プラグイン

Runs project checks described in a dsh-checkup.json file (files, syntax, JSON, HTTP health, asset counts, and any node/npm scripts or commands you list) and falls back to auto-detected defaults when no profile exists.

Corner card that watches Tibo (@thsottiaux) on X, shows new posts and replies with a Chinese translation, and marks reset-related posts; extra JSON sources can be added. Reads the third-party tiboreset.com text mirror rather than X itself (a few minutes behind, text only); translation goes through the host translation service, which may send post text to a public provider unless translate is set to false.

DeepSeek Harness port of the Impeccable frontend-design skill: 24 design playbooks covering shaping, critique, audit, polish, layout, typography, motion, accessibility and design-system documentation, driven by the reference files shipped in the skill plus PRODUCT.md / DESIGN.md in the project instead of a downloaded engine binary. Registers the impeccable skill; reads project files and writes design artifacts under .impeccable/. The binary-only features (live browser bridge, pixel measurement, font matching) fall back to static variants, screenshots and explicitly declared manual estimates. Ported from pbakaus/impeccable (Apache-2.0).

Render any Markdown report or DSH session excerpt into shareable PNG images with no npm dependencies, driving the Chromium-family browser already on the machine and falling back to PDF, then HTML. Reports come out as a summary card plus a full verbatim long image; a report written for dsh-fact-check also gets the tier badge, the two rails and the per-claim tiering, all taken from the report itself, and a report following one of the other report skeletons it defines gets the section that skeleton designates as its conclusion lifted onto the summary card. Conversations come out as a message flow, exportable by turn range, event range or the most recent N turns.

DeepSeek web search with its own API key: a left-sidebar search page and a deepseek_search tool that every model in the profile can call, falling back to the DeepSeek key already configured for the model provider. It registers its own tool and routes instead of replacing the built-in web_search and web_fetch providers, so an existing search setup keeps working.

Lightweight fork of DDDMUC/dsh-free-search: multi-engine web search and fetch with automatic fallback, failure cooldown and API key rotation; takes over DSH's default web_search/web_fetch providers. Cannot be installed alongside DDDMUC/dsh-free-search (same package name and bundle row).

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.

Connects DSH Foxy to Foxy Memory for cross-platform recall, with local JSONL fallback, platform filters, and a memory browser. Requires dsh-foxy.

macOS (Apple silicon) fork of aureth-labs/dsh-aurabot-qq: the same QQ persona bot (chat, voice notes, stickers, memory, song covers, modules) with the Windows-only code removed. It connects to a OneBot v11 gateway such as SnowLuma running in OrbStack or Docker, falls back to the macOS `say` voice, sends files over 30 MB to a containerised gateway through a loopback-only temporary download link, and installs the song-cover environment (Applio, audio-separator, uv, Rubber Band, separation models) at pinned versions. It keeps the upstream package name and plugin id (dsh-aurabot-qq), so it replaces upstream and cannot be installed alongside it; use upstream on Windows.

Provider pool for DeepSeek Harness. It runs a local OpenAI-compatible proxy that tries configured LLM providers in priority order and falls back on authentication, quota, rate-limit or server errors, with per-provider usage statistics.

AI news panel for the sidebar: 24-hour picks, recent stories, a top-ten trending board, the daily digest and keyword search, all read from the public AIHOT JSON API. Picks and All filter by six categories; feed cards show category, source, time, heat score, summary and, when AIHOT supplies one, its recommendation note. The Picks, All and Search feeds group stories by Beijing-time day and page ten at a time with cursor endless scroll plus a Load more fallback. Read-only and client-side only, with no API key, no build step and colors taken from the host theme tokens.

CPU, GPU and memory load in the DeepSeek Harness Session header: three compact cells with a percentage and a 2 px bar, coloured at 75 % and 90 %; CPU from os.cpus() deltas, GPU from nvidia-smi (Windows counter fallback).

A crypto ticker card in the DeepSeek-Harness sidebar footer, above the balance and Settings, showing BTC, ETH and SOL prices, 24-hour change and 24-point sparklines, sourced from OKX with automatic fallback to Binance then CoinGecko (CoinGecko has no sparkline history); clicking the header folds the whole card into a single small button, and the folded state is remembered. A fork of 18569663yz-web/dsh-plugin-crypto-ticker: do not install both.

Drag a folder onto the sidebar workspace list to register it as a workspace and open a new session. The Desktop app supplies the folder path; a plain browser page falls back to the host folder picker.

Explains Chinese internet memes to an agent: registers a meme_explain tool that returns the meaning, tone, origin and lifecycle of a sticker, which memes can and cannot be combined, occasions to avoid, and both a meme-version and a plain-version reply. Answers come from a local memory store with spacing-effect decay, falling back to a user-configured OpenAI-compatible vision endpoint (BYOK).

Native desktop notifications for DSH on Windows, macOS and Linux. While the DSH window is in the background, permission requests arrive as system notifications; where the platform supports notification actions, Reject and Allow-once buttons return the decision to the host approval waterfall, and returning focus or a timeout hands the request back to the in-app panel. Approvals and questions also play a separate input cue that is independent of window focus. Turn-end, question and error reminders, notification sounds and the cue are configured on a Notifications settings page.

An annotated browser tab: pages render in headless Chromium on the host and are served same-origin, so you can click an element or drag a region, write what should change, and send the notes into the conversation with selectors, visible text and rectangles. Requires a Chromium-based browser on the host (auto-detected, or set CHROME_PATH / browserExecutable). By default it provides the `browser` tab type itself; set `takeover: false` to keep the stock browser and use a hand-off button instead. Picking continues until you press Esc.

File dedupe and unify toolkit, shown as "小视频整理" in the DSH sidebar: exact matches by head/middle/tail sampling or full byte compare; near-duplicate MP4s found by a prefix frame-level multi-candidate matcher (only the first N seconds, default 40, max 60; a frame-level index plus multiple candidate tracks handle shifted intros and repeated segments). The "ignore intro" scan maintains a SQLite fingerprint cache keyed by absolute path and size, with MAD-first frame lookup and dHash/pHash fallback; incomplete cache rows are removed and rebuilt automatically, and a full scan rebuilds the band index. A "quick file lookup" takes one file, uses the cache when present, generates its fingerprint on demand when missing, and reports a task with live logs, progress and cancellation without rescanning the source folder. Duration-bucket matching always probes and captures fresh, and does not use the fingerprint cache. Batch rename or move by extension, UUID, file size or video length; empty-folder cleanup; MOV/AVI/WMV detected by header and transcoded to MP4 keeping resolution, frame rate and audio. Requires ffmpeg and ffprobe. Scans run in a sidecar worker with live progress, A-B single-direction dedupe and asynchronous fingerprint writes; one task at a time; local-only API. Every task that moves files has one-click restore: per-group/target _dupfc_restore.json records, deleted files skipped and counted, files outside the record renamed to UUID and moved into resetFail.

Rui Xi's Memory (dsh-memory-v2) — a local long-term memory plugin for DeepSeek Harness: four memory collections (semantic / conclusion / profile / temporary), per-day scene archives, two-phase consolidation and literal trace. When structured recall finds nothing, it falls back to searching the daily scene archive: time-window first (yesterday / last week / N days ago / dates), then a 2/7/30/180-day tiered sweep, with line-numbered citations. All data stays in the local DSH directory; semantic recall, diagnostic logs and the sensitive debug route are off by default.

Elysia's memory plugin for dsh. One shared, human-editable MEMORY.md across every session: two-disk mirror, history snapshots, cross-process write lock; entries tier between the main file and a searchable archive by hit frequency; retrieval is keyword co-occurrence plus BM25. Nine sections (persona, requirements, corrections, lessons, methods, requests, sessions, method library, ingested docs) plus a dedicated self-identity section for a character's own self-cognition, injected above the persona. Ships an opt-in memory pack (packs/elysia: 61,903-char identity + 2,482-char persona, distilled from 1,696 officially-confirmed voice lines and checked against a 112,175-line game-dialogue corpus) that is off by default, backs up before writing, only touches identity/persona, and can be uninstalled. Also a skills library and a sticker library (you import your own images; none are bundled) that sends a fitting sticker at the right moment. On by default: a text bubble with her avatar after each reply, which costs one extra small model request unless the agent already announced (voice.bubble: false turns it off). Opt-in voice playback needs a local GPT-SoVITS engine (about 27 GB, installed separately, NVIDIA GPU required; the plugin downloads about 200 MB of weights) and is Windows-only; the automatic second-drive mirror is Windows-only too (elsewhere set mirrorPath, otherwise there is no mirror).

A named-agent-team chat workbench for DSH: per-member model routes with fallback and 3D mascot avatars; private and group chats with mention relay and bounded parallel lanes; a whole department as one agent whose lead dispatches sub-agents that run in parallel on their own models before the lead synthesizes; record search with hit highlighting and today/yesterday/earlier buckets; a four-layer memory system (raw chats → atomic facts → scene summary → user persona) recalled by hybrid BM25 + embedding RRF; plus an avatar library and local/cloud speech input. Hybrid recall needs an embedding endpoint (default: local Ollama qwen3-embedding:0.6b; otherwise BM25 only); speech input needs a local jianchang512/stt service or a cloud ASR key, otherwise it falls back to the browser speech API.

Waifu mode for QQ: splits every reply into several short chat bubbles with a human-like typing delay, drops trailing periods, and keeps tables and code blocks intact. Works by patching @tencent-connect/dsh-qqbot in the desktop profile: run node outbound/apply-patch.mjs once, set im-qqbot streaming: false, then restart — until then only the waifu_* tools work.

A Feishu/Lark channel for DeepSeek Harness: WebSocket inbound messaging that steers a running turn instead of queueing behind it, Markdown tables rendered as native Feishu table components, code blocks kept on their own lines, 13 Feishu tools, voice input and output, media upload, a model fallback chain, and a web settings page split into six grouped tabs. Runs on a self-built Feishu app over the long connection, so no public endpoint is needed.

Blocks the end of a turn while temporary files from that turn are still lying around: it registers a listener on the serial agent/turn-stopping hook and, if anything is left, hands the model the file list plus two ways out — delete them, or justify each one and move it somewhere deliberate. Candidates come from two narrow temp-file sources, plus a backup-retention channel that is on by default and flags excess .bak copies of files touched this turn. The first is the .agent-tmp/ convention directory under the session cwd: whatever is left in it is reported outright, because that directory exists only for throwaway files. The second is the paths this turn actually wrote, captured through the fs/write-intent, fs/edit-intent and fs/delete-intent hooks; those are reported only when they sit under that cwd (compared after realpath normalisation), fall outside .git, node_modules, out, dist, build, target, vendor, .venv and the Python bytecode cache directory, match a temp-file pattern that requires a separator after the prefix so checkout.js and benchmark.ts are not taken for scratch files, and the file is still present — which keeps junk already sitting in the workspace from being resurrected as a complaint about this turn. One gate per turn, subagent sessions skipped, nothing deleted unless the backup mode is explicitly set to prune, and the injected wording follows the host locale instead of assuming a Chinese one.