Plugins

macOS desktop control for DeepSeek Harness: agent operates non-browser apps (CapCut, Photoshop, WPS, native clients) like a human — 14 tools for mouse, keyboard, scrolling, app activation, windows and screenshots, plus a vision closed-loop (desktop_see for image understanding + desktop_locate for UI element coordinates) that reuses dsh existing vision credentials. GUI-session service starts on demand, no login autostart.

Desktop-style UI enhancements for the DSH Web client: adaptive wallpapers, a desktop pet companion with an activity feed and feeding economy, @file workspace references, and the resizable file/preview right panel.

Replay real DeepSeek Harness turns against Standard, Minimal, Anchored, or plugin candidates with frozen request-surface evidence

Multiple chat panels for DeepSeek Harness: view and interact with several agent sessions side by side in a Mission Control grid

Mobile-viewport fixes for the dsh web GUI: overlay drawer sidebar with swipe gestures, icon-only composer with a mode · model · effort status line, centered popovers, and horizontal-overflow/scrollbar centering fixes.

Adaptive Reversible Context (ARC) for DeepSeek Harness — provider-aware, cache-conscious context governance with reversible recovery.

Permission mode between workspace-write and full access: shell commands are checked by deterministic rules and a subagent review; irreversible or system-level actions require explicit approval.

Human-in-the-loop tools (`human_task` / `human_task_ready_check`): pause the agent until the user performs a real-world action or returns an observation (GUI operation, game testing, visual verification, hardware checks), gated by session consent and an AFK presence check, returning a structured JSON result via a web task dialog.

A DeepSeek Harness plugin that interviews the user about a project, generates a project timeline / Gantt chart, and exports it as Word or Excel.

Recover deleted or corrupted dsh sessions (session.jsonl.zstd) and memory (memory.db) from the raw disk, and repair rebuilt sessions that fail to resume via the /session-repair command in the web UI.

Ranked subagent delegation: role-based model routing with failover, per-route reasoning effort, model-transparent labels, fork delegation, and a visual route editor.

Tiered model routing for DeepSeek Harness: think/build model tiers, vision delegation (the vision model only describes images; the working model acts), image generation, and a high-impact guard.

Floating workspace file-preview window for the Web client: lazy file tree, CodeMirror preview and in-window edit with save, Markdown and image preview, Quick Open search, and opening paths clicked in the conversation.

Adds a "History" conversation view tab listing the session's user messages as a clickable directory, with auto-paging and click-to-jump back into the chat.

Desktop profile for dsh: native Tauri window over the harness web surface with tray, native notifications, a file-manager panel, and one-click updates.

Select text in a finished assistant reply to add numbered comments beside the quote; collect drafts and submit the batch through the official composer as one user message.

Message timeline rail for DeepSeek Harness web chat: evenly spaced ticks for each user message on the right edge, click to jump, hover to preview.

Type normally; about two seconds after you pause, it rewrites your draft into a clearer, first-person version more likely to get a good answer, leaving pasted code, links and long text untouched, marking missing details as placeholders instead of guessing, and keeping multi-part questions numbered.

Composer GitHub picker: click the GitHub icon at the bottom-right of the input box to search the workspace repository's issues and pull requests via the gh CLI, and insert the reference as a URL or @owner/repo#number mention.

Detached-broker supervision of long-running services (dev servers, watchers, mock APIs) that survive turns, sessions and DSH restarts, with a Service sidebar tab (registered via better-sidebar's extension API) and service_start/stop/restart/list/logs model tools.

Falsification-driven correction loop for agent sessions: risky work commits an evidence-checkable claim first, deterministic gates verify it, falsified claims force mutually exclusive replacement hypotheses with discriminating experiments, and every event lands in an append-only evidence ledger.

Transactional data-only configuration pack manager with integrity, ownership, preview, diff, and rollback safeguards.

Draggable whale companion with local progression, achievements, skins, and privacy-safe activity tracking.

Per-file-type multimodal chains: preset processing models per wildcard convert image/video/audio files into prompt tokens before they reach the text-only session model, with per-preset fallback chains and a Multimodal settings page.