Plugins

Plugin development toolset: mark files read by shell commands, sync code into a profile, mock a plugin backend, preview a plugin frontend, verify plugin loading, spin up an isolated test instance, run named shell shortcuts, and list the plugin's tools.

Command resolution forensics: which file a command actually runs, which PATH entry won, what it shadows, how many entries are duplicates or dead, and whether the copies behind one name report the same version.

Fork and delete buttons on each workspace row of the Harness sidebar, plus a /fork-worktree command: fork creates a git worktree and registers it as a new workspace, delete removes the workspace folder through git worktree remove (asking again before discarding uncommitted work) and drops the registration.

Bidirectional dispatch for DSH: opens visible desktop sessions for tasks dispatched in by Grok Bot / Yobiru, adds /dispatch-grok and /dispatch-zcode slash commands to send tasks out, and reports the session result back to the dispatching peer on completion (peers are user-configured; nothing is sent until configured).

A machine-checked reasoning ledger: every fact must resolve to a real file:line and be quoted verbatim from it, every assumption must state what would falsify it, and a conclusion may not rest on an assumption that never comes back for review. The ledger can carry a review date, a confidence and a kill condition. Ships the checker tool, an /fp command family, an optional gate that asks before irreversible shell commands, and a web panel listing the error codes each round fixed.

GPT-SoVITS voice output for DSH: a speak tool and /voice command, a read-aloud speaker button under each reply, and a settings tab to switch voices, tune synthesis, manage the local service and one-click install a voice pack. Windows only; needs a local GPT-SoVITS checkout with trained weights and its Python environment.

Requires Claude Code installed and signed in on this machine. Delegates coding tasks to the local Claude Code agent through the Claude Agent SDK, reports which files changed, supports one-command revert, and adds a Claude Mode toggle to the composer.

Turns a Minecraft Java save into a Blender cover or animation: scans the .mca region files for flat open ground, exports the terrain to OBJ, and drives Blender headless with MCprep materials, the BPS character rig and three-point lighting. After each render it reviews the frame with OpenCV for facing, ground contact, exposure, lighting and composition, then re-renders with adjusted parameters when a check fails. Requires Blender 4.2 LTS or newer and Python 3.10+ with opencv-python and numpy; the first run assembles the Minecraft assets with the bundled setup command.

Runs a Claude Code project's .claude/ slash commands and permissions.deny rules inside DeepSeek Harness. Command bodies expand $ARGUMENTS, positional arguments and @file references into the agent's next turn; matching tool calls are denied before they run. Every host service is probed, so a host without a command registry, working-directory service or TUI status line degrades one feature instead of failing the boot.

Genspark tools for DeepSeek Harness - web search, crawl, image/video/audio generation, media analysis, X/Instagram/Reddit search and model listing through @genspark/cli, with 17 typed tools and 18 /genspark-* commands. Requires @genspark/cli installed and logged in (gsk login), or GSK_API_KEY set.

Exposes a stdio MCP server as native DeepSeek Harness tools. Ships with the official Microsoft Playwright MCP preset, so the agent browses through the Edge or Chrome you already have open, with your logins intact; any other stdio MCP command can be configured instead. Host-only, no client bundle, no npm dependencies. To drive your own browser, start Edge/Chrome with remote debugging and allow npx to fetch @playwright/mcp.

Adds extra working directories to a workspace, each read-write or read-only, from a folder button next to the composer or the `/add-dir`, `/remove-dir` and `/dirs` commands; the model sees them in its runtime context, and `write` and `edit` calls into read-write directories skip the sandbox escalation prompt by default while shell commands still ask.

A DSH conversation workspace: turn actions, provenance quotes, local notes and summaries, in-conversation find, focus lock, and per-workspace command buttons.

Registers /compact_queue, a compaction command that does not drop the request when the session is busy: it compacts immediately when idle, otherwise queues the request in that session and runs it at the session's next idle window, with a 2-second fallback tick beside the agent status event so the trigger is not missed, a 30-minute wait cap, and /compact_queue cancel to withdraw a queued request or abort a running one. Each session keeps its own queue state from a single command registration, completion and failure arrive as a collapsed notice that does not wake the model, and every ManualCompactionError code except busy is terminal so a failure is not retried. Host half only; the core /compact command is left untouched.

DeepSeek API wallet balance in the composer dock with a popover showing topped-up and granted credit, the refresh time, and usage and top-up links, plus a /balance command and an api_balance tool; the Host reads a signed-in account wallet or the DEEPSEEK_API_KEY credential, so the browser only receives amounts.

Editable quick-action buttons beside the DSH composer: send preset text or insert it into the draft, then reorder, move, or resize the buttons. Text-only: runs no host commands and never overwrites an existing draft.

Visible terminal panel that streams the agent's shell commands into an in-app terminal and lets the agent capture the terminal screen as text.

Adds a ✨ button and an /enhance command to the composer that rewrite the draft into a structured prompt, flag open questions below the input, follow the current session model, and offer an undo.

DuckFleet: one flock of machines, one command. Add a machine once and its ACP delegation tool appears — drive one, several, or all of them from natural language. ／ 鸭群：一群机器，一声令下。添加一台机器即自动获得它的派活工具，用自然语言让某个／某几个／全部机器干活。

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.

Context firewall for DeepSeek Harness: credentials in successful plain-text tool results become opaque ⟦cloak:category:n⟧ placeholders before they reach the model, with originals kept only in a process-memory, agent-scoped vault. 36 built-in signature families (shared with the cloakpack git pre-push guard), key-name rules for unknown formats, custom regex rules, dry-run mode, an audit log, and the /cloak command.

Safety guard that makes self-killing commands impossible: a safe_kill tool that always excludes the caller and every ancestor process, a bash pre-execute check for pkill self-matches, sudo -S stdin grabs, rm -rf / and similar patterns, and an extensible rule library with block / warn / hint severities.

Copy-on-write permission preset: shell commands run inside a per-workspace Sandboxie box, so writes outside the workspace land in the box copy while the real disk stays untouched (Windows only, requires Sandboxie-Plus).

Adversarial review gate for tool calls. It hooks `tools/pre-execute`: a local regex pass over destructive commands (`rm -rf`, `git push --force`, `git reset --hard`, `DROP TABLE`, `format`, disk writes) and sensitive paths (`.env`, `id_rsa`, `.ssh`, `.git/`, `$DSH_HOME`) runs first with no latency, and only a hit is escalated to one review-model call that returns allow or deny. Plain reads and ordinary writes never enter that path, so everyday work pays no added delay. Every verdict is shown live in the Web GUI.