Plugins

Vision MCP and DSH bundle for text-only DeepSeek: analyze_image, analyze_clipboard, compare_images and vision_status tools, a visual settings page, free GLM-4.6V-Flash by default, result caching and rate-limit tolerance; keys stay out of logs.

Plugin store: npm authoritative catalog + awesome curated list (550+ plugins, 11 categories), dsh-field quality verification, official `dsh plugin add/remove` one-click install, sidebar + settings entry.

Connects DeepSeek Harness to QQ over OneBot 11/12, Satori, Milky or the official QQ bot API, with a separate session per group or private chat and nine built-in tools: group management, chat history with image understanding, web search, page fetching, reverse image search and image generation. Optional proactive chatter reads recent messages on an interval and decides whether to join in, with the judging and speaking stages able to use different models. Models, proxy and allowlists are configured in the DSH settings page. First run requires an owner (auth.owners): with it empty the plugin drives no sessions at all (deny by default, so installing it does not expose the bot); once set, private chats and groups work immediately, guests only reach the guest tool allowlist while qqbot and qqconfig stay owner-only.

ShaderToy backgrounds for DeepSeek Harness: a WebGL2 player with a background library and five bundled presets, adjustable opacity, brightness, resolution and frame rate, playback options, and an in-app editor for the running shader.

DeepSeek Harness plugin that stops hand edits to generated files (lockfiles, protobuf/codegen output, files marked @generated or DO NOT EDIT) and tells the model which source to edit and what to run instead.

Game asset production pipeline: generate images, video and audio, turn a video into a coherent sprite animation, derive recoloured variants, adapt assets to an engine schema, install them into the project and verify the code actually references them. Requires ffmpeg for the video-to-sprite path, and an image/video API key (Agnes AI / OpenAI / fal) for generation.

Model settings panel for DeepSeek Harness: set the default model for new sessions, plus separate provider/model/reasoning-effort overrides for subagents and goal rounds, with live reload.

An OpenCode Go model source for DSH: it reads the opencode-go catalog from Models.dev and serves it on a separate opencode-go-live route, leaving the built-in opencode-go route in place. The catalog snapshot is cached under the DSH home and a failed refresh keeps the last successful one; its own Models settings card edits the route key and can render OpenCode Go subscription usage.

Generate images from the agent on the user's own GPU by queueing a workflow on a local ComfyUI install over its HTTP API: Z-Image-Turbo for fast drafts and Qwen-Image-2.1 for final assets and legible in-image text. Registers comfyui_generate and comfyui_status plus a skill covering model selection and prompt craft, has no runtime dependencies, and starts ComfyUI headless when no server is listening.

Command Code subscription tiers (Go / GOAT / Pro / Max) published as `llm-pi-ai` route profiles rather than an adapter: the live /provider/v1/models list and the plan page capability catalog are joined, each model lands on the OpenAI Chat, OpenAI Responses or Anthropic Messages route the gateway's own supported endpoints call for, switching tier deletes the routes the previous one generated, and the same credential drives five-hour and weekly usage windows, credit balances and a web-search provider.

Binds a separate DeepSeek API key to each workspace or session, warns when an effective key is stale or belongs to another provider, and never swaps a key on its own.

Circuit breaker for idle agent turns: watches agent/assistant-stream and cancels a turn that produces only reasoning — no text and no tool call — after 45s, after 80000 reasoning characters, or on a degenerate repetition pattern (exact repeat unit, sentence-template repeat, sparse repeat, or 6-gram density). On trip it injects a continue instruction that requires the next action to be a tool call, and ships 11 regression cases plus per-gate tuning.

Skill that fact-checks a claim against the open web rather than model memory: it splits the claim into independently checkable fact points, routes each point to the primary-source ladder for its domain (policy, academic, medical, financial, disaster and accidents, food safety, science, education, sports, entertainment, judicial, and more), searches domestic and international ecosystems in parallel, fetches primary pages, grades sources by type, re-searches with debunking keywords, and reports conclusion-first with a full URL and citation position behind every statement, keeping unverified and undecidable points separate from confirmed ones. Three factual judgements can be routed to a TypeSafe Jev (System One) decision model when a key is configured, falling back to the skill's deterministic rules when it is not; a model verdict can only lower a confidence tier.

Go (weiqi) review coach: parses SGF records, backfills winrate and score analysis with a bundled KataGo engine, explains the costly moves in the chat, and writes the commentary back into an SGF copy; a web panel shows the board with problem moves, AI choices and variations.

Orchestrates durable multi-agent teams from one DSH session: a captain spawns member subagents, drives a dependency-aware task DAG with quality gates, and coordinates through persistent mailboxes, with a live activity panel.

Unit conversion toolbox covering 20 categories — length, mass, temperature, area, volume (incl. US cooking units), speed, time, data sizes (decimal MB vs binary MiB), data transfer rates (Mbps vs MB/s), acceleration incl. g-force, illumination (lux / foot-candle), pressure, energy, angle, frequency, power incl. three horsepower definitions, force, torque, typography (px/pt/em/rem), and fuel economy (mpg ↔ L/100km). Two tools — convert_unit and list_units — zero runtime dependencies, pure arithmetic; each result includes the formula applied.

Records real clicks, inputs, navigation and network traffic in a user-guided browser session for DSH, then generates a Markdown timeline report (report.md + events.jsonl) to build API-level skills.

After official llm-retry (n/5) gives up, keep retrying transient API failures (timeout, transport, rate-limit, 5xx, dropped SSE) on the same open step until success, cancel, or pause; AUTH, QUOTA, and user Stop are never retried.

AI-driven presentation workflow packaged as a dsh skill: generates editable PPTX decks and SVG snapshots from PDF, DOCX, URLs or Markdown, fills native PPTX templates, and enhances finished presentations.

LLM request retry plugin for DeepSeek Harness: when a model request fails with a configured HTTP status code, machine code, or provider field=value match, it sleeps for the per-rule delay and re-issues the request automatically, so subscription quota and rate-limit errors (429 / insufficient_quota) do not interrupt a run.

Input-bar prompt optimizer: rewrites drafts with 6 strategies × 3 languages plus custom instructions, optional goal/todo/compaction-summary/tool-result and chat context, a confirm dialog, last-5 history and three-tier settings persistence.

Competition math (IMO/Putnam/USAMO/AIME) solved with a pure-reasoning pass, an adversarial verifier in a fresh subagent context, and calibrated confidence output (high / medium / honest "no confident solution"); optional LaTeX-to-PDF rendering.

Separate slash-command group in the Web UI composer: /steer injects text on the agent's next step without interrupting the turn, /new starts a blank session, /skill and /docs wrap /steer with post-turn skill or docs tasks, and custom /steer aliases are configurable in Settings.