DeepSeek Harness 表情包外掛——純文字 `[表情: 描述]` token 發圖，內建兩套圖庫，提供 `send_meme`/`learn_meme` 工具，零第三方依賴（僅 node:sqlite）。
DSH 整合
原生執行環境
作者聲明
安全稽核
未稽核
最後核實
2026-08-30
授權條款
MIT
01它能幫你完成什麼？
Send and reply with memes through pure-text chat in DeepSeek Harness
Fetched meme images rendered in chat bubbles as `[表情: 描述]` text tokens — no image-input capability required
DSH chat / QQ- and WeChat-style bot users who want expressive meme replies
Let the model automatically learn and catalog new memes from user attachments
`learn_meme` records the most recent user attachment (no attachment id needed) and auto-detects category/description; also accepts a URL
Users who want to grow a personal meme library without manual tagging
Proactively send a meme that matches the conversation mood
Pick an emotion bucket, randomly draw several captions, and the model pastes one into the reply
Companion-style agents that should react emotionally rather than only answer text
02如何將外掛接入 DeepSeek Harness？
先決條件
- A DeepSeek Harness runtime with a target DSH profile to install into — the README installs into e.g. `~/.dsh/profiles/web/`
- Node.js with the built-in `node:sqlite` module — the only dependency, zero third-party deps
安裝步驟
- 01
Install from npm into a DSH profile: `dsh plugin --profile web add dsh-meme` (equivalent to `pnpm add dsh-meme`)
$ dsh plugin --profile web add dsh-meme
- 02
Or install directly from GitHub / local: `pnpm add github:yyh-001/dsh-meme` or `pnpm add file:/path/to/dsh-meme`
- 03
Switch the active library in Settings → 当前图库 dropdown — built-in `official-001` works out of the box, no config needed
驗證整合成功
作者未說明
03DSH 整合程度與能力邊界
Native DSH plugin — registered via `dsh plugin --profile web add dsh-meme`; exposes `send_meme`/`learn_meme` tools to the model and mounts a settings panel + 😊 picker through `cordis.patch.yml`, with dsh-companion providing the message channel.
Text-based meme sending (send_meme)
emotion tag (e.g. sad/happy) + optional limit→candidate `[表情: 描述]` tokens pasted into the reply; the frontend renders the matching image in the chat bubble
Auto meme learning (learn_meme)
optional attachmentId / imageUrl→records the most recent user attachment into the library and auto-detects category/description
Mood-driven proactive sending
emotion bucket (happy/angry/sad/shy/confused/daily)→random sample of captions from the bucket; the model picks one to paste into the reply
Library management & import/export
upload / edit / delete memes; ZIP file→persisted library with category management; one-click ZIP export/import of a meme pack
settings stored in `~/.dsh/dsh-expression.json`, survives plugin upgrades
04適合誰？何時不該用？
適合
- DSH chat / QQ- and WeChat-style bot users who want expressive meme replies
- Users who want to grow a personal meme library without manual tagging
- Companion-style agents that should react emotionally rather than only answer text
不適合
- `learn_meme` auto image-recognition depends on the current default model supporting image input; if unsupported, you must manually specify the tag/caption.
- The bundled `official-001` library's upstream provides NO LICENSE (no explicit redistribution permission); personal default use is fine, but public redistribution must keep the upstream source attribution in manifest.json.
05相容性、維護與安全提醒
- `learn_meme` auto image-recognition depends on the current default model supporting image input; if unsupported, you must manually specify the tag/caption.
- On disk memes keep fine tags (angry/baka/…) but the model only sees six emotion buckets; when captions don't align, the frontend folds quotes before matching.
- The bundled `official-001` library's upstream provides NO LICENSE (no explicit redistribution permission); personal default use is fine, but public redistribution must keep the upstream source attribution in manifest.json.
MIT licensed · published to npm as `dsh-meme` · last push 2026-08-20 · no tagged release yet
06常見問題
如何在 DeepSeek Harness 中安裝 dsh-meme？
執行 `dsh plugin --profile web add dsh-meme`（等同 `pnpm add dsh-meme`）即可裝進任意 DSH profile（如 `~/.dsh/profiles/web/`）。也可從 GitHub 直接裝：`pnpm add github:yyh-001/dsh-meme`，或本地路徑 `pnpm add file:/path/to/dsh-meme`。
安裝前需要什麼？
需要一個裝有目標 DSH profile 的 DeepSeek Harness 執行環境，以及具備內建 `node:sqlite` 模組的 Node.js。dsh-meme 零第三方依賴，裝完即用。
dsh-meme 是怎麼接入 DeepSeek Harness 的？
它是 DSH 原生外掛：註冊 `send_meme` 與 `learn_meme` 兩個模型可呼叫工具，透過 `cordis.patch.yml`（純 insert、熱掛載免重啟）掛載設定面板與 😊 懸浮窗，並搭配 dsh-companion 提供發圖通道。
它和普通圖片機器人有什麼不同？
模型全程不處理圖片——前端展示表情，只有 `[表情: 描述]` 文字 token 走 DSH，因此不會觸發 dsh 的圖片准入檢查。選擇由六個情緒桶（happy/angry/sad/shy/confused/daily）驅動，而非原始圖片輸入。
排障：pnpm 裝成了舊版本？
pnpm 有「新包安全期」（預設 24h），剛發佈的版本會被靜默回落到舊版；急用可在 `pnpm-workspace.yaml` 的 `minimumReleaseAgeExclude` 裡加上 `dsh-meme`。
07相關的 DSH 工作流程
dsh-pet
作者 pc2005-cloud
DSH 桌面寵物：一行命令裝好即用的透明動畫小桌寵，支援多開、大小位置隨心配置；還內建 DIY 素材鏈，能用 AI 影片自造專屬寵物
whale-girl
作者 vlln
DSH Web GUI 桌面寵物外掛（QQ 寵物形態）：右下角懸浮、可拖拽/投餵/玩耍的積累型夥伴。
dsh-dafeiyu
作者 qcytsn
DeepSeek Harness 的桌面原生大飛魚陪伴外掛，即時展示智慧體狀態，Windows 上置頂顯示。
dsh-kun-like-pet
作者 liyupi
Kun Like 桌寵 —— DeepSeek Harness 桌面寵物外掛：右下角小坤寵隨 Agent 工作狀態切換 9 種動作，任務完成播放「你幹嘛~哎喲」
08資料與來源
**dsh-meme**（原 `dsh-expression`）是 DeepSeek Harness 的表情包插件——找得到、发得出、学得会：
dsh plugin --profile web add dsh-meme
已发布到 **npm**（`dsh-meme`），一行装进任意 DSH profile（如 `~/.dsh/profiles/web/`）：
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-30。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。