Plugins

Deterministic loop guard: denies a tool call repeated with identical arguments and caps per-agent calls, in code outside the model; incident log lets a parent interrupt a stuck subagent.

Real-time repetition guard for DeepSeek Harness (DSH): stops model generation when the same string repeats >=10 times in the streamed output. 实时检测 DSH 大模型流式输出中的重复内容，同一字符串重复十次以上立即停止生成。

Global prompt-injection / context-virus defense for DeepSeek Harness: scans tool arguments, tool results, pre-model messages and the outbound stream; quarantine/block/monitor modes, canary trap, and an evolvable on-disk signature library with learn/import/export.

Installs a skill-router skill that semantically searches a local skill corpus on demand, keeping the corpus out of the per-turn model catalog.

Brings the codebase knowledge-base capability of langchain-ai/openwiki into DSH — one-click generate/read/update of the repository Wiki, Grounded Claims (source-traced knowledge cards) and a project knowledge base built from uploaded local files, reusing the model DSH already has configured, with no second API key.

Probes Windows-host Ollama, LM Studio, vLLM, and llama-server from WSL, compares ctx, and reports /v1/models apiReady versus TCP.

Reports Docker CLI, context, and daemon reachability in WSL, plus vLLM :8000 /v1/models health and GPU runtime hints.

Share the DSH workspace over the local network: phones, tablets and other computers browse, download and upload files via browser with no DSH login. Token gate, path boundary, read-only mode and rate limit. Zero-dependency (node:http).

Text-to-image and image-to-image generation via OpenRouter's unified Image API: configurable model aliases with parameters gated against the live model capability listing, reference-image inputs, a settings card, and inline chat display of results.

Turns a Xiaohongshu, Douyin, TikTok, YouTube or X link into text: transcript with timecodes, on-screen text, a description and OCR of every image, caption and metadata. Mounts the hosted MCP server over streamable HTTP.

Install and call external Agent CLIs (Codex, Claude Code) from DSH without touching the system install. Each CLI gets its own Provider, model, reasoning effort and permission preset, callable directly or as a native subagent.

Image generation workbench and agent tool for DSH: one tool (ark9_generate_image) paints text-to-image and image-to-image via any OpenAI-compatible API. Channels are entirely user-configured - no bundled endpoint, you bring your own baseURL, key and model name (manual input supported). Agent-initiated generations are approval-gated by default: requests queue in the floating panel and nothing bills until you approve; approval is non-blocking (the agent replies "awaiting your approval" immediately, and results, denials or timeouts are injected back so the agent reports them in the next turn). Floating glass workbench with five tabs (Generate / Approvals / Prompts / History / About), aspect-ratio grid with quality-budget + 16px-alignment sizing, transparent background, batch up to 10 images as aggregated sub-tasks, prompt library with custom JSON sources fetched through a host proxy, persistent generation logs with one-click retry, multi-channel aggregation with per-channel model fetching, bilingual UI (zh/en), stroke-SVG icon set, config import/export. Registers as a Better Sidebar tab when dsh-better-sidebar is installed; stacks above the dsh-cua FAB when both are present.

Cipher memory platform as a DSH bundle: four memory types (attention, experience, preference, cognitive) sedimented per turn by four memory agents, cross-session deterministic recall injected via systemPrompt.context, insight review kept out of user-facing output, per-session UNNI/LOOP modes with memory-driven auto-continuation, memory search/ingest tools, plus a settings page (SOUL editor) and an AgentPool panel.

Secret-redacted session handoff with review, auto-start, and brief-only modes.

Neutralize tool calls with invalid JSON arguments on the wire so one malformed model generation cannot brick a session against strict OpenAI-compatible servers.

Governs multi-model access, monthly credits, manual/quality/credit/auto routing, failure rerouting, and usage auditing for DSH.

Command Code LLM provider plugin adapted for DSH-Desktop 0.7.1: registers a commandcode route, with a live model catalog, reasoning-effort support, and real-time plan usage display.

Keeps long DSH sessions alive when a model channel starts failing: daruma takes over after the in-box retry gives up, trips a self-healing circuit breaker on repeated 429/5xx or a terminal error, and moves the next request to another channel in your chain. A panel beside the composer picks the backup by hand, switches show up as rows in the conversation, and every decision lands in a durable JSONL audit log.

Three independent language channels for model output — tool-call descriptions (every tool-call card title, including run_code in PTC-style presets), thinking, and replies — each set to follow the web GUI language, force a specific language, or turn off (default: only descriptions, following the GUI). One global runtime-context prompt directive on the host; the client half reports the GUI language; a Settings plugin card switches the modes.

OpenAI-completions-compatible adapter for custom gateways (vLLM / LM Studio / self-hosted proxies): always role:"system", thinking driven by the model config, Qwen-style response split, and vision-model image input (single / multiple).

Rule-based multi-provider model routing for DeepSeek Harness with pre-first-token failover, cooldown circuit-breaking, usage accounting, and a status API.

An F1 Race Control theme for DSH Web with four switchable team skins, broadcast-photo backgrounds, light and dark modes, and native appearance controls.

Token usage statistics by model, session and date in a sidebar panel, plus OpenCode Go quota, DeepSeek balance and Z.ai quota monitoring.

A lightweight plugin that adds model thinking token statistics to the bottom Dock and the end of each conversation.