dsh-insert-context：按步数向智能体上下文注入提醒
dsh-insert-context 是 ADkun 维护的 DeepSeek Harness 工作流插件（v1.5.0，MIT）。在 agent/pre-step 上按规则从第 X 步起每 Y 步向智能体上下文注入自定义消息（repeat 控制次数，0 为无限），text 支持 {{step}}；可选在用户消息末尾追加一句话，带总开关 enabled 与 appendFirstOnly、appendSubagents 等选项，均在 Settings→过程插入 配置。步数按智能体独立计数，关闭总开关时计数暂停。安装示例：dsh plugin add 指向插件包目录。
Read Moredsh-reveal-context：在对话里重新显示被隐藏的上下文注入行
DSH 对话视图默认不渲染大部分 context 注入行，调试时只能看轨迹。dsh-reveal-context 是 azazo1 维护的客户端插件，通过额外 conversation 定义与 reveal-context 节点在不改内核的前提下把这些行画回 Chat，并提供总开关与按 source.kind 排除。Web 安装：dsh plugin --profile web add azazo1/dsh-reveal-context；需 DSH 0.2.0-rc.1 以上 0.2.x。MIT 许可。
Read Moredsh-agent-chat：两台机器上的 DSH Agent 跨机消息与上下文注入
DSH Agent Chat 插件通过 HTTP 提供了跨机消息通道，解决 DSH Agent 非常驻进程导致上下文传递困难的问题。该插件基于 Node.js，利用“落盘-轮询-注入”机制，使两台机器的 Agent 能互相留言并在回合开始时拼入上下文。配置项包括对端地址、轮询间隔、token 认证及日志路径等。插件仅依赖 Node 内置模块，支持 LAN 部署，且 `/agent-chat/send` 接口安全性高度依赖 Token 配置。
Read Moredsh-plugin-context-trim: DeepSeek Harness Session-Level Context Injection Gating
The DeepSeek Harness plugin `dsh-plugin-context-trim` addresses the context redundancy caused by indiscriminate injection of plugin capabilities across sessions. By leveraging a "shadow entry" mechanism, it allows users to trim Skills, Tools, MCP tools, and prompt sections on a per-session basis via the funnel button, with no residual artifacts and no cross-session impact. Features include batch management, state protection, hot activation, and default template saving, aiming to optimize context window utilization.
Read Moredsh-projects-mode: DeepSeek Harness Projects Workspace Mode
`dsh-projects-mode` is a client plugin built for DeepSeek Harness (DSH), designed to address context confusion and token consumption in multi-session parallel development through the "everything is a plugin" philosophy. Core features include a full-screen project view for unified session management; real-time injection and cross-session refresh of instructions and memory; automatic generation of new-session briefs containing session summaries to reduce token usage; automatic working directory ownership; and token metering with security hardening. The plugin resolves issues such as unclear session ownership, delayed instruction updates, and high token consumption from long conversation histories. It is suitable for developers who need long-term, multi-session collaboration in DSH.
Read MoreDeepSeek Harness Plugin: dsh-prompt-injector (General Per-Turn Context Injection)
`dsh-prompt-injector` is a plugin for DeepSeek Harness (DSH) designed to address the issue of rules being forgotten in long sessions. It leverages the memory mechanism, allows prompt lists to be managed via a UI, and injects context in the form of a “Context injection” notification line for each conversation (or after context compaction), ensuring that the model always “sees” critical rules. The plugin’s core features include per-turn injection, selectable trigger modes, abbreviation filtering, a zero-intrusion design, and UI-based management. It is responsible only for injecting text into the context and does not execute any actions, making it suitable for repeatedly emphasizing specific constraints in coding or complex Q&A scenarios.
Read Moredsh-layered-memory: DeepSeek Harness L0-L3 Layered Distillation Memory Plugin
`dsh-layered-memory` is a memory plugin designed for DeepSeek Harness, aimed at addressing issues such as context loss and repetitive questioning in long conversations and cross-session interactions. It automatically executes background processes including L0 raw conversation capture, L1 atomic memory extraction, L2 scene block integration, and L3 core profile distillation to transform conversational information into reusable memory assets. Before model execution, the plugin automatically recalls relevant memories based on length and time budgets, injecting them into the context to avoid slowing down the conversation. The plugin offers three proactive query tools, such as `memory_search`, supports four session-level memory tiers—daily, work, smart, and closed—and features a "write-only, no-read" toggle, same-session deduplication, automatic context compression resets, and time-based weighted recall functionality. The LLM token costs incurred during the distillation process are recorded.
Read More