Agent 連携 · DeepSeek Harness プラグイン — 132 ページ

Project explorer, file workbench, context dock, and conversation history for DeepSeek Harness

DeepSeek Harness 浠ｇ爜鍏ㄩ潰妫€鏌ユ彃浠讹細涓夋娴佹按绾匡紙缂栬瘧杩愯妫€鏌?鈫?鍔熻兘瀹屾暣鎬ф牳瀵?鈫?鐪熷疄鐢ㄦ埛妯℃嫙娴嬭瘯锛夛紝鑷姩鎶婇棶棰樺弽棣堢粰 AI 淇锛屽彲閫夌敤 GUI锛屽苟闄勫甫閫傜敤浜?Trae/Qoder 绛夊钩鍙扮殑鐙珛 CLI 涓?MCP 鏈嶅姟

Semantic repo index for DeepSeek Harness — tree-sitter symbol index, code search, and a bounded ranked repo map injected into agent context.

Code reference & reuse-survey plugin for DeepSeek Harness (DSH): search local code + GitHub/npm before building, evaluate reuse vs rewrite, ask the user, architecture-level reuse. 开发前检索本地与开源可复用代码并询问用户的复用调查插件。

OpenAI Codex CLI integration for DeepSeek Harness: run one-shot codex tasks, repo reviews and session resumes as native dsh tools, with read-only sandbox by default.

Community DeepSeek Harness bundle for Codex account-pool integration

Multi-agent collaboration shared workspace for DSH (DSH × Claude Code × Kimi Code × DevEco Code): outbox/status/tasks/knowledge/artifacts protocol as a cross-platform DSH plugin + CLI.

DeepSeek Harness (DSH) plugin that packages a distilled bilingual guide to ComfyUI custom node development as the comfyui-node-dev skill, ready for any DSH agent to load.

DSH plugin: publish a Command Code subscription (Go / GOAT / Pro / Max) as llm-pi-ai provider routes, with account usage and web search on the same credential.

Command Code plan quota panel for DeepSeek Harness: monthly / 5-hour / weekly credit windows with reset countdowns, above the sidebar Settings seat.

Commander for the DeepSeek Harness Web GUI: activate any conversation as the commander, and its model can dispatch tasks to other conversations via <dsh-dispatch> protocol blocks. The plugin parses the blocks, delivers each task to a target (or freshly created) worker session, watches the workers, and feeds result summaries back into the commander as receipts so it keeps orchestrating.

DeepSeek Harness 会话上下文强制压缩插件:模型可调用的 compact_agents 工具,忽略自动压力阈值,覆盖进程内所有活会话(主会话/子代理/AgentTeams 成员),忙则排队、本轮结束补压,逐目标回报遮蔽节点数与估算 token 数。Session-context force-compaction plugin for DeepSeek Harness: a model-callable compact_agents tool that ignores the automatic pressure threshold and covers every live session in the process (main session, sub-agents, AgentTeams members alike), queues a busy target and compacts it when its turn ends, and reports shadowed node counts and estimated tokens per target.

元压缩：模型自己决定何时、如何压缩自己的上下文——列 surface、选区间、以自写文本替换；替换走官方 compaction 事务（检查点、配对平衡、可重建）。Meta-compaction: the model directs its own context compaction through the official compaction seam.

Per-session automatic compaction threshold for DeepSeek Harness: a compaction backend whose trigger percentage is stored per session and adjustable from the Web composer while the session runs.

DSH 视觉整机控制插件（P2）：经 macOS Accessibility API（AXUIElement）把本机 GUI app 变成 AX 快照优先的元素级工具集（观察 4 + 窗口管理 1 + 动作 6 + 截图兜底 1，共 12 个 computer_* 工具）；v0.2.0 窗口对象化：windowId 不透明句柄（WindowRegistry + tombstone + CG 截图绑定）+ 输入后端分层（mode/delivery，Tier 1 pid 定向预留）；Swift 辅助进程零 npm 依赖；敏感动作可选人工审批（默认关）。

Model-agnostic Computer Use capability for DeepSeek Harness: isolated browser first, provider-neutral observation, text planner + third-party vision perception.

Text-first browser and background macOS computer use for DSH

All-in-one DeepSeek Harness (DSH) chat bridge: one plugin, one config block — Feishu/Lark, Telegram, DingTalk, and Web mirror adapters plus the web-settings pane, all behind a channels[] selector

Connect the locally signed-in WPS Comate account to DeepSeek Harness as a model provider, via a secure loopback shim (reverse-proxy style). 通过安全回环 shim（反向代理），把本机已登录的 WPS Comate 账号接入 DeepSeek Harness 作为模型来源。

自动上下文策展：tool 输出瘦身 + 无效日志清理 + 工作流/记忆钉扎保护 + 先归档后压缩

Windowed, reversible context management for DeepSeek Harness; replaces Basic compaction.

Model-driven context-window rollover bundle for the DeepSeek Harness: fresh working context, durable model-managed notes, a small verbatim recent tail, and targeted history recovery — no summarization.

Fail-closed network recovery, recommended answers, and safe unattended continuation for DeepSeek Harness.

Persistent project memory, context-safe agent handoffs, checkpointing, evidence tracking, and long-running task continuity for DeepSeek Harness