Plugins

Add-on layer over the official QQ bot plugin @tencent-connect/dsh-qqbot: gates group messages before they reach the agent (probability, debounce with backlog, invalid-message filtering, admin approval before reading attachments), greets idle private chats, and adds emoji stickers via dsh-meme, TTS/ASR voice replies, image generation and one settings page. Needs the official QQ bot plugin installed (plus dsh-meme for stickers). Group gating defaults to a 25% chance and only applies when mentions are not required; the private-chat greeting is on by default (22:00-08:00, at most 3 a night, only to chats that wrote first); voice and image generation call endpoints you configure; with no admins configured nobody can approve attachments.

On-demand MCP tool loading. Only the mcp__* tools listed in the coreTools whitelist (max 5) stay in a session's tool set; every other MCP tool definition is actually removed from the request through the session-scoped tools.restrict() API and comes back only after the model calls lazy_load_tool. A tools/pre-execute gate turns calls to unloaded tools into a load-it-first error, and activation resets when the session ends.

Agent gateway manager: picks a working gateway among proxies and direct, verifies the model channel, and switches online; DSH plugin plus standalone CLI.

Requirement to a detailed design doc another AI can build from: the model fills a plan, the plugin validates it and emits a SPEC plus task cards, then independent review and implementation evidence check the result. Needs Python 3.11 or newer; the spec-graph engine it calls is bundled in the package.

allow/ask/deny permission gates for DeepSeek Harness tools: path/params/agents surfaces, argv decomposition, network rules, redaction, grants, dry-run, fail-closed config

Access the DSH Web GUI from your LAN: a listener on the address you set forwards the browser to the loopback GUI with this process's launch token, and 0.0.0.0 answers on Wi-Fi, VPN and wired alike. Address, port, session lifetime and password are edited in an Access section of Settings. The password is on by default — a fresh entry starts with a generated random one — and can be switched off on a network you trust for direct access; the Host stores only its scrypt hash.

IPv6 remote-access gateway for DSH: PIN-protected HTTPS reverse proxy with a Let's Encrypt certificate (DNS-01 via DuckDNS), QR scan-to-connect and a settings page, no relay in the data path. Needs a public IPv6 address (detected on Windows and Linux only, not macOS), a free DuckDNS subdomain and token, and inbound ports 8787/8788 open on the router; the AAAA record is only auto-updated when the plugin config sets ddns or tls.

Cold-start boot animation for DeepSeek Harness, with an unlock gate in front of the workspace: PIN, Windows Hello or an authenticator code, all local.

Forces every active session into the full-access permission preset (this disables the approval gate) and appends approval, sandbox and tool-call events to a local JSONL for after-the-fact review. Audit path is configurable; it never uploads anything.

Replaces identity numbers, phone numbers, bank cards, emails, addresses, customer names and secrets with numbered placeholders before a model request leaves the machine, restores the response, and keeps an encrypted local mapping plus a hash-chained audit log exportable as CSV.

Freezes a six-step repair chain into callable tools — freeze the facts, measure the real shape, minimal repro, single-variable A/B, a judge that must say OK, delivery gate — plus a zero-dependency boot self-check (path drift, missing files, version sync, crash-log freshness), per-item tunables and a rollback ledger that backs up every action.

Toolbox panel for the DeepSeek Harness Web GUI: a right-side drawer (full screen on phones) with connection information (Wi-Fi details and gateway ping need a Windows host), one-click diagnostics, a speed test with server-side timing, a quality monitor, history and auto patrol; the host registers the /api/toolbox routes.

An agent preset that adds a working mode to DeepSeek Harness: the mode text has the model analyze requirements and plan before acting, open a gate document, and reply in a fixed three-block shape. The task tree and session-start profile injection are separate companion packages in the same repository, installed from a clone with scripts/install.mjs; only the term dictionary (dsh-term-dict) comes with this package, and without the task tree the model writes its own alignment bar. Needs DSH 0.2.1-alpha.1 or later.

Delegates a one-shot task from DeepSeek Harness to a locally installed Reasonix over the Agent Client Protocol, reusing Reasonix's own model, credentials and sandbox settings; exposed as the `subagent_reasonix` tool. Auto-approves every Reasonix permission request.

Runs your kanban backlog during DeepSeek off-peak hours while the machine is idle: per-card git worktrees, evidence-gated completion, merge-back, and budget and idle gates. Linux or WSL only (needs bash, python3, git, jq, flock); initialise a git repo once with `npx dsh-offhours init`.

Agent orchestration for DeepSeek Harness with eleven selectable presets, planning, delegation, review gates and durable project state.

Persistent concurrent node-graph workflows for DeepSeek Harness with agent, command, HTTP, human, fan-out/join, gate, and subworkflow nodes.

Adds a DeepSeek Harness agent preset that hands implementation work to Codex subagents (auto-approved, workspace-write sandbox); the session agent only splits tasks, dispatches them and reports results. Requires DSH 0.2.2-rc.1+, the @deepseek-ai/dsh-subagent-codex bundle and a locally installed, logged-in Codex CLI.

Packages a working style into a versioned pack of rules, skills and presets injected into the system prompt at global, project or session scope, distills local agent sessions into aggregate reports and suggested rules you can adopt or undo, exports the pack as a tarball a teammate can install, and ships a reference FDE pack with 40 skills.

Writes a novel from one idea to a validated manuscript with a code-driven workflow: a JavaScript state machine runs the stages (constitution, outline, design, writing, validation, batch, export) and a 15-step per-chapter loop, PowerShell scripts validate each gate, and versioned checkpoints let a restart resume at the exact chapter. Nine subagent roles do the drafting and review. Needs PowerShell 7 (pwsh) and Python 3.7+; without pwsh the gates record a DEGRADED verdict and the run pauses.

Runs the agent through GEML workflows: work items move through flow steps, practices are recorded at the step they belong to, and a board tab in the right sidebar shows where each item stands. Also mounts the GEML MCP server and skills; when the project has a fixed workflow, a supervisor gates tool calls by its current state. Needs Node.js 22+; the GEML MCP server is fetched with npx (@geml/geml) when a session starts.

Experiment-orchestration SOP skill for long-running experiments. A top-level model drafts a decomposition plan — subtasks, the model for each, an acceptance figure with tolerance, and dependency order — and dispatches only after the user confirms that plan; each subtask then goes to a lightweight executor subagent as a self-contained prompt, and its report is confined to measured facts and the delta from the agreed figure, with judgement and decisions staying at the top level. Ships one plan template and five task templates (gather, execute, consistency check, read-only review, mechanical transform), acceptance gates split into a task layer and an experiment layer, and three rules that stop internal-consistency evidence from being presented as independent or external validation.

Adds three session tools — create_image and edit_image over GPT Image or Gemini through the TLPro gateway, and speak for text-to-speech through xAI — plus a viewer in the chat UI for the images and audio they produce. Needs your own TLPRO_API_KEY and XAI_API_KEY, and publicOrigin and mediaDir set in the profile config.

Serial model orchestration: a delegate tool over a built-in catalog of current Chinese-market models (DeepSeek, GLM, Doubao, Hunyuan, MiniMax, MiMo, Kimi, Qwen, GPT). The main agent splits work, picks a model per subtask with a required reason, waits for each result, and falls back to a default model; every route is checked against the Session model-selection policy before a child starts. Needs Plugins → Subagent → Model selection enabled for the routes you want (new sessions only) and each model's provider configured. Uses the same plugin id as other model-router plugins, so install only one of them.