Plugins

Knowledge-base-driven agent preset: wiki long-term memory, a web KB UI, a management page and optional UI polish.

Cross-session archive: overview workbench, folded timeline, subagent lineage, full-text search, rule+LLM summaries, ZIP export, live tracking. Native session-page embeds: turn rail, docked right panel (live mode, pinned to original session on switch), rename (official-synced), conclusion digest, health monitoring + persistent header chip (idle / active / tool-wait / stalled). dsh v0.1.2+: fuses with the official TurnNavigatorRail — vector-icon info chips (duration, tool calls, errors, thinking) plus lazy tool distribution inside the official preview card, right-click opens the workbench; the classic piano-key strip is on by default and delegates unloaded-turn jumps to the official auto-load engine; the official rail can be hidden entirely. Settings dialog section with 10 tunable behavior params.

Auto-saves large pasted text as an attachment file in the DeepSeek Harness web input box; with dsh-better-sidebar installed, the paste pill opens the file in the sidebar editor and saves it back.

Codex-style desktop pet: a floating animated sprite in the corner that mirrors the agent's running state (working, waiting, failed, done).

A workbench beside DeepSeek Harness conversations with source navigation, Python class hierarchies, Git diffs, breakpoint debugging, and searchable Markdown memory cards. Go and Java navigation need their own language servers; breakpoint debugging needs separately installed debug adapters (debugpy, Delve, vscode-js-debug); Python class relationships are Python-only and in beta.

DSH side of Blind Whale Farmer, a SMAPI mod in the same repo that lets an AI play Stardew Valley from a text game state through MCP tools: syncs play limits and extra instructions to the running game, adds an optional session preset or a single stardew gateway tool (both off by default), and sends "keep playing" when a Stardew session ends on a tool call that came back as plain text. Needs the mod installed in the game.

Windows PowerShell 5.1 skill for downloading large GitHub release assets: it first read-only probes the host (listening local proxy ports, system proxy and env vars, an installed Clash Verge core and its config, curl version), measures KB/s of four routes in parallel (existing proxy, headless-spawned core, public mirrors, direct), runs a short time-boxed concurrency ladder on the winner, then downloads the asset in parallel curl byte ranges and verifies the byte count plus SHA256 against the release digest. Skips the ladder when the asset is under 100 MB or a single connection already reaches 3 MB/s, because concurrency only pays off when a single connection is throttled: measured 6.3x on a throttled node (1,234 to 7,759 KB/s) but 0.72x on a fast one (5,829 vs 4,175 KB/s). Needs no aria2 or 7z, never changes the system proxy, never asks the user for ports or concurrency, reports ASSET_NOT_FOUND with the real candidate names instead of silently picking a different file, and degrades to a size-only check with DOWNLOAD_SIZE_OK_NO_DIGEST when a release predates digests.

Big Fish Voice: makes the DSH web desktop pet speak. Whisper bubbles are read aloud automatically, left-clicking a bubble reads it, right-click/long-press reads it again, and clicking the pet replies with lines like 「主人，小鱼在这呢」. 10 offline Chinese voices: 2 local replica voices (鱼鱼酱 / 鱼鱼姐, no API key, ~58MB each) plus 8 Kokoro female voices (4 moe / 4 onee-san, models download on first use ~348MB, then work offline). Auto-reading can be throttled (unlimited / 30s / 60s / 3min); within one DSH instance only a single page plays audio. About 35KB package, inference in a worker thread, no dsh-pet files modified. The two replica voice models are for personal non-commercial use only.

Five tools for reading a video as raw material rather than as something to watch. video_outline maps a whole file as text and draws nothing at all: 4-60 equal windows, each with its cut count, speech share, intensity bar and tags, then the windows worth opening and the exact start/end for the call that follows -- zero image tokens. video_probe returns metadata and a reading plan without decoding frames. video_storyboard turns a window into one contact sheet plus a text index binding every cell to its timecode, cut count and frame time; cells whose text would be illegible at that scale are marked zoom, and dedupe global folds a picture that comes back after a cut. video_zoom re-renders 1-9 cells from a short window at a larger scale. video_read decodes only keyframes, OCRs each screen, and returns an outline plus \[mm:ss] text -- the path for lectures, slide decks and screen recordings, where the words are the content. It also reads a file with sound and no picture as a transcript (subtitle track, then DSH_ASR_CMD, then a transcript file you pass; long audio is cut at 45 s onto silence), composes six changed frames into one sheet when a file has no readable text at all, and can write the result out as SRT, WebVTT or text. Cuts, dissolves and one-frame flashes are found in a single ffmpeg pass; from ten minutes up a file is analysed on its keyframes only, 2.7x faster on a two-hour lecture; and a cell is never shrunk below a fifth of the source, so a 1440p sheet spends the same tokens on fewer, bigger cells. OCR is the engine Windows ships, tesseract with line boxes, or DSH_OCR_CMD. It takes a B station link (bilibili.com/video/BV..., a bare BV id, an av id or a b23.tv short link) wherever it takes a file: up to 300 seconds is read from the CDN in place with nothing written to disk, longer videos are downloaded once and cached within a 4 GB / 24 video budget, play URLs are cached for ten minutes, and a link that is not a single video says what it is. Needs ffmpeg, uses ffprobe when present, and has no other runtime dependencies. It is a wire-compatible reimplementation of SAOS-3624/dsh-video-storyboard: it registers the same video_probe / video_storyboard / video_zoom tool names, so the two cannot be enabled together.

Registers DeepSeek Search and Fetch API providers for the ctx.web seam, replacing both the web_search and web_fetch backends with DeepSeek's Search and Fetch JSON APIs, with configurable retries and backoff, settings that apply to the next call, and a configuration page on the bundle page. Requires DSH 0.2.1-rc.5+ and a DeepSeek API key with access to /v1/tools/search and /v1/tools/fetch (calls are billed by DeepSeek; account sign-in is not supported).

Owner workbench for independently connected family WeChat assistants, per-member project and tool permissions, scheduled services and delivery status; API acceptance does not confirm receipt.

Always-on-top notch for Windows that shows which sessions are working, which need you, and which are done or failed: one bot per session, never takes focus, hides under full-screen apps, with its own settings page.

Agent state machine plugin with a todo ledger, stall guard, phase trails and a process-isolated per-step Python worker with GUI automation lanes. Windows only; needs Python 3.10+ on PATH, and pyautogui, mss and pywinauto for the GUI lanes.

A long-form Chinese novel workbench inside DeepSeek Harness: read, edit and annotate the manuscript in place; AI reports problems with ready-to-adopt revision drafts, and the text changes only when you accept. A ledger for characters, facts and foreshadowing plus a world map keep a long novel consistent, and AI can continue, expand or rewrite chapters.

WeChat personal assistant that manages Apple Calendar and Reminders with confirmation before writes and sends scheduled WeChat messages, for macOS 14+ on Apple Silicon.

Bundles 19 short-video production skills (1 entry point, 1 orchestration, 17 atomic) into the agent skill catalog: Chinese TTS pronunciation fixes, text-to-image and text-to-video, image panning, composition, covers, media verification, data review and four-platform publishing. Ships skills and scripts only, so a runnable pipeline still needs ffmpeg, Python 3 + Pillow, Node.js and playwright installed locally, your own Bilibili biliup credential and WeChat Channels / Douyin / Xiaohongshu login states, and the placeholder values under skills/*/scripts/ filled in with your own. An installer's own skill of the same name keeps precedence.

Usage statistics: a usage and reference-cost panel for DeepSeek Harness with today, 7-day, 30-day, all-time and custom ranges aggregated by hour or day; a model filter and a per-model breakdown of requests, input and output tokens, cache, reference cost and share; a model-share card showing the top four models by total tokens with the rest grouped as others; a reference price list generated from opencode Zen and DeepSeek's published rates, with DeepSeek models billed by Beijing weekday peak/off-peak windows and Chinese public holidays; a pricing-match dialog to point an unmatched model at a reference model or type per-million rates by hand; reference cost switchable between ¥ and $ at a fixed 1 USD = 7 CNY; and a yearly heatmap drawn only with theme variables, with no third-party chart library.

Standalone Network Engineer Assistant agent preset plus a bundled 32-doc command knowledge base skill for DPtech, H3C, Huawei and Ruijie switches and routers: KB-first retrieval, cross-vendor command translation, topology and CLI-output screenshot reading, online manual lookup with cited sources, and mandatory risk annotations for dangerous commands; answers networking topics only.

Reversible delivery contract: rewrites operator authority and execution rules when enabled, byte-identical to stock when off, with host and profile clean.

Cross-session usage and cost observatory for DSH — token trends from hourly lines to 90-day and 1-year heatmaps, per-model DeepSeek peak/off-peak pricing reconciled against the official balance, GLM/Kimi/MiniMax subscription-quota windows with burn-rate forecasts, cache-hit ring, project rankings, CSV export, and a dsh-tui terminal face.

Executor/advisor workflow ported from pi-advisor-flow: the executor calls ask_advisor for a second opinion at key decisions, after repeated failures and before completion; equivalent repeated tool calls are intercepted before execution for an advisor review with a proceed/revise/blocked verdict, and hard mode force-reviews every turn before it closes. Per-call token and cost ledger with /advisor status readout, privacy tiers for repo context and file contents, six-shape secret redaction, and a web settings card; advisor failures never stall the main loop.

WorkBuddy bridge: registers the WorkBuddy LLM provider plus a multi-account ledger ball with realtime credits, daily auto check-in, rate-limit cooldown tracking and one-click account switching.

Composer and sidebar helpers in one plugin, each switchable in Settings: a git Changes panel (uncommitted / unstaged / staged / branch diffs), ↑/↓ input history, Shift+Tab /plan toggle, one-click ▶ Continue after a stopped or failed turn, drag-and-drop queue reordering, Copy deeplink, Auto Review "Allow for this chat", project status markers with active projects first, background notifications, 11 extra UI languages and automatic updates from GitHub releases through the plugin manager.

WeChat Official Account typesetting workbench: renders Markdown into inline-styled HTML you paste straight into the WeChat editor, with a sidebar panel, 11 themes and five model tools for block-level revision.