DeepSeek Harness 플러그인 — 940 페이지

DeepSeek Harness 的三合一能力台：提示词定制（按名称屏蔽 / 替换 / 注入 / 排序，按会话阶段与 agent 预设分层）+ 技能集合管理（导入 / 分组 / 开关）+ MCP Server 接入（分层遮蔽与工具级启停）。纯插件注入，不动官方源码。

防提示词注入与 LLM 防御体系：外部内容标记 + 注入特征检测 + 危险动作人审门控 + 审计侧车

System-prompt surface drift gate for DeepSeek Harness: baselines every section contributed to the assembled system prompt and warns loudly when a plugin or preset upgrade silently adds, removes, or rewrites what the model is told

In-place prompt editing, branching and history rewrite for DeepSeek Harness

DSH 插件：在「更多设置」里查看、修改并保存所有 agent 共用的全局系统提示词，保存后立即生效。

提示词工程模式：结构化推理（CoT/ToT）、few-shot、模板变量、生产级提示词优化。受 wshobson/agents（38k★ MIT）启发。

A sparkle button in the DeepSeek Harness composer that rewrites your rough draft into a clear, well-structured prompt.

增强提示词魔法棒:DSH Web 对话输入框的灵活自适应提示词增强插件(磁盘常驻版,重启自动加载)

Prompt enhancement for the dsh web GUI: rewrite the composer draft into a well-structured prompt through the harness LLM service, preview the before/after, then fill back, copy, or undo. Original draft is always preserved.

Prompt enhancement for DeepSeek Harness: a composer button and an enhance_prompt tool that rewrite a draft into a clearer, more specific prompt, following the same rules as WorkBuddy's enhance-prompt feature.

Sparkle button in the DSH composer: rewrite the draft into an executable prompt through the harness LLM, with a multi-level undo stack and a token-level before/after diff.

One-click prompt enhancement for the DeepSeek Harness composer: a chip beside the model selector that rewrites your draft into a precise, structured prompt using the session's own model.

本地加固 fork（源自 zzy6-a/dsh-prompt-enhance，MIT）：输入框旁一键把口语草稿改写成结构化专业提示词，可撤销；五种风格、模型可跟随或固定、芯片置灰保护；零文件写入、严格 Origin 校验、内置限流。

Composer prompt enhancer for DeepSeek Harness: one ✨ button rewrites the draft into a structured prompt, flags open questions below the input, follows the current session model, and can undo the rewrite; /enhance does the same rewrite as a slash command.

RooCode-style Enhance Prompt for DSH: a wand control in the composer toolbar that rewrites the draft through one auxiliary LLM call (switchable model, defaults to zai/glm-5.3-flash), replaces it in place and stays undoable.

One-click prompt enhancement for the composer: rewrite the draft with prompt-engineering rules, write it back undoably, and keep @file references verbatim.

Prompt enhancer for DeepSeek Harness: a composer toolbar button that rewrites the current draft through an auxiliary LLM call, with one-click revert. Prompt text reverse-engineered from CodeBuddy CN's 增强提示词.

提示词文件注入

DSH 提示词锻炉:零配置流式优化(复用会话模型)+ 轻量词库(JSON 文件)+ 一键存技能。Prompt Forge: zero-config streaming optimization + file-based prompt library + one-click skills.

Prompt optimizer for the DSH web GUI: one button beside the composer rewrites the current draft through a configured LLM channel (host-side, no key duplication), streams the result into a side-by-side dialog, and only writes back on Accept.

DSH composer plugin: ↑/↓ prompt-history navigation — browse the current session's sent prompts right in the input box, with a "历史 x/y" badge.

Prompt history navigation for DeepSeek Harness (dsh) web: browse previously submitted prompts in the conversation composer with the Up/Down arrow keys, shell-style.

System-prompt template injection for DeepSeek Harness: manage named prompt templates and inject the selected one (global default plus per-workspace overrides) into every session's system prompt.

DSH plugin: register a configurable prompt variable so agent presets that use complete:true personas (which drop every other system-prompt section) can still receive an injected text block.