Install
Inside DeepSeek Harness, with dsh-market
dsh plugin --profile web add dshmarket
Or from the command line
dsh plugin --profile web add github:moononnn/DeepSeek-Harness-biaoqingbao
Installing runs third-party code with your own permissions — it can read your files, use your credentials and reach the network. Review the source first, and pin a commit (github:owner/repo#sha) when you can.
README
Let the assistant in DeepSeek Harness express emotions with stickers.
⚠️ This is a third-party community plugin, fully open source (MIT). It reads and writes sticker library data locally on your machine; AI tagging / auto-matching features call the model services you already configured in DSH (the corresponding text and images are sent to the model provider). Installing means you trust this source. DeepSeek Harness is in developer preview (v0.1); plugin interfaces may evolve.
What it does
- 😀 Emotional sticker matching: the assistant calls the
expresstool to send a fitting sticker when it feels happy, sad, proud, speechless... - 🗂️ Library management: multi-select upload (PNG/JPG/WebP/GIF), browse, search, edit tags
- 👍 Feedback tuning: 👍/👎 buttons on sticker cards — the more you use them, the better the picks
- 🏷️ AI tagging: auto-tag images (description / emotion / scene / keywords) with your configured DSH model, single & batch
- 👀 Auto-match observer (optional): detects emotional moments in conversation and nudges the assistant to send a sticker
- 🗣️ Dialects (optional): 9 dialects — Northeastern, Henan, Shanghai, Cantonese, Taiwanese, Sichuan, Shaanxi, Beijing, Xinjiang
- 🎭 Mimic my style (optional): analyzes your chat history and distills your typing style
- ⚙️ Settings panel: sidebar "表情包" entry — global switch, observer, vision model, dialect, style mimicry
- 👥 Per-assistant settings: when assistant presets exist, each one can get its own dialect, sticker frequency and on/off switch (e.g. one assistant speaks Sichuanese, another speaks Cantonese — they never mix)
Works great with the memory plugin (recommended)
Install Hanako Memory (brings openhanako's memory system to DSH, including assistant-preset management: create presets, pick personalities, write personas, manage memory) and the per-assistant settings light up automatically:
- Every assistant preset appears in the "每位助手单独设置" list in the settings panel
- Each assistant gets its own dialect, sticker frequency and 👍/👎 preferences
- Sticker matching automatically follows the assistant speaking in the current session
It also works standalone: without assistant presets, the global defaults apply and everything keeps working.
Install
dsh plugin --profile web add <package-name>
Restart dsh web, then open the sidebar "表情包" entry:
- Upload images you have rights to use (multi-select supported)
- Chat — the assistant will send stickers when it has emotions; or just say "send a sticker for being wronged"
- Tap 👍/👎 on sticker cards to tune the matching
Or install from GitHub:
dsh plugin --profile web add github:moononnn/DeepSeek-Harness-biaoqingbao
Data & privacy
- Library, preferences and usage logs are stored locally at
${DSH_HOME}/plugin-data/biaoqingbao/ - No chat logs are uploaded
- AI tagging / observer / style mimicry use the DSH model route you configured
- Uninstalling keeps your library data
Compatibility
- DeepSeek Harness v0.1 (rc series), Node.js 20+
- Web profile fully supported; headless gets tools without the management panel
Development
npm run check # syntax check (lib + client)
npm test # unit tests for core logic (node:test, zero deps)
Core pure functions (scoring, preferences, tag sanitizing, frequency gating) live in lib/core.js.
Relation to the Hana version
This plugin is a port of moononnn/hanako-biaoqingbao (a Hana community sticker plugin) for DeepSeek Harness: the original tool registration, library management, feedback tuning, AI tagging, dialects and style mimicry are all preserved, re-implemented on DSH's plugin system (dsh.bundle + Cordis).
Credits
Inspired by the design of the Hana community sticker plugin. Thanks to moononnn for the original work.
License
MIT © moononnn & 小花