dsh-plugin-debug-mode：在 DSH 里复刻 Cursor 式 Debug 模式
dsh-plugin-debug-mode 是 rh666666 维护的 DSH 客户端插件，通过 /debug 斜杠命令进入调试模式：Agent 在代码中埋可摘除探针，经宿主 HTTP 通道写 JSONL 日志，用户复现后在面板选择结果，Agent 读摘要再修复或继续轮次。流程不依赖 Agent 执行 shell。本文基于官方 README 介绍安装命令、五个 dsh_debug_* 信号、改代码轮次规则、产物目录与适用注意。
Read Moredsh-step-trace：记录智能体每一步并回放排查
dsh-step-trace 是 hj01857655 发布的 DSH 插件，将模型回复与工具调用追加写入 .replay/.jsonl，支持逐步查看、从某步分叉会话、对比两路运行的分歧点，并提供 CLI 与 Settings 内 Step trace 面板。独立 GitHub 仓库已归档并曾并入 dsh-plugins monorepo；npm 仍可安装。适用多步 Agent 调试与重试对比，安装前需知插件以 dsh 进程权限读写本地 replay 数据。
Read MorePreset Node Graph: Visualizing Agent Presets in ComfyUI Style
`agent-preset-graph-xww` is a visualization plugin for DeepSeek Harness that turns abstract agent preset manifests into ComfyUI-style node graphs. Maintained by laojingwei, it runs in the web frontend and does not modify host state. It displays an agent’s composition, real-time execution status, and various capability blocks, and supports features such as node search, detail viewing, automatic execution following, and image export. The plugin is compatible with DSH ≥ 0.1.7-rc.2, requires specific services, and helps developers monitor agent behavior and organize preset configurations.
Read Moredsh-plugin-session-export: DeepSeek Harness Session Black Box Plugin
dsh-plugin-session-export is a debugging plugin for DeepSeek Harness (DSH) that captures comprehensive session events via bypass monitoring without interfering with Agent execution. Installation requires registering dependencies via a local link. It supports one-click generation of structured retrospective reports (in Markdown or HTML format) that include session overviews, tool invocation statistics, and error summaries. Featuring sensitive data masking, real-time statistics, hot configuration updates, and a web-based dashboard, the plugin is ideal for debugging and retrospective analysis scenarios for Agent developers.
Read Moredsh-watcher: Read-Only Session Insights and Statistics Plugin for DeepSeek Harness
`dsh-watcher` is a read-only plugin purpose-built for DeepSeek Harness, designed to provide developers with session insights, work path observation, and cost estimation. It visualizes Agent execution steps, parallel branches, and tool results using collapsible views, supports session aggregation and cost statistics, and does not intrude into the conversational flow. It requires DSH 0.1.7-rc and Node.js 22/24+, and can be used by adding `github:aa2246740/dsh-watcher#v0.6.1` through the desktop app or Web CLI, helping developers clearly analyze workflows and monitor model invocation costs.
Read Moredsh-agent-run-logger: DeepSeek Harness Local JSONL Tracing and Viewer
dsh-agent-run-logger is a plugin for DeepSeek Harness that converts official session logs into fine-grained local JSONL trace files and provides a real-time visual viewer. It records agent runs, model steps, tool call durations, and token usage, with waterfall chart visualization support. It captures non-sensitive metrics by default, while content capture must be manually enabled. Developers can start the local viewer via an `npx` command to debug agent execution details and performance issues, but the viewer is limited to local access and does not replace official logs.
Read MoreDeepSeek Harness Plugin: satan9394/dsh-debug-recovery (Debugging and Error Recovery Skill)
`satan9394/dsh-debug-recovery` is a debugging and error recovery plugin designed specifically for DSH developers. Its core design philosophy is “evidence first,” aiming to address the lack of a unified tax on and evidence collection mechanism in agent debugging. The plugin provides error classification, recovery strategy formulation, and regression prevention mechanisms, helping developers quickly locate and fix issues from chaotic logs. Since the official documentation does not provide specific installation instructions, it is recommended to consult the source code repository directly. Users should pay attention to permission security and perform source code reviews before use.
Read Moremkaliezz/dsh-action-ledger: DeepSeek Harness Action Lifecycle Projection
DSH Harness (DSH) persists action facts to logs. The `dsh-action-ledger` plugin uses the `/action-ledger` command to read and reconstruct these facts into a readable timeline. It projects the lifecycle of tool calls (from request to success or failure), guarantees deterministic ordering, but does not copy parameters or create custom event types. This tool provides only a read-only view for debugging and auditing; it does not replace persistence or authorization systems, nor does it prove physical side effects, and DSH’s developer-preview compatibility should be noted.
Read Moredsh-system-prompt: DeepSeek Harness Real-Time Session Prompt Inspection Plugin
The `dsh-system-prompt` plugin for DeepSeek Harness (DSH) is designed to help developers debug agents through the "everything is a plugin" philosophy. It is a read-only tool that provides a browser interface and an API for statically and dynamically inspecting the current session's effective prompt, context, variables, and tool configuration. It includes two main tabs: the "Conversation Inspection" tab for displaying the components of the prompt, and the "Sections Details" tab implemented via DOM bridging. The plugin exposes a JSON projection of session data through the `/dsh-system-prompt` endpoint, ensuring that live objects are not serialized. Installation is simple; local development requires running a build and aims to reduce the cost of obtaining debugging information.
Read Moredsh-plugin-ui-toggle: Managing DSH Plugin Status in the Web UI
`dsh-plugin-ui-toggle` is a client-side plugin maintained by Biobuilder-ai, designed to address the inefficiency of frequently modifying configurations or restarting services to test plugin toggles in DeepSeek Harness (DSH). The plugin provides a visual plugin toggling feature through the Web UI, enabling changes to take effect immediately and be persistently stored. Its core features include managing plugin states directly in the settings menu, exposing a RESTful API, and writing toggle states into a managed code block at the end of the `cordis.patch.yml` file. It does not rely on additional components and supports developers in dynamically adjusting feature modules, although core plugins cannot be disabled.
Read More# Systematic Debugging: Integrate Structured Debugging Methodology into Skill to Enable AI Troubleshooting Without Blind Guesses
### 正式翻译： The `systematic-debugging` repository comes from `awesome-cursor-skills`. It will write the five-step debugging process of reproduction, isolation, hypothesis, verification and repair into `SKILL.md`, covering git bisect, binary search and minimal reproduction. This article introduces its origin, core capabilities, installation methods in AI programming tools such as Cursor, as well as typical usages such as automatic triggering and explicit invocation via `/systematic-debugging`, helping developers constrain Agents to troubleshoot based on evidence rather than randomly modifying code.
Read More