dsh-plugin-tool-management

Managing MCP servers, skills, subagents, prompts, and memory in DeepSeek Harness has usually meant juggling config files, toggling tools one by one, and mentally tracking which persona goes with which memory set. dsh-plugin-tool-management folds all five domains into a single interface with eight tabs—Scenes, MCP, Skills, Subagents, Prompts, Memory, Sessions, and Host Compatibility—so configuration stops living in scattered YAML blocks. The core problem it solves is the friction of switching work modes: build a dedicated scene for coding, writing, or general work, bind a specific set of MCP servers, skills, subagents, and memory files to each, then flip one switch to activate the whole bundle. Turning it off restores the previous state automatically—no copy-pasting prompts, no manual tool toggling.

At a granular level, the plugin makes model visibility an explicit, observable control. Memory markdown files are injected into context automatically. MCP servers accept decision hints in their remarks that the model actually reads and follows. Individual tools inside a server can be disabled in isolation. Subagents run and-return without polluting your session history. Prompt presets support multiple baselines with one-click application and five-generation backups. On the safety side, credentials are masked by default and mask values are never written back into patch files; skill source files remain read-only; deleted items land in a recoverable trash; and scenes can be locked to freeze add-modify-delete across all five domains.

This plugin suits DSH users who frequently shift between roles—coding agent on one tab, long-form drafting on another, MCP tool chains in play—while wanting each environment to stay independently configured, auditable, roll-back-able, and exportable as a portable bundle rather than a tangle of terminal flags.