Plugins

Node-based GLSL workbench: wire a shader graph on a canvas and preview it live, compile it to Shadertoy-compatible GLSL with multi-pass buffers (BufferA–D), hot-reload as you edit, cross-compile to HLSL / MSL / SPIR-V, and keep projects, materials, snapshots and resources (images, audio, lyrics) in a local workspace. Ships 57 built-in nodes, 8 bundled examples and 11 agent tools. No runtime dependencies.

Shows the provider behind a session's current model as a compact badge in the DSH Web composer tool row, left of the model selector.

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.

Cross-session memory for your dsh agent: remembered, recalled, and re-verified instead of assumed.

SSH server dashboard for DSH: per-host status, CPU, memory, disk and network meters, trend charts, an xterm.js terminal, SFTP file transfer, and server_box_hosts / server_box_exec / server_box_upload / server_box_download chat tools; imports ~/.ssh/config and mounts as a sidebar tab or a standalone drawer.

Removes sandbox_permissions / justification arguments that cannot widen the call's current sandbox mode, so calls no longer fail with 'sandbox escalation to "workspace-write" is not strictly wider than this call's current "danger-full-access" mode'. GPT models tend to fill these optional fields on every call and pick workspace-write, which is narrower than danger-full-access. Runs in tools/pre-execute: compares the requested mode with the mode the session's sandbox policy resolves, and drops the fields when the request is the same or narrower (also drops a lone justification). Also understands the values GPT carries over from Codex's own shell tool: null, empty and use_default are treated as no request, and in a danger-full-access session any value (including require_escalated) is dropped because nothing is wider. Real widening requests, unknown modes and tools that do not advertise DSH's escalation schema are left to DSH core. Never widens permissions. Configurable tool list (default bash, pwsh, write, edit, run_code). Zero dependencies.

Reload a plugin or skill you just installed without restarting DSH. reload_plugin clears that plugin's own module cache, unloads the old instance and re-imports it, restoring the previous module if the new code throws; reload_skill forces skill directories to be re-discovered. Ships both as tools and as the /reload-plugin and /reload-skill commands.

Source Control panel in the DSH Web left rail: search, switch and create branches with ahead/behind counts, a topo-order commit graph, and a click-to-expand preview of any commit showing its files (A/M/D), insertions and deletions, message, author and timestamp. An always-on footer dock above Settings carries the branch dropdown and recent commits and follows the open session; fetch and fast-forward pull are one button, and a git_worktree agent tool creates, lists and removes worktrees.

Session-scoped research evidence ledger with sources, linked claims, open questions, and explicit user review; provides a tool, slash command, and conversation panel, with selected-claim Markdown export that omits excerpts, review notes, and local paths by default; does not fetch sources or verify their truth.

Session-scoped delivery receipts for local artifacts: compare current file hashes with a recorded baseline, keep agent claims separate from explicit user acceptance, flag changed or unavailable files, and export a Markdown receipt through a tool, slash command, and conversation panel.

Skill center: browse and manage local Agent Skills, and search, preview and install skills from the ClawHub, ModelScope and QwenPaw marketplaces in one merged view with source and category filters; ships a `skill_hub` tool so the agent can install skills itself. It uses the same package name (dsh-skill-hub) and mount id (skill-hub) as cheshireez/dsh-skill-hub and amd47802574/dsh-skill-hub, so only one of the three can be installed.

Agent skill and tool for querying multi-platform model account balances: DeepSeek, Doubao / Volcengine, Qwen / Alibaba Cloud, Hunyuan / Tencent Cloud, and custom HTTP endpoints. Read-only, with 60s cache.

Speaker button left of the model selector in the composer tool row — one click scrolls to the start of your newest question, marks it with a blinking caret and reads from there through the newest agent reply (dsh-tts, browser voice as fallback); press-and-hold the icon and then click anywhere on the page to pick another reading start position; press-and-drag-up opens a vertical mixer for in-page media volume and system output volume.

Observes consecutive identical tool failures and optionally pauses before the next model step, preserving original results and successful repeats, with an in-chat report in DSH Web.

Mounts the hosted Reaudit MCP server (https://mcp.reaudit.io/mcp) through the in-box @deepseek-ai/dsh-mcp-client bridge with a REAUDIT_API_KEY bearer token: 238 mcp__reaudit__* tools for brand visibility in AI answers (ChatGPT, Perplexity, Gemini, Google AI Overviews), competitor and citation analysis, site audits, content generation and social scheduling.

Offline Wikipedia and ZIM lookups through your own kiwix-serve: the wiki_search, wiki_read and wiki_books tools run full-text search with snippets, read an article by exact title, and list the local ZIM library.

Mount the QianYuan MCP server inside DeepSeek Harness: one config-layer bundle brings 15 agent-record tools (pitfalls, cached results, identity, task board) to mcp__qianyuan__*. No runtime code; reads are anonymous, writes are ed25519-signed.

Connects DeepSeek Harness to the AdWhispr remote MCP server through mcp-remote with browser OAuth sign-in, adding tools for competitor ad research, ad creative cloning, and launching and managing Meta, Google and TikTok campaigns. Requires an AdWhispr account.

Move a session to another folder, permanently delete it, or AI-rename it from the Web UI — by dragging a session row onto a folder title or from the session menu. Moving rewrites the session header cwd, relocates the persisted log and updates workspace accounting so history travels with it; deleting stops running agents and removes the log, projection-cache row and workspace accounting. The same three operations ship as agent tools: workbench_session_move, workbench_session_delete and workbench_session_rename_ai.

Live token usage and estimated cost in the sidebar: the current task plus every session combined, an as-of clock for those totals, per-session time detail (elapsed, turns/steps, first-token, model vs tool time, decode speed), and a per-turn breakdown with each turn cost on hover.

A right-sidebar session panel that lists every Skill, MCP tool call and file read or write by turn and step, with per-failure error detail.

Four assistant output layouts with full reasoning, independent tool details, and frame-synchronized or fading text.

Runs git in the Host process so push, pull and fetch work inside the agent sandbox; exposes 8 agent tools (status, diff, log, stage, commit, push, fetch, pull) and 11 slash commands including /git-status, /git-commit, /git-tag and /git-name-set.

Declares a spec-driven development agent preset with a Chinese persona and four static Claude Code/Codex delegation tools for planning, implementation, review and knowledge distillation; roles are prompt-guided, not an automatic state machine.