Agent 連携 · DeepSeek Harness プラグイン — 110 ページ

The dsh one-shot bundle: a direct core Agent/Session runner over dsh-base with no Host, HTTP, or browser layer

The dsh one-shot bundle: a direct core Agent/Session runner over dsh-base with no Host, HTTP, or browser layer

The dsh one-shot bundle: a direct core Agent/Session runner over dsh-base with no Host, HTTP, or browser layer

Coordinated boot-layer configuration reloads: one exclusive queue, consecutive-change merging, failure broadcast, and disposal drain over chokidar exact-path watching

Workspace file browsing Remote: one-level list with metadata, text preview reads, and stat primitives for the web GUI file explorer

Feishu (Lark) IM bot channel for DeepSeek Harness agents — per-chat sessions, native thinking process, interactive approval cards, WebSocket long connection.

ACP LLM adapter + ACP Servers settings UI: drives external Agent Client Protocol servers as model providers on the DeepSeek Harness LLM seam, with a web settings page for browsing the ACP registry and managing configured servers

本地 LM Studio 视觉识别插件：连接 LM Studio OpenAI 兼容接口（默认 http://127.0.0.1:1234），用视觉模型（自动使用最后加载的模型，不预设具体模型）识别本地图片与剪贴板图片。agent 工具 lmstudio_vision / lmstudio_clipboard / lmstudio_models + 设置页「LM Studio 视觉识别」面板；配置持久化到 ~/.dsh/dsh-lmstudio-vision.json。Hot-pluggable — 挂载方式同 dsh-ssh：profile node_modules link + dsh.profile.bundles，无需修改 dsh 源码。

macOS desktop-control capability seam (ctx.macosUse): include-only foreground-window screenshots, exact-window observations, guarded native GUI actions, and exact app launch

Memory provider registry for self-evolution

Local layered memory provider

Host-plane memory service for DeepSeek Harness backed by a local OpenViking server: automatic session capture (write-behind with retry + dead-letter), session-scoped recall, and profile injection primitives.

Deployment-level Docker plugin for DeepSeek Harness (DSH): a 容器 (Containers) tab next to chat/trajectory with observe/intervene two-mode UI, 18 model tools (docker_info/ps/images/pull/run/start/stop/restart/rm/rmi/logs/exec/inspect/stats/network/volume/build/cli), read-only live shell watching, and /dock-api/* HTTP routes served via webServer.

Popper: a falsification-driven correction loop for agent sessions — claim commitment, gate falsification, and mutually exclusive hypothesis revision, with an append-only evidence ledger

Qwen-MM-Plugins integration bundle for DeepSeek Harness: bundled multimodal skills, MCP server rows, and an image attachment bridge for text-only model routes

The dsh SDK profile bundle: stdio JSON-RPC serving and process lifecycle over dsh-base

The standalone minimal SDK profile bundle: JSON-RPC, one DeepSeek adapter, persistent shell, editor, and JSONL sessions

Skill usage telemetry service

Remote SSH operations for the dsh web GUI: host config store (~/.dsh/dsh-ssh.json, import from ~/.ssh/config), a persistent ssh2 connection pool with jump-host support, exec / PTY web terminal / SFTP transfer / local port-forward tunnels / cluster execution, plus agent tools (ssh_list, ssh_exec, ssh_upload, ssh_download, ssh_tunnel, ssh_cluster). Hot-pluggable — mounted via ~/.dsh/cordis.patch.yml + a profile node_modules symlink, no dsh source changes.

Governance-backed SupervisorExecutorProvider adapter (CDH Phase A bridge)

System-level task notifications for DeepSeek Harness (DSH): approval requests, awaited replies, and finished background jobs / subagents / workflows — pushed to the OS notification center with chimes, plus an in-app bell panel, toasts, and a clear button.

Model-facing consumer of the ctx.macosUse / ctx.guiModel / ctx.browserUse capability seams: bounded desktop and browser primitives plus vision-model-driven computer_use_task and browser_use_task tools

Model-facing memory tool and prompt context

Agent-preset tool package for the host `memory` service: thin memory tools plus an automatic compact user-profile prompt section and per-input targeted injection.