uAcharGG/dsh-vision

README 摘要

@uachar/dsh-vision-plugin Vision recognition plugin for DeepSeek Harness : paste images into the Web composer, recognize them via GLM-4V on the host side, and inject the result into the conversation as a hidden context message. A native dual-face (host + browser) plugin installed as a standalone package. Features - Paste & send images — one or more images pasted into the composer are forwarded to the host for recognition. - GLM-4V recognition — the host calls Zhipu GLM-4V and appends the result as a hidden context user message, so the main model answers based on what the image actually shows. - Visible thumbnails — recognized images are stored as durable attachments and rendered as thumbnails in history. - Model / API key settings — a dock above the composer lets you add models (default glm-4v-flash ) and API keys; persisted to $DSH HOME/vision-config.json . - Self-healing dsh patch — on load, the host half checks the dsh checkout for the attachment-authorization patch and applies it automatically (idempotent one-line change + apiproxy rebuild). How to use 1. Open a session — the 视觉识别 dock appears above the composer. 2. Click the gear icon, add a model (default glm-4v-flash ) and a…

分类