dsh-plugin-durable-subagent：崩溃后子代理委托不重跑
dsh-plugin-durable-subagent 是 Tim-ReJet 维护的 DeepSeek Harness 工作流插件。它在 ctx.subagents 注册 durable provider，并暴露 durable_subagent 工具，驱动 pi-durable-subagents，使 Harness 在崩溃前已接受的子代理委托在重试时读回首次结果而非再次执行。安装需先 npm i -g pi-durable-subagents，再用 dsh plugin --profile web add github:Tim-ReJet/dsh-plugin-durable-subagent。工具路径用 exec.callId 推导 request id；provider 可选 unique 或 content 幂等策略。适合需要 durable 子代理编排且关注 checkpoint 重放的 DSH 用户。
Read Moredsh-process-kit：用文件状态机与人签 Gate 约束 AI 协作开发
dsh-process-kit 是 SHADOW-LI0327 维护的 DSH 工作流插件：以 docs/process 为唯一真源，提供 process_init、process_status、gate_summary、process_journal 等工具，在 Gate0/1/2/4 做人签，内置多角色 Subagent 与门禁脚本，仅做读取、渲染、安全追加与脚手架落地。本文介绍定位、安装 target、快速开始与 PROJECT.md 配置要点，并说明 npm 0.5.0 与未发布 0.6.0 的角色差异。
Read Moredsh-session-tree：把 DSH 左侧栏变成可嵌套的会话树
orphiczhou/dsh-session-tree 是 DeepSeek Harness 的客户端 profile bundle，将左侧栏替换为按 workspace 分组的会话树，嵌套展示 subagent 后代并支持一键进入多轮对话。提供四态状态、搜索、归档与行内 Pin/Rename/Fork 等操作，并注册 Host 工具 tree_send，向树内任意可继续 subagent（含兄弟节点）发消息。安装可通过 Plugins 页粘贴 GitHub URL，或执行 dsh plugin add orphiczhou/dsh-session-tree。MIT 许可，安装前请审阅源码与权限。
Read Moredsh-plugin-subagent-model-route：在子智能体目录与会话头显示实际模型路由
DSH 客户端插件 dsh-plugin-subagent-model-route（v0.1.2，MIT）在父会话 subagent catalog 每一行与子会话 header 标注最后一次请求的真实 model route（provider/model · reasoningEffort）。数据来自已有 modelSelection 投影，无新 RPC。通过 priority -1 shadow 官方 UI 两个 slot 实现；安装用 dsh plugin --profile web add，需重启 host。需注意行高增加、上游 UI 改进需自行 re-vendor，以及与 subagent-director 可能重复显示路由。
Read Moredsh-subagent-model-director：为 subagent 按智能分级与并发分派模型
DSH 插件 dsh-subagent-model-director 向父会话注入模型档案（高/中/低智能与并发数）、复杂度判据与分配策略，并提供 subagent_assign_model 做批量并发感知分派。需 DSH≥0.1.7-rc.1，且 subagent 模型须先在全局 Model selection 中允许。含设置页、分派测试与 MIT 开源实现。
Read Moredsh-subagent-delegation: DSH Subagent Delegation and Acceptance Plugin
`dsh-subagent-delegation` is a DeepSeek Harness (DSH) plugin designed to address the clutter and resource waste that can occur when a main Agent handles complex tasks. The plugin uses built-in decision logic to determine when to delegate tasks, such as multi-file work, multiple searches, or independent items, and supports parallel execution and background running of sub-Agents. It also emphasizes verifying results by checking actual file changes and chains of evidence. Developers need to craft task-specific prompts to define sub-Agent behavior, while the plugin integrates through the `cordis.patch.yml` mechanism, effectively improving workflow efficiency and controllability.
Read MoreDeepSeek Harness Plugin: dsh-subagent-route-badges
`dsh-subagent-route-badges` is a DeepSeek Harness debugging plugin that directly displays routing information in the Header Actions area of sub-agent sessions. It does not intercept requests; instead, it reads `request/header` data through a projection mechanism to display the Provider/Model and thinking level (Effort) in the UI layer. Registered in a purely additive manner, it supports rollback, reads only the necessary data, and effectively resolves the issues of fragmented and inconvenient configuration visibility during debugging.
Read Moredsh-subagent-setting: Unified configuration of subagent default model in web settings page
## Introduction If you run multi-subagent task flows in DeepSeek Harness (DSH), you might encounter a specific issue: the default configurations for a subagent's provider, model, and reasoning effort are scattered across configuration files. To switch to a different default model, you must manually edit the YAML file and then verify that every subagent has adopted the new values. `dsh-subagent-setting` brings this to a web settings page: select the default provider, model, and reasoning effort on the interface, and newly created subagents will take effect immediately. You can also choose to apply this to existing subagents.
Read Moredsh-subagent-model-router: Optional Models and Routing Configuration for DeepSeek Harness Subagent
`dsh-subagent-model-router` is a plugin designed for DeepSeek Harness (DSH) to unify the management of model routing, delegation, and background result aggregation for sub-agents. This plugin addresses the issue of scattered configurations caused by the varying model capability requirements for different tasks (such as retrieval, reasoning, and review). Core features include: the `subagent_model` tool restricting model selection to configured aliases; the `wait-for-subagents` tool used to wait for terminal results from background sub-agents; embedding model tags and routing instructions into system prompts to assist orchestrator decision-making; and providing a read-only model directory view as well as namespace-scoped configuration tools. Additionally, it supports manual routing management via a Web settings page and configuration in `~/.dsh/settings.yaml`.
Read MoreClaude Code 2.1.220 Defaultly Switches to Opus 5, Nested Subagent Depth Expanded to 3 Layers
On July 24, 2026, Anthropic released the 2.1.220 series updates for Claude Code: version 2.1.219 switched the default model to Claude Opus 5 (1M context window), and expanded the default nested subagent depth from 1 to 3; version 2.1.218 changed `/code-review` to run as a background subagent. An early August evaluation showed that the gap between Opus 5 and GPT-5.6-Sol on Terminal-Bench 2.1 was less than 0.5 percentage points, and the competition for AI programming agents has shifted to focusing on both models and orchestration layers. This article sorts out the version changes based on the official CHANGELOG, and provides configuration instructions for subagent depth limits, concurrency upper limits and other parameters.
Read MoreVS Code 1.131: At a Glance of Subagent Running Status, Built-in Voice Input and Hybrid Markdown Editor
On July 29, 2026, VS Code 1.131 was released: the Agents window now allows real-time viewing of the models used by Subagents, their running duration, and current tool calls; built-in experimental dictation is now available across Chat, the editor and the terminal; a mixed Markdown editor has been added to the Agents window, supporting in-place editing and executable annotations for Agents. Combined with the Agent Host, Git Worktree and review optimizations introduced in version 1.130, this article sorts out the context and activation methods of the Agent infrastructure updates in July.
Read MoreClaude Opus 5 Released: 1M-Context Window Becomes Default Flagship Model of Claude Code
On July 24, 2026, Anthropic released Claude Opus 5, which features native 1 million-token context window, default thinking mode, and the same pricing as Opus 4.8. For Claude Code v2.1.219 and above, accounts such as Max/API will parse "Default" as Opus 5, while the default setting for Pro tier remains Sonnet 5. This article sorts out the specifications of Opus 5, Fast Mode and Effort levels, as well as the toolchain updates of Claude Code in July including Subagent background running, nested delegation, /code-review background review and /doctor self-check, and provides step-by-step guides for version checking and model switching.
Read More