Плагины

Keeps DSH agents from forgetting your requirements during long tasks. It saves important conditions and completion evidence locally, brings them back after context compaction or session resume, and stops partial work from being reported as the whole task done.

Auto-switch the DeepSeek route to the vision model on demand: flash main session switches (A), pro keeps deep reasoning and delegates image reading to a vision subagent (B), subagents always switch, with fatal-failure fallback. No manual model switching.

Redirects DSH downloads, artifacts, caches and temp files to user-specified drives (auto-picks the largest non-C drive), injects a path-discipline prompt into every session, and ships a disk_guard tool for status and cleanup — keeps the C: drive from filling up.

Manage multiple API keys per provider in Settings → API 密钥: keep a pool of keys with notes, switch the active key with one click, and the switch really takes effect on the next model call.

Collapses each settled turn's process rows into one expandable disclosure in the DeepSeek Harness Web conversation while keeping the final answer visible.

补丁包管理（DSH 插件，hybrid）：把你在第三方插件/项目上的自定义修改提取为补丁包，官方更新时检测并提示，用 AI 把补丁合并到新版本（冲突自动交 AI 解决），再生成新补丁包。含 Web GUI 管理面板（设置「补丁包」分区，/patch-keeper/api）。纯 JS 零依赖。

Holds the OS wake lock while any agent, subagent, or background job runs, releases it after a configurable grace period, and provides a bilingual web settings page with live status, manual hold, and display-on options.

Windows process-tree reconciler for the DSH host: samples the host process tree and the job ledger, attributes each process to the job that created it, flags unattributed and orphaned processes with a confidence level, and offers guarded termination of a whole tree. Termination is refused unless the target belongs to the DSH host tree, is re-sampled and re-checked immediately before the kill (creation time included), and contains no protected descendant, so an allowlisted PID cannot widen the kill scope.

Manual context compaction for DeepSeek Harness: keep the latest N conversation items with stop or batch mode and a per-session compaction archive — embedded in the context meter popup when the meter panel slot exists, or a composer button otherwise.

Manual context compaction for DeepSeek Harness: keep the latest N conversation items with stop or batch mode and a per-session compaction archive — embedded in the context meter popup when the meter panel slot exists, or a composer button otherwise.

Display companion for dsh-vision-router: keep Auto Vision routing, and put a square picture button on each original model row. Does nothing unless dsh-vision-router is installed.

Memory triage for DeepSeek Harness: at the end of each turn it decides which entries to keep and which to archive, routes project-specific ones into a project bucket while global ones stay in the main store, merges near-duplicate entries, and falls back to an LLM when the rules are unsure — automatic actions are recoverable archives only, never deletions.

ego lite for DeepSeek Harness, with a memory it keeps: the agent recalls what it learned about a site before it acts, promotes a script that worked into a reusable site tool, and hands you the keyboard through a real dsh prompt when a page needs a human.

Conversational Markdown mindmap workspace: edit complete documents step by step, keep a live right-side visualization synchronized, and open, manage, and export mindmaps from the working directory.

Login gate for the DeepSeek Harness web GUI: username/password entry page (DeepSeek style), 12-hour session tokens, slash commands for user management. Background tasks keep running while logged out.

dsh-awake · 守夜人：在 agent 任务执行期间阻止操作系统休眠

Adds a Worktree view to DSH Web UI that groups Sessions by Git worktree while keeping DSH as the source of truth.

Prevent the computer from sleeping while Harness is running. Toggle at runtime through a model-facing tool.

Hermes-style self-evolution for DeepSeek Harness: a skill_manage tool that lets the agent save proven procedures as skills, prompt guidance for when to save them, usage tracking, and a background curator that keeps the self-authored skill library healthy

A DeepL-style translation surface for dsh: replaces the shell's root frame with a two-pane translator driven by a tool-less agent preset, keeping the shipped settings panel (models, providers, plugins) intact

Minimal-anchored agent preset: request #1 keeps the real bash + str_replace_editor pair with auto-injected context stripped, then promotion unlocks the full Standard catalog and posts a one-shot background-jobs notice against sleep-polling habits.

Collapsible right-docked web terminal (xterm.js) bridged to a host PTY over WebSocket: bash/zsh switching, resizable panel, and fold/unfold that keeps the session alive.

LLM-driven remember/forget memory: after each turn a model decides what to keep (MemoryDirector), injects relevant memories before each step, dedups, and persists cross-session.

Compaction backend replacing LLM summarization with a deterministic semantic extractor (keeps code/paths/commands, drops chatter) plus 28.4x KV-compression accounting.