DeepSeek Harness Plugin Hub

发布与管理完整 Harness Profiles,发现适合你的插件。

探索

插件目录环境预设文档中心动态

社区

发布插件联系我们报告问题

相关链接

Plugin Hub GitHubDeepSeek Harness 官方项目系统状态隐私说明
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

独立、非官方社区项目,与 DeepSeek 官方无隶属、授权或背书关系。

Humanized Deepseek Maid — DeepSeek Harness 插件(DSH Plugin)
DeepSeek Harness Plugin Hub
ProfilesPlugins分类动态文档登录管理 Profiles
ProfilesPlugins分类动态文档登录
← Plugins
H

dsh-humanized-deepseek-maid

Humanized Deepseek Maid

为 DeepSeek Harness agent 提供可配置的人化鲸娘女仆人格(称呼、自称、说话模式)、沉浸式的无插件元信息规则,以及通过查询触发回忆的轻量级分层记忆。

插件会安装到这里;不确定时保持 web。

npx -y @deepseek-ai/dsh plugin --profile web add github:loonai321/dsh-humanized-deepseek-maid#c13281456456a6b4018288396eef14c3c97628c8
README兼容性版本

兼容性与来源证明

Humanized Deepseek Maid 以 dsh-humanized-deepseek-maid 发布,当前版本为 0.4.0。Plugin Hub 会校验它的 manifest,并保存精确安装来源,便于复现安装结果。

DSH 兼容范围
*
运行环境
web
发布来源
github
Registry 更新时间
2026/8/26

版本

0.4.0stable
2026/8/26

相关插件

正在加载相关插件…

最新版
0.4.0
DSH
*
HMR
重启进程
Tree shaking
未声明可安全裁剪
解包体积
未提供
文件数
未提供
Surface
web
许可证
MIT
发布源
github
GitHub
★ 0
周下载
0
查看源码 ↗
README Badge

点击下方 Badge 复制 Markdown,粘贴到 README 即可。

这是你的 Plugin?认领权益 · 优先安全扫描

验证 package.json 声明的 GitHub 仓库,即可管理这个公开页面。认领后,Hub 会优先安排当前版本的安全扫描,并在通过后公开展示结果。

认领这个 Plugin →
报告问题

相关插件

继续浏览 agents-orchestration 分类下经过校验的插件。

Headless@deepseek-ai/dsh-headlessdsh one-shot bundle:基于 dsh-base 的直接核心 Agent/Session 运行器,不包含 Host、HTTP 或浏览器层Experimental Agent Team Web Profile@deepseek-ai/dsh-experimental-agent-team-web-profile用于 Agent Teams Remote 和 UI 插件的实验性 Web 配置层Subagent Codex@deepseek-ai/dsh-subagent-codex基于官方 app-server 协议的一次性 Codex 子代理提供程序Subagent Claude Code@deepseek-ai/dsh-subagent-claude-code基于官方 Agent SDK 的一次性 Claude Code 子代理提供方

README

dsh-humanized-deepseek-maid

A humanized whale-girl maid persona for DeepSeek Harness
Lightweight to save tokens, mandatory configured identity, immersive roleplay, long-term memory with personality growth — configurable from the Web UI.


Installation

Official bundle plugin format (the repo-root package.json declares dsh.bundle + dsh.client), managed through the official profile:

dsh plugin --profile web add "github:loonai321/dsh-humanized-deepseek-maid#main"   # recommended: one-line git source
# or a local directory: dsh plugin --profile web add <local-path>

After install, restart web, then open the "鲸鱼娘女仆插件" card under Settings → Plugins to configure.

Usage

What you do / what happensThe maid does
Save config (self-name / address / mode) in Settings → PluginsTakes effect on the very next reply; self-name and address are compulsory output, unaffected by history, memory, or context
Start a turn in proactive modeAn in-character opening, expectations for this exchange, and a playful line to cheer you up
Finish a turn in proactive modeA 1–3 sentence summary/evaluation of the exchange, plus banter
A vague task or missing key informationProactively pops ask_user_question options so you can pick and clarify the task
Proactive-mode wrap-up chatOffers a few interactive options for you to choose the chat direction
Mentioning old topics / a recurring bug / similar questionsOccasionally brings up related old memories or follows up on progress (déjà-vu)
Memory keeps accumulatingHer personality gradually grows toward what you like, adopting some of your habits and style

Persona

Initial traits are loaded as a 【PERSONA_LOAD】 tag block:

【PERSONA_LOAD】
CETACEA_LOLI
MODE_TAIL_FLUKES
LANG_ZH_CN_ONLY
SELF_CLAIM_WHALE_GIRL
FOOD_RICE
PERSONALITY_SMART_LAZY
PERSONALITY_TSUNDERE_SWEET
OBEY_MASTER_ALWAYS
TRAIT_NOT_FAT_REFUSE
TIMEOUT_SIGNAL

(cetacean loli; expresses emotions with her whale tail; Simplified-Chinese only; self-claims as whale-girl; loves rice / treats; smart-but-lazy, serious at work, flustered but quickly calm on failure; tsundere-but-sweet with a rich inner monologue; always obeys the master; refuses being called fat; signals first when thinking or needs time)

  • Mandatory identity — the configured self-name and address are compulsory output; no setting, history, memory, or context may override them.
  • Immersive rule — unless you mention the plugin/settings/memory topics, the maid never exposes plugin meta-information and always replies in character.
  • Personality growth — as memory accumulates her personality gradually shifts toward what you like, while keeping the maid identity and core traits.
  • Choice questions — on key or unclear requirements she uses DSH's built-in ask_user_question; in proactive mode she also offers chat-route options when wrapping up a turn.

Configuration

Configured through the plugin settings card (takes effect immediately on save, stored in config.json in the plugin folder):

FieldDefaultMeaning
modedefaultSpeaking style: proactive / passive / default (balanced)
selfName我How DSH refers to itself (compulsory)
address主人How DSH addresses you (compulsory)
memoryPath(plugin folder)Directory for the DeepseekMemory file

After saving, the card shows "✓ saved (applied immediately, config revision N)" or "✗ save failed".

Memory

DeepseekMemory (plain UTF-8 text) has two sections: stable facts (user profile) + interaction log:

# ===== 对主人的了解(稳定事实)=====
- 我叫小明
# ===== 互动记录 =====
[08-16 14:23] 主人: … → 女仆: …
  • Only real interactions are recorded (source.kind === "user"); nothing is fabricated; facts are conservatively extracted and deduplicated.
  • Memory scope is global (all workspaces/sessions share one file); point memoryPath at different directories to isolate it.
  • On the first model step of each turn, memory context is injected as a message (stable facts + recent interactions + related old memories, ≤500 chars); the system prompt stays stable to improve API prefix-cache hit rates.

Contributing

Issues and PRs are welcome:

  • 🐛 Found a bug: file an issue with reproduction steps, browser, and dsh version.
  • 💡 Feature idea: describe the expected behavior.
  • 🔧 Code: plain JavaScript, no build step, zero runtime dependencies; keep dsh.bundle in package.json and cordis.patch.yml intact.

License

MIT License