플러그인

Agent 行为守护桌宠：跨 agent 监控可疑外传（加密打包直传对象存储 / DNS 外泄线索 / 敏感密钥被读），发现即告警，明确确认后按「一次一个」终止目标 agent；事件写入只追加 guard-events.jsonl，并通过 DSH host 工具 + MCP(stdio) 供其它 agent 查询

Consistency guard for long-form writing: scans conversation prose for character, timeline and numeric setting conflicts across chapters, then jumps straight to the conflicting message.

Layered cross-session memory for DSH that pushes only what changed - the first round injects every active entry, later rounds push only added / updated / removed ones, and nothing at all is injected when nothing changed, with keys that merely repeat an entry id left out of the text so the baseline is about 26% smaller. Facts and decisions carry semantic keys (one active truth per key), the model may only write to an inbox that a human promotes, an entry can carry a review date that raises a one-time reminder when due, and search is relevance-ranked with journal lines down-weighted and Chinese matched by bigram. A sidebar tab tells the lifecycle story instead of listing files - entries are grouped by stage (awaiting confirmation / in use / archived) along four axes (topic - one level, human-assigned, type, tag, date), each group can be ticked for batch promote / withdraw / archive / restore / delete / assign topic, and topics can be renamed store-wide or searched on their own. The archived stage reads in three levels - the category says why an entry left (distilled into the docs / useless for good / not needed right now, the first two being defaults you can add to), below it the entry topic stays collapsed until you click it, and entries with no topic sit right under the category. The tab also shows the global and workspace instruction files plus the current injection size. A zero-dependency CLI ships with it, so the store is usable without DSH.

Declare any number of LLM providers, each with its own endpoint, credential and per-model parameters (reasoning-effort vocabulary, image input, retry policy), across the OpenAI Chat, OpenAI Responses, Anthropic Messages and Google Gemini wire formats, configured from a settings page.

Composer key remap for the DeepSeek Harness web UI: bind Enter, Ctrl/Cmd+Enter, Shift+Enter or any custom chord to the Send, Newline or Interrupt actions, with cross-action migration, durable persistence and native behavior preserved by default.

Large-image recognition for vision-exp models: lossless 800×800 tile recognition (smart/pipeline/full), local OCR with preprocessing & handwriting routing, optional multi-vendor GPU (DirectML/CUDA/OpenVINO) with auto CPU fallback.

dsh 插件：把手机 / 平板 / 第二台电脑上的文件一键投递到会话——插件自带独立端口接收上传（不走 dsh 登录鉴权，只认上传令牌），文件进入输入框正上方的待添加浮窗，点一下即加入输入栏。

Cross-project experience recall for DeepSeek Harness: when the running work matches something your memory store already learned elsewhere, that lesson is injected before the next step. Runs fully locally, judged by a local small model.

Read web novels inside the DSH web GUI: import legado book sources, aggregate search across sources, a bookshelf with reading progress, continuous-scroll reading, and five agent tools for search, chapter reading, source import and probe, and shelf lookup.

Deny an AI agent access to .env files, credential stores, keys and any path you hide — across file tools, shell commands, search selectors and run_code. A DeepSeek Harness (dsh) plugin.

A CLI application manager for discovering, monitoring, and managing installed command-line tools across package managers. DSH plugin: 7 AI-callable tools plus a web dashboard at http://127.0.0.1:3080/app-manager.

NPM-installable DSH Web plugin for safely loading and managing skills across DSH and common local Agents.

DSH plugin: sync model settings, plugin manifests, patch layers and lockfiles across devices through a private Git repository (manual push/pull; https+token or ssh public-key auth).

Cross-platform desktop pet for DeepSeek Harness — a transparent, always-on-top whale that swims on your desktop and reflects agent state. Ships native Tauri binaries for Windows, macOS and Linux.

Unofficial DSH plugin — long-term, cross-project memory for coding agents: an embedded project-portfolio skill plus portfolio_status / portfolio_log tools over plain markdown files the user owns. Not affiliated with DeepSeek. 非官方社区插件，与 DeepSeek 无关联。

Starts a fresh session automatically when the app is opened (page load or tab/PWA foregrounded) 30+ minutes after the last time it was opened - avoids dropping the user back into a stale, possibly unrelated conversation after a real gap.

Global real-time Token HUD for DeepSeek Harness Web: draggable floating window showing live output tok/s, cumulative tokens, and per-session running/retrying status across all DSH sessions.

Teaching board view: agent-generated self-contained HTML is projected into a sandboxed iframe as a board, annotated by hand (brush, four colours, eraser, undo), extended in place across several tool calls, and screenshotted back into the conversation as a real user message.

Persistent memory across DeepSeek Harness (DSH) sessions using Supermemory. A native port of the Claude Code supermemory plugin.

MCP server management panel in Settings: add, edit, enable/disable and delete servers across user and workspace scopes with form or JSON editing, and auto-register discovered MCP tools as native conversation tools.

Peer communication between conversations in one DeepSeek Harness process: list the conversations the workspace shows, link one under a nickname, message it without waiting for its turn to end, read its progress without waking it, and declare rules for your own tool calls.

Turns DSH permissions into 9 selectable presets (3 sandbox modes x 4 approval tiers), with rule plus LLM risk classification across four risk levels, password-gated approval for high-risk operations, and a JSONL audit log.

Genshin Nicole PV wallpaper skin for the DSH web GUI — 14 subtitle-free frames rotate on a timed crossfade queue with pause and manual switching.

Three-party collaboration protocol for dsh: Step List state machine, five-tier review chain (Gemini/web/Claude Code/dialog/manual), alternatives adjudication, cross-restart resume with heartbeat, and capability persistence.