Plugins

A Conductor tab and commands for a chat's context: /new replaces the context in place with a model-written handoff, /rewind restores the chat and the agent's file edits to an earlier prompt, plus /drop, pins, an automatic compaction level, recovery after failed turns, silent messages between chats and agent and task orchestration. Supersedes dsh-orchestrator, dsh-force-compact and dsh-continue, so do not install it alongside them. Requires DSH >= 0.2.1-rc.5, Node >= 22.5, the web profile and dsh-better-sidebar.

DSH → MAX messenger plugin: compact UX, briefings, status, reliable replies, remote control, GUI settings, reply threads, tables, schedules, webhook, cost & budget. Zero runtime deps.

WSL World for DeepSeek Harness: the whole session runs in a local WSL distribution with Linux paths, clean PATH, native @/search/diffs/worktrees, plus a connection panel and project doctor. Not a wsl_* tool set.

Serial-port toolkit for DSH - 9 tools plus a web UI where an AI agent and a human share one UART line, with auto port matching, hex/text frames, timestamps, scheduled sends, rule-based auto-recovery and appearance themes. The serial engine runs on Python 3.8+ — DSH's bundled interpreter is used when present, otherwise python / py -3 on PATH.

Before closing a coding session or the DSH app itself, checks what would be lost — uncommitted work, unpushed commits, scratch files, missing handoff notes, pending restarts, broken profile bundles — and prints a three-level verdict; when a handoff is needed it writes the seven-section handoff file into the workspace ledger. Read-only by default.

Appends one short user message before the next step when the model is circling: the same absolute path written three times with no measurement landing in between, the same error episode (tool, target and normalised fingerprint) returning twice, or a write followed by two steps with no measurement — the last one only arms, and the message lands before the next write rather than interrupting reading. A measurement has two states: intent is judged from the tool arguments at call time, and the write streak is cleared only when the result comes back, only for that command directory. A failed test still counts, because it yields the line that is wrong. Hard measurements are `lsp_diagnostics`, `verify`, `run_tests` and `run_test`; `ops_bash`, `bash`, `shell` and `pwsh` count only when the command matches a measurement pattern, so `ls` no longer launders a blind edit streak. Signals fire once per episode, re-arm after a real measurement, and are capped per turn; a successful `present` sleeps the gate until the next write, and subagent sessions are skipped. Generated paths (build output, lock files, logs) are excluded. Error targets are read out of real `file:///` ESM stacks, so two different files never collapse into one episode, and `cd <dir> && cmd` counts as a measurement of that directory rather than its parent. Ships the `stuck-gate` skill at bundled rank 600 (a user-level copy at 400 overrides its wording) plus a read-only observation panel at `/better-debug` reporting firings, sessions covered, the first action after each message across six classes, and how many measurement loops actually closed. The log defaults to `$DSH_HOME/logs/better-debug.log` and is created on first write, so the panel has data right after install. Optional host services are subscribed dynamically, so a host without them still loads the gate. It does not estimate tokens saved, and it does not claim the message caused the change. The panel banner counts only the failures after the most recent load (`log.current`), so one old mount failure no longer reads as "currently disabled". Test suites wait for log conditions instead of fixed delays (test-only change in 0.2.5).

Reusable workflows: orchestration scripts saved as files under .dsh/workflows, discovered from the project, and run by name with arguments, so they live in git and can be shared. Four tools: list, show, run, save.

DSH 本地记忆插件：自动记忆、遗忘曲线、关键词关联、可检索可编辑，数据留在本机。

Native macOS tools for DeepSeek Harness agents, with 21 local tools for notifications, clipboard, screenshots, speech, keep-awake, dark mode, Spotlight, Shortcuts, AppleScript and Safari tabs. Zero dependencies.

Bridges the shared ~/.agentloom store between DeepSeek Harness and Codex: injects the shared memories and the AGENTS.md entries the context budget truncates into the system prompt, and adds six loom_* tools for status, sync, memory search/list/add and skill listing. Requires the agentloom CLI installed and a one-time agentloom init; the seven-tool MCP bridge in the same repo is a separate package for MCP clients.

Gives the model in DeepSeek Harness a real new_conversation tool: creates a session in the current workspace, delivers a self-contained handoff as its first user message, optionally attaches a read-only dsh-session: reference to the calling conversation, pins the title, and switches the Web UI to the new session.

Intercepts DSH tool results before the model reads them: dedupe cache for repeated calls, a declarative decision graph that replaces large outputs with structured digests, an injection gate, and an honest token ledger that includes a negative-return account for over-compression. Measured 1400-1600 tokens saved per hit; the cumulative ledger is currently negative and the README says why, including that it cannot skip model turns.

Records a session as a run, replays it in an isolated workspace copy with another model or prompt, diffs the two trajectories side by side, aggregates runs into a task × variant matrix, and gates regressions with task assertions and a CI exit code.

Control tower tools: dispatch a task to a role agent, read its report, verify and accept or send it back. Requires the separate agent-control-tower Python CLI (tower; it is not on PyPI — install it with pip install -e from the same repo) on the DSH process's PATH plus a tower init-ed workspace; without it every tool call fails with ENOENT.

Windows desktop control for DeepSeek Harness: accessibility-tree snapshots, click by control name, delivery verdicts, and repeatable task files.

Adds lasting bots with their own name, photo, persona, memory, voice and model, a Bots panel in the sidebar, and a hidden helper subagent per bot that follows the conversation, searches ahead and leaves notes the bot reads; voice uses your own ElevenLabs, OpenAI, Groq, Mistral or Deepgram key. Interface and prompts in Brazilian Portuguese.

Project spaces, group chats, task boards, memory and usage for teams of DSH bots

本插件仅供娱乐使用，非实用性插件，下载前请自行选择。DeepSeek Harness 非官方社区插件（Token-Reducer?）：模拟执行任务，生成 SIMULATED 轨迹与 0 字节产物，不调用真实模型或工具/For entertainment only. Not a practical plugin — download at your own discretion.Unofficial DeepSeek Harness plugin that simulates task execution with SIMULATED traces and zero-byte artifacts.

Unofficial DeepSeek Harness port of the Academic Research Skills plugin (CC-BY-NC-4.0, non-commercial): 5 skills, 16 /ars-* commands, 4 agent roles, write-scope guard, and deterministic Python gates.

Tool-execution efficiency suite for DeepSeek Harness: read-only result cache, speculative prefetch, and concurrent readonly shell commands. Honest scope: helps when tools are slow and calls repeat; model latency dominates otherwise.

Registers the hosted SpecStream MCP server with the harness MCP manager and keeps the API key in the credential store. The server returns live web data as structured rows, each value with its source URL, and a run starts only after the quote's maximum is approved.

Fixes web_fetch failing with "non-public IP address" while Clash, Surge or v2rayN runs in TUN/fake-IP mode: finds your local HTTP proxy by itself, sends web_fetch through it while the proxy is on and directly when it is off, and adds an Auto/Proxy/Direct switch to the sidebar. Model calls are left alone by default.

内置 zk 的 DSH 知识库 / 日记插件：装好即用 —— 自动装 zk、建笔记本，给 agent 一组 zk_* 工具（搜索、读写、日记、双链），设置页里管路径、索引与自检。

Sidebar node canvas for DSH where each node is an agent task and each link a dataflow dependency. One deterministic import (no model calls) produces both a task pipeline and a file-level module graph; file nodes carry a coupling score, and every run is snapshotted so it can be undone as a whole.