, and a 12-item settings page (11 switches plus a slider) persisted to the profile.

Runs a chosen power action (shutdown / restart / hibernate / sleep / lock / logoff) once every Agent session and background job has gone idle: the plugin samples in-process sessions once per second, counts running sessions as always busy, checks running background jobs through a configurable option (considerJobs, on by default, shown with its live count in Settings), waits a confirm window and then a cancellable countdown, and shows a draggable bottom-right card with the current phase and remaining seconds; the enable toggle never survives a host restart, any newly started session or Cancel aborts the countdown, and if the agents service is unavailable the machine is treated as busy so nothing ever fires. Windows only (the built-in shutdown / rundll32 actions); every action except none degrades to none elsewhere.

A status bar in the composer footer showing the current phase, how long it has been running, the current step token count and a no-progress warning, projected from session logs without polling the host, plus a detail panel that opens on click.

Highlights sidebar session rows by state (pending approval, running, just finished) with color, animation and a matching status dot.

Two games inside DeepSeek Harness where the model supplies intent and a local engine owns the rules: a twin-brain snake that periodically draws a strategy card, and Chinese chess where the model only comments on a move picked from locally generated legal candidates. Both keep running when the model is slow, offline or out of quota.

Control Proxmox VE virtual machines from the conversation: 16 tools for the VM console over VNC (screenshots, mouse, keyboard, batched actions, Unicode text and clipboard) and for running commands and reading or writing files through the QEMU guest agent, all over one ssh connection to the node, so the host needs no X11 or Wayland.

Connects the DeepSeek Harness agent running on your own computer to the HarnessMate platform, so you can chat with it, exchange files and approve its tool calls from a phone or a browser.

Opens the Obsidian vault a DSH session is working in, on the first human message — so your sync plugin (Nutstore sync, Obsidian Sync, Obsidian Git, …) starts running and the files the agent edits actually get synced. Skips a vault that is already open, and opens a new window behind the current one instead of stealing focus. Also serializes writers per vault — a second conversation's write is refused before it runs and queued until the holder is done, so a sync client cannot splice two conversations' edits together. Windows only.

Windows balloon notifications for session lifecycle events (new / running / finished / error) across all workspaces, with a session-store watch that catches headless and A2A dispatch sessions whose events never reach the host bus. Windows only.

Bridge between DSH and the local WorkBuddy desktop app, in four parts: read, search and append WorkBuddy memory, skills, plans and daily logs; run tasks through the local CodeBuddy CLI with four permission profiles; drive the running WorkBuddy window over CDP (open a conversation, read its replies as sanitized rich HTML, type and send, switch the account-billed model, relaunch it with a debug port when it is missing); and a sidebar panel that puts that conversation surface inside DSH. 25 tools and 15 HTTP routes in one plugin.

Mounts the tsm-core-dlr semantic service (LanceDB vectors + graph) as MCP tools and ships the scenario-independent DLR/TSM modeling skill, a service-status overlay and the agent rules; the service is a hard dependency and must be running before the bundle activates.

Chooses which skills and MCP servers each workspace may use, from a "Skills & MCP" item in the workspace row's "..." menu; unchecked skills leave the session catalog, the `skill` tool and `/name`, and unchecked servers have their `mcp__<server>__*` tools denied for that workspace's sessions, running ones included from the next turn.

Session map: a fork-tree canvas over DSH sessions, and a Settings page that clones a Git repository into a new workspace. The canvas is reachable from a session-header toggle and two right-side tabs, and from a conversation tab that is off by default; it starts on the current session's own tree, growing downward, with wheel pan, Ctrl/⌘-wheel zoom, a subagent-session filter and a per-session overview of turns, steps, model time, tool time, output tokens and running state. The clone page takes a repository address and a remembered parent directory with a native folder picker, offers a per-clone shallow switch (off by default), runs git clone on the host, and registers the cloned directory as a workspace.

Adds a Delete session action to the sidebar session menu that permanently deletes the local records of that session, while refusing sessions that are running.

Shows a Spark Scope card at the bottom of the dsh sidebar: model server decode and prefill rates with sparklines, running and queued requests, KV cache, TTFT and TPOT p95, and GPU temperature, power, load and memory for each DGX Spark node. Read-only; needs a running Spark Scope server. Works in dsh web and the desktop app.

Orders and folds the DSH left sidebar by what is actually running — a workspace floats to the top when one of its sessions starts and keeps that place when the session finishes, only the N most recently active workspaces stay expanded, and six switches plus one count control each rule.

Adds a one-click Continue button beside Send that resumes a stopped turn by sending the "continue" prompt; hidden while a turn is running or the session is empty.

Persistent agent-preset (mode) selector docked in the composer input bar: switch the mode for new tasks, and set the new-task default from any running session.

Marks models in use by other running sessions in the DSH model menu, and restores the remembered reasoning effort and group collapse state for each model.

A small pet in the corner of the interface that reads the real event stream instead of guessing: it drifts when idle, speeds up while the agent is thinking, draws a flowing line per running tool call, turns red and shakes with the error summary on a failure, and celebrates a turn that ran longer than ten seconds. The same widget is the at-a-glance busy and error indicator, so there is no spinner to stare at. The host half folds real harness events into a snapshot and the client half only draws it; it performs no content recognition, and it positions itself away from the composer so it never sits over what you are typing.