dsh-skill-panel：把本机技能摊开成可搜索、可多选的面板
dsh-skill-panel 是 HUIlv0 维护的 DSH 客户端插件：在左侧导航增加「技能」入口，以悬浮气泡按功能分类列出本机已安装技能（名称、简介、来源），支持搜索与多选，一键将 /技能名 写入对话输入框；窗口大小与字号可调并持久化。本文基于官方 README 介绍安装（pnpm add github 依赖、bundles 与 cordis.patch.yml）、用法、原理与已知限制。
Read Moredsh-mcp-skill-panel: DSH MCP and Skill Management Panel
The DeepSeek Harness (DSH) developer plugin `dsh-mcp-skill-panel` provides visual management of MCP servers and the skill catalog. It supports adding, editing, and testing connections to MCP servers, as well as searching, filtering, and previewing skills. The plugin modifies `cordis.patch.yml` directly; changes take effect immediately if live reload is enabled in the host. Installation requires running the specified command and restarting `dsh web`. Note that MCP settings are not shared between Web and Desktop modes, and skill management is view-only.
Read Moredsh-toolbox-ui: MCP and Skills Management Panel for DeepSeek Harness Web
The `dsh-toolbox-ui` client plugin for DeepSeek Harness (DSH) provides a floating panel that lets users manage MCP servers and local skills directly within conversations, supporting bidirectional operations between global and project-level configurations. The MCP management panel allows viewing status, starting or stopping connections, importing configurations, and managing credentials; the skill management panel supports viewing scopes, locating files, one-click toggling of the model auto-invocation switch, and running diagnostics. The plugin supports hot updates, enabling changes to take effect without restarting a session, making it suitable for developers who frequently adjust configurations.
Read MoreLilyBlessing/DSH-MCP-Skill-Panel: DSH Web Plugin for Managing MCP and Skills
The DSH Web plugin `lilyblessing/dsh-mcp-skill-panel` (v0.5.3, MIT License) aims to centrally manage MCP servers and Skills to optimize context usage. Its core features include: MCP real-time start/stop, tool-level disabling, JSON quick migration, and project-level configuration; Skill enabling, disabling, creation, and manual editing. The plugin provides an optional AI middleware layer that allows models to invoke disabled MCPs on demand via `mcp_search` and `mcp_call`, while preserving the user's manually enabled state. The activation mechanism supports immediate effect or activation at the next session, requiring users to weigh the KV-Cache invalidation cost of immediate activation (approximately 5 to 12.5 times the hit rate). Installation requires using a git source or np.
Read More