dsh-lark-connector

Feishu (Lark) is the daily communication hub for many teams, yet DeepSeek Harness (DSH) had no native bridge to pull those conversations into Agent sessions. dsh-lark-connector closes that gap: messages from a Feishu DM or a group chat that explicitly mention the bot flow directly into an isolated DSH session, are processed by the Agent, and the final reply is sent back to the original message. The entire pipeline runs on the official WebSocket long-connection, so no separate Webhook server is required.

Core capabilities span message routing, image recognition, session isolation, and credential management. DM messages are answered immediately; in group chats, only text and rich-text messages that @ the bot trigger a response. Images (PNG, JPEG, WebP, GIF) are downloaded, persisted to DSH attachment storage, and forwarded as image content blocks so vision-capable models can analyze them. Each Feishu chat maps to its own DSH session with FIFO ordering within a session and concurrent processing across sessions; history is retained through session persistence, and archived chats automatically spawn a fresh visible session. The bilingual Web settings page manages App ID, App Secret, brand, and lark-cli path; lets you choose the default Agent preset, model provider, and workspace for new sessions; and fine-tunes card streaming, collapsible thinking-process display, and tool-call timeline rendering. Interface copy follows the current DSH language automatically.

This plugin suits teams running multi-Agent workflows on DSH who want Feishu as a lightweight conversational entry point, developers who prefer @-mentioning a bot in a group chat over opening a web console, and organizations that rely on lark-cli and need unified credential management with automatic CLI sync, setup checks, and status verification bundled into the Agent toolset.