Плагины

Radial right-click menu on conversation turns in the DSH web GUI for forking a session from a chosen turn, interrupting generation, copying the working directory path or session ID, opening the working directory, and exporting session logs.

Read-only MCP server over the DSH plugin-certification registry, exposing a plugin certification grade, its snapshot and the five-dimension evidence behind it.

DSH plugin: browse local Codex projects & sessions and continue them in DSH — read ~/.codex rollouts, pick a session, resume with full context (goal + transcript + cwd).

Adds per-route context window and compaction threshold fields to the Models settings page and a vendor-aware thinking-effort picker to the composer, writing llm-pi-ai capacities, reasoningEfforts, and a compaction-basic patch row.

Adds a right-click menu to DeepSeek Harness inside desktop shells that ship no context menu: cut, copy, paste and select all in the composer, copy and copy-as-plain-text on selected text elsewhere, and a search entry that opens the default browser with a configurable engine.

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.

dsh (DeepSeek Harness) plugins for project-level persistent context: session archive with a mechanical index, memory consolidation (CONTEXT.md + MEMORY.md), low-frequency skill autolearn, and automatic handoff.

Keeps notes as local Markdown files grouped by directory under a root folder you choose, with a sidebar page, AGENTS-style rule files surfaced as writing context, and six agent tools.

Measures what an llm-pi-ai endpoint really accepts — context window from its own model list, output-token limit and reasoning levels read out of its own range errors, image support from a generated test image — then reports declared values that contradict that evidence and rewrites them only after confirmation.

DSH 实时库文档：写代码前自动拉取项目安装版本的官方文档，消灭幻觉 API。零配置零 Key。Version-pinned live library docs for DeepSeek Harness — kill hallucinated APIs

会话注入门控：给 DeepSeek Harness (dsh) 加一个会话级上下文裁剪闸门，按会话取消/恢复插件贡献的 skill、tool、系统提示词段落——3 条影子通道、5 条本地路由、零依赖零配置。Per-session injection gate for DeepSeek Harness (dsh): trim skills, tools and prompt sections per session. Zero-config, zero dependencies.

MULTIVERS engine for DeepSeek Harness: Zero-Bloat Payload Diet, Context Synthesizer, Flatline Rolling Window, and Settings Dashboard.

Memory plugin with built-in governance for DeepSeek Harness

CWL — Context Window Lifecycle for DeepSeek Harness: structured context eviction (arXiv:2606.11213). Graduated, deterministic, zero-LLM eviction of exploration/action episodes when context pressure exceeds budget — no summarization lossiness, no hallucina

dsh-custom-context-injector 提供「自定义上下文」设置页并注入到模型上下文，支持落盘持久化。

Integrated memory and context management plugin: four-layer memory system (L2-L4), context compression/trimming/freezing, prompt orchestration, input box UI buttons, and enterprise security with role-based access control.

PowerContext for DeepSeek Harness: WorkBuddy-style project memory and inspectable handoff workflows - bounded context recall before each model step, 19 pc_* tools for memory, handoff, experience and skills, with a Windows server watchdog for resident operation.

Multi-agent frugality defense plugin: read-ledger dedup metrics, compaction-immune rules, completion gate, cheap-model review lane.

Git sidebar panel with commit graphs, ahead/behind tracking, branch operations, and one-click chat context injection.

Scoped persistent memory for DSH: memory_save/search/replace/remove/list across global, user, project and failure scopes, DSH storage-domain persistence with secret and prompt-injection scanning, session-scope search, retention cleanup, and optional session-start injection; standing context via memory_pin/memory_pins/memory_unpin.

Upload any file to DSH, isolated per workspace and session — a composer file library with server-side thumbnails and a context menu to open, copy, re-download or delete.

Explains which layer killed a failed turn: times every model request (first byte, longest silence, total) and combines that with the error to name the culprit (dsh idle watchdog vs Node undici timers vs SDK timer vs server crash, context overflow, gate queue, tool timeouts, failed compactions) and the exact setting to change; verdicts in the DSH terminal plus a /why command, subagents included.

Context-window attribution for DSH Web: a Mini Utility Dock panel that ranks the user messages, assistant turns, reasoning blocks and tool results of a session by the tokens each one occupies, so it is clear which entries hold the window. It also shows the share by entry type, the heaviest entries across the live sessions of the host, and, where the host provides them, provider usage and the estimated system/tools/messages composition. The panel is read-only: it does not estimate cost, modify a session, or trigger compaction.

Task Runner mode for DeepSeek Harness: the main agent decomposes large tasks into independent subtasks, dispatches them to subagents (default local model) in fresh short contexts, and synthesizes results; concurrency and worker/orchestrator models are configurable in the settings panel.