Plugin

Full workbench fork of dsh-better-sidebar: explorer, editor, terminal, Git, browser and split panes, plus a VSCode-like shell refactor with a full-width compact header and activity rail; publishes as @all3cn/dsh-better-sidebar-n23 on npm.

Provider health dashboard as a Settings page: read-only per-provider status, health (time-decayed from recent errors), thinking capability, max tokens and latest error, sourced from the llm-pi-ai config and the compat-healer error log.

Keeps a local model's context at a size the GPU handles well: measures prefill speed on every request and, before each step, warns or compacts early on a hard token ceiling, an observed time-to-first-token budget, or a predicted cold-prefill budget; /context-budget shows the numbers and the cost of compacting now.

Live plugin-topology viewer for DeepSeek Harness: the wiring diagram of every mounted plugin, its runtime activity, a composer-side star map — and a composition workbench that edits the wiring on the graph itself (previewed, validated, backed up, reversib

DeepSeek Harness (DSH) bundle: wires Kimi Computer Use's stdio MCP server through the generic dsh-mcp-client bridge, plus a bundled skill for driving it.

DSH Desktop plugin safety guard: self-repairs plugin load crashes and runs 8 core safety checks (2 optional: deep-config & smoke), prompting you when a plugin changes. Non-commercial source-available.

Use local MLX-LM or MLX-VLM models in DeepSeek Harness through a loopback OpenAI-compatible provider, with optional DSH-owned server startup.

One-click prompt enhancement and structuring button for DSH composer.

Roadmap ideation for DeepSeek Harness — generate and compare distinct roadmap options via headless agent runs, pick one, export as goal.md

dsh plugin: token-efficient prompt-side shim for the official MCP client — folds mcp__* tool schemas out of every assembled prompt and exposes two constant meta-tools (mcp_list / mcp_call) instead

A Live2D pet companion for the DeepSeek Harness web UI: the Vivi (芊芊) character with petting/feeding, a daily-treat bond economy and mood-driven expressions.

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.

Native desktop alerts when dsh needs the human — approvals, questions, plan reviews, turn completion, goals, errors and workflow runs — plus a Settings tab for every option, hot-applied.

Task-completion notifications: desktop alerts via macOS Notification Center and Windows Toast, plus mobile push through Bark, ntfy, ServerChan or any webhook when an agent run goes idle, errors or blocks.

Linux X11 computer use for DeepSeek Harness: 9 tools (screenshot, move/click/drag/scroll mouse, type, key combos) driven by a zero-dependency Python XTest helper — native-pixel coordinates, per-action verification, optional auto-observation screenshot.

Tiered compaction hub: replaces the single 80% compaction cliff with a configurable ladder (default 30/50/70/90) whose tiers escalate from gentle to maximum consolidation laws, folds a deterministic keyword index into every checkpoint so facts accumulate across cycles, and summarizes through a model chain — primary plus up to 3 fallbacks, each with its own reasoning effort. Adds a Compaction module sub-tab and a Settings card; toggling off always restores stock compaction.

Adds confirmed DSH restart actions and durable automatic continuation for eligible top-level sessions through an existing compatible dsh-restart Host provider.

DeepSeek balance and per-session cost in the DSH web composer dock, with official peak/off-peak pricing (peak hours Mon-Fri BJT 9-12 / 14-18, weekends off-peak) and built-in deepseek-v4-flash-vision-exp vision model rates billed at flash price (images billed as tokens, max 384 tokens each).

Add independent opacity controls for the sidebar, message panels and composer in dsh web.

Turns attached files into versioned evidence: `search_documents` builds a private local index (SQLite FTS5 after a startup capability probe, dependency-free JS fallback otherwise) and returns compact evidence blocks carrying an exact coordinate — PDF page, PPTX slide, text/DOCX line range, or quoted XLSX `Sheet!Range` — which `read_document` expands only while the content version still matches. Contiguous CJK runs are indexed as overlapping bigrams and queried as phrases, so word order is preserved; uploaded raster images take the native vision attachment path instead.

Obsidian knowledge-base ingest workflow as a DeepSeek Harness plugin: a bundled skill provider + an obsidian_validate_vault model tool, all declarative and composable.

Prompt optimizer for DeepSeek Harness: a composer tool-row button that rewrites the draft into a structured prompt via the session's selected model, with one-click restore

DeepSeek Harness Web 客户端插件：接管 composer 的【模型与推理等级】座位，在「模型」子列表首位插入一个大小写不敏感的连续子串模糊搜索框（div>input），实时过滤已配置模型；弹框宽度锁定防抖动。

DeepSeek Harness (DSH) bundle: wires Kimi Computer Use's stdio MCP server through the generic dsh-mcp-client bridge, plus a bundled skill for driving it.