Plugins

Let the Blue Big Fat Fish browse Bilibili for you: it comments on your behalf, DMs you interesting finds, writes a diary, and keeps a memory for long-term summaries. It filters videos to your taste, creates Markdown digests, and can watch comments and reply for you. In Circle Mode, it replies only inside threads explicitly invited by the video uploader in a top-level comment — ordinary viewer requests do not count, and it never starts a new top-level comment. Commenting, DMs, auto-run, and comment watch are off by default. Needs Node 22, DSH 0.1.7-rc.1+, a signed-in Bilibili account, a model route and Python.

Image generation and image reading for DSH: an Images page in the sidebar, labelled 图像 in Chinese (one row per model, 10 ratios, 1K–4K sizes, a contact sheet with detail cards, zoom, reuse and delete), three preset providers plus custom endpoints — one needs no key at all — and a vision-review skill. A pasted key is written to ~/.dsh/secrets/media-tools.env (mode 600) and only sent to the provider you picked. The page follows the harness language (Chinese or English). Requires Python 3.9+ (generation runs through scripts/generate.py); the page needs desktop DSH 0.2.1+.

Creates, reads, and edits DOCX, PDF, PPTX, and XLSX through `/documents`, `/pdf`, `/presentations`, and `/spreadsheets`; renders pages to check layout, and turns selections in the sidebar preview into edit requests for the session. Python 3.10-3.13 required.

Windows-only plugin that watches the foreground window title, recognizes Bilibili video courses, classifies them by subject, fetches metadata/danmaku/comments, and generates Markdown + Mermaid knowledge maps and Word study manuals (Word output needs a local Python install).

Promo studio panel for DSH plugins: type a plugin name, match it to a case file, and render a 28-second real-UI film plus a poster locally. The V0.1 film pipeline renders on the local Chrome with its own frame renderer (shutter-interval sampling for motion blur, hard-cut handling) and ships the composition assets, so an npm install can render without an Onetake runtime. Needs Node.js 22+, Chrome or Chromium and ffmpeg; python3 for the sound mix.

Evidence-aware research writing skills for DeepSeek Harness: academic paper revision against supplied experiment records and references, rebuttal preparation, and presentation storyboarding, each with bundled Python tools that package the evidence and check the edits locally.

Local literature knowledge-base RAG: 10 tools (kb_ingest, kb_search, kb_rag, kb_zotero, kb_dedup, kb_clear, kb_fetch, kb_scope, kb_stats, kb_status), hybrid BM25+vector+reranker search, cited QA with clickable DOI links, all-local bge embeddings over a single-file SQLite index. Needs Python 3.9+ with pymupdf, faiss-cpu and sentence-transformers; the `npx dsh-kb-rag-install` installer sets those up and pre-downloads the model.

Smart Charts skill: turns CSV/TSV/TXT/XLSX/XLS/JSON data files into standalone interactive ECharts HTML - 32 chart types including offline maps, 3 themes, no CDN. A --profile data profile (per-column dtype, cardinality, missingness, samples, statistics, entity grain, insight signals) lets the agent read the data and pick the chart itself; each chart embeds a fact sheet (plot_stats + data_preview) whose numbers are the only quotable values. LLM transform code runs in a restricted sandbox (blacklist + AST whitelist + safe builtins); --doctor self-checks the Python environment. Fully offline, ECharts bundled.

Directed fuzzing driven by an LLM agent: from a target repository and a bug (CVE description, patch, crash trace, or file:line), it runs a PLAN, IMPLEMENT, EXECUTE, REFLECT loop in which the model writes Python input generators, until an input triggers the campaign's stderr-pattern oracle or the campaign stops.

Connector only, it renders nothing itself: subscribes to DeepSeek Harness session events and folds them into a 12-phase work state (planning / thinking / executing-tool / streaming / waiting-input / voice / awaiting-approval / complete / error / paused / idle / heartbeat), then POSTs each phase to a desktop-pet backend at http://127.0.0.1:8000/api/plugin/dsh/ingest. It holds no window. No API key or token is needed. Requires the Mao Live2D pet application to be separately installed and running (that is a Windows Electron app with its own PowerShell 7 / Python 3.10+ / Node 18+ prerequisites); without it this plugin has nothing to push to.

Reads bilibili video as structured text for AI: metadata, comments, danmaku, subtitles, local ASR, frame OCR. Local processing (no cloud upload), decoy-subtitle detection with frame check, instant re-reads via cache, AI-ready manifest + summary. Frame OCR needs ffmpeg; local ASR/OCR need Python 3.10+ venvs (plus the VC++ runtime on Windows); AI subtitles need a Bilibili login.

ShiGuang Memory Engine (SGME) bridge: multi-agent shared long-term memory via HTTP — L0/L1/L1.5/L2 distillation, scenario-based injection, unified search, and proactive care signals (memory_search / wiki_search / signal_pull / signal_claim / signal_ack), installable as `dsh-sgme`.

Aggregated multi-upstream web search for DeepSeek Harness: replaces native web_search with model-native grounding and agent search APIs, source attribution, and rich citation cards.

Find Android game packages and guide Google Play dependency analysis and modification. Needs Python 3 and the Android tools you pick (JADX/apktool/adb).

Third-party DSH bundle carrying the original deterministic-pixel-art Skill for editable JSON/PNG pixel art, Python rendering and recoloring within declared regions. Needs DSH 0.2.2-rc.1 (other DSH versions are not claimed compatible) and Python 3.10+ with Pillow.

Low-cost video understanding tool: a video_understand tool turns a Bilibili link / BV / local video into an AVIS info layer (ASR + scene structure + object tracks + YOLO labels) and returns summary + Q&A. Question-driven dynamic routing across layers (L0 ASR / L1 object tracks / L2 key-frame VLM), semantic-layer reuse for repeat questions, per-question budget cap. Python engine: core layer needs faster-whisper / opencv / yt-dlp (~200-300MB); the optional semantic layer adds ~2GB of torch / transformers / ultralytics. A bundled doctor --fix sets up the venv and installs both.

Registers two MCP servers that drive the installed AutoCAD and SolidWorks desktop apps over COM, plus two on-demand skills. Windows only, Python 3.11+; the pinned upstream servers install themselves on first load, and only the sides whose CAD application is registered get a row. Each server advertises two meta-tools instead of its full catalog (a discovery shim here for SolidWorks, upstream DISCOVERY_MODE=search for AutoCAD), taking the schemas loaded before the first request from 23,404 to 256 tokens and from 83,732 to 365 respectively.

Patent-writing gatekeeper for dsh - screens the idea against Chinese prior art first, then runs five-party alignment interviews, eight-chapter disclosure drafting, figures and simulation runs, seven-dimension review and docx/PDF export. Skills work out of the box; the search, export and experiment tools come from a bundled Python MCP service that ships disabled - one line in ~/.dsh/patent-services.yaml enables it, and uvx fetches the wheel from PyPI.

DSH-native integration layer for the Minta memory engine: composes the official dsh-mcp-client row, registers the minta-memory-governance skill plus a session-start memory prewarm, and ships a Minta agent preset; requires the separately deployed Minta engine (Python/Docker) to provide the 19 MCP tools - not usable without it.

Let your coding agent dream: strategies must survive counterfactual replay before they earn trust. An offline self-improvement loop over an evolving world model — no fine-tuning, the LLM stays frozen. Speaks MCP. Ships its Python server inside the package

Open source links from DeepSeek Harness conversations at the referenced line, review Git diffs beside the chat, inspect paused variables, and turn explanations into searchable Markdown memory cards. The workbench also shows Python class relationships (beta). Go and Java navigation require language servers, and breakpoint debugging requires a debug adapter.

Controls the Chrome you are already logged into: the model clicks, types, fills in forms, reads pages, takes screenshots and downloads files in your real tabs, including inside same-origin iframes. Runs on macOS, Windows and Linux; needs Chrome 116+ and Node 18+. Works over native messaging and chrome.debugger instead of a listening port. Ships 12 browser_* tools and a cb CLI, dispatches real CDP input events, draws an on-page cursor that also appears in screenshots, and adds a toolbar panel with live status, the controlled tabs and a pause switch. browser_download saves a link, an element's target or a URL with the site's login cookies (following redirects, keeping the server's file name) to the Downloads folder or the session's working directory; cross-origin iframes are named in the snapshot but not entered. The native host and the MV3 extension travel inside this tarball: on first load the plugin copies them into place and writes the native-messaging registration itself, so there is no clone, no install.sh and no python3 - the one step left is adding the extension to Chrome: install DSH Bridge from the Chrome Web Store (https://chromewebstore.google.com/detail/bippcbbbipppmhjkjomgdnjbooecbimo, no Developer mode), or load the bundled unpacked copy at chrome://extensions (the plugin opens that page and puts the folder on the clipboard for you). Until then the tools fail loudly with those steps. On Windows the registration goes into the registry (for Chrome and Edge) and the host talks over a named pipe. A daily CI run on GitHub's macOS, Windows and Linux machines installs the plugin, loads the extension into a real Chrome and reads a page through the browser_* tools.

FormatForge: forge 30+ file formats (PDF, DOCX, PPTX, XLSX, EML, EPUB, TOML, archives) into AI-readable text via a drag-and-drop inbox, ff_translate/ff_formats/ff_result tools and a CLI; scanned PDFs fall back to local OCR or an enhance hint for the session model.

Registers the minas-webdav MCP server as one stdio row, giving the agent 14 tools for a Xiaomi Smart Storage (MINAS) NAS without mounting it: list, stat, tree, find, du, df, cat, upload, download, mkdir, move, copy, delete and a capability probe. The Python package behind it is installed separately with pipx.