Plugins

Turns a local Markdown vault into two folders the user picks in Settings: a knowledge folder DSH writes notes into, and a memory folder used as the agent's long-term memory with per-turn recall and end-of-turn capture. Pure Markdown with frontmatter, BM25 search, reversible archive, and path isolation that rejects traversal, absolute paths and symlink escapes.

Local long-term memory for DSH: browse, search (BM25 + synonyms) and edit entries in a panel; folders, archive and a recycle bin; choose per session what to inject; model-written memories need your approval; import/export backups; optionally sync across devices via your own Git or WebDAV.

Semantic context watcher for DSH agents: maintains a journey ledger with flash summarization and MinHash novelty detection, folds milestones, and can inject long-term context with a three-level stall response; injection and the stall-2/3 responses are off by default.

Adds a permanent Delete session action to the sidebar session menu, behind a confirmation dialog that reports the log size, running turns, subagents, background jobs and reminders bound to the session; the id is then released from the live, archive and workspace registries, so the row stays gone on refresh instead of folding back in.

Context observability and control for DeepSeek Harness: cross-session token analytics, session-level budget/prompt/tool controls, and optional agent scaffolds. It reuses the official tokenMeter, compaction, systemPrompt, and tools.restrict seams; monitoring is read-only.

Deletes a session from the sidebar session menu: its log and replay cache go to the Recycle Bin on Windows (a local trash folder elsewhere); a live session is queued for removal on the next start.

Gives the model a private glossary: reads 「滚木」 as "none" and 「棍母」 as "doesn't exist", and decides on its own whether to phrase a sentence with those words when expressing that meaning. Registers one system-prompt section and rewrites nothing; word lists ship as per-category YAML (moba/fps/comments/ai/dev/private) with a settings page to toggle files, change reminder strength and open the glossary folder.

Windows only. With Wallpaper Engine installed it follows the active wallpaper or a pinned pick from the Workshop library; without it the plugin ships four starter wallpapers in a drop folder. Translucent panels, per-block blur and wallpaper-matched text contrast.

A draggable balance and usage card for the web UI. Clicking the character button slides the card out from the left side of the image and clicking it again retracts the card; it shows the official DeepSeek account balance split into granted credit and top-up, and token usage folded from local session logs for today, the last 7 days and the month, priced per call with the official peak/off-peak rates (cache hits, cache misses and output are priced separately).

Replaces the composer model picker with a compact panel for provider-to-model navigation and an adaptive reasoning-effort slider, with a Default stop and no slider for models without adjustable efforts.

Expands the assistant reasoning block while it streams and folds it back when done, animated in both directions; the sidebar switch only controls whether that happens automatically.

Collapses the sidebar global-panel rows to a few visible entries and adds a picker for choosing which ones stay, so a long panel list stops squeezing the workspace and session area; rows are hidden only, never unregistered.

Adds an action to the session menu and the session title bar that opens the artifacts folder of that conversation: it reads the session log, works out which directories the conversation wrote files to, and opens them in Explorer.

Replaces the host-side folder chooser in the Add workspace flow with a name dialog, so a remote browser can create and open a workspace instead of an OS dialog appearing on the Host screen.

Publishes the desktop shell's file-path bridge as the host-path seam the composer reads, so files and folders dropped into the composer become @path references instead of uploaded copies (images still upload through the native image channel).

Folder-scoped sidebar for DSH Web: by default only folders you load are listed with their sessions; unload a folder without deleting its registration, mark folders to load at startup, and opening a session loads its registered workspace so it does not fall into Ungrouped.

Folds every panel row in the sidebar (slot-registered and plugin-injected alike) into one Plugin Hub entry, giving the workspace list its vertical space back.

Live preview inside the sidebar session hover card: what the session is doing right now (running tool with elapsed time, thinking, last tool duration), an attachment thumbnail with click-to-enlarge, the latest user prompt and assistant reply with injected system scaffolding stripped out, a recent-tool icon strip, turn count plus token totals plus model name on one line, and a 12-item settings page (11 switches plus a slider) persisted to the profile.

Folds the durable todo and goal events the dsh pet ignores into a percentage bubble on the pet, and falls back to a one-line activity summary for a session that never wrote a todo list. Requires the dsh web pet plugin (@linxin666/dsh-pet >= 0.4.0).

Adds a Windows Explorer folder context-menu entry that registers the folder as a DSH project, with a switch for the menu entry in Settings → General. Windows only, and nothing extra has to be installed.

Local file manager served by the DSH web server itself: whole-disk browsing, previews for images, PDF, video and audio, text, Markdown, JSON, CSV, sandboxed HTML, docx and xlsx, Range downloads, folder ZIP, drag-and-drop upload, copy/paste, conflict-checked text editing, create/rename/delete-to-trash, and xlsx cell editing saved as a copy.

Adds upload, download, new folder, delete, drag-to-move, drag-to-chat and in-place text editing with auto-save to the workspace Files tab and file previews.

Available on npm as dsh-doudizhu-lan. Three-player LAN Dou Dizhu (斗地主) played from a floating window: one machine opens a room on its own port, the other two join with a six-character room code plus an address. Empty seats can be filled by computer players in four switchable strengths (easy, medium, hard, hell), so one or two people can still play a whole hand; the deal starts as soon as the three seats are filled. 54 cards in four suits, bidding of 1/2/3 or pass, the landlord takes the three bottom cards and holds 20, play runs clockwise, and the first-bid seat rotates clockwise between hands. The rule engine is pure and covers 11 legal shapes, with big joker > small joker > 2 > A > K down to 3, suits never compared, straights, pair-straights and planes excluding the 2s and the jokers, and a rocket that beats everything. The table itself is built to be read at a glance: three seat cards side by side in play order (you, then your next and previous seats) with role, hand count and a bar that runs out, one line that says whose hand is on the table and what would beat it, the bottom cards as text, an in-place play history that folds into a single summary line and unfolds into every trick including the passes, plus hint and auto-play buttons that pick the cheapest legal answer out of the local rule enumeration. Nicknames can be changed mid-hand (twelve characters, duplicates get a suffix). Computers run inside the authority and are handed only their own hand, the public record and the seat counts, so no hand can leak; the authority is the only place that decides which card ids leave it. Computers are driven by the room itself, so a one-person table keeps playing with nobody else on the LAN port. Humans have no turn timer, the collapsed floating pill can be dragged around the screen, and 179 unit tests pin the rules, the wire format, the theme tokens, the drag gesture, the history grouping and the computer seats.

Host-side long-task detection for small-context local models: long outputs are scaffolded into a parts skeleton (one part per reply, then assemble), and multi-step work into a PLAN.md ledger with notes/ (plan first, one step per reply). Detection is heuristic and host-side, so it does not depend on the model judging its own limits.