dsh-messaging

Getting an AI Agent talking on Feishu, Telegram, Discord, and WeCom used to mean writing a separate adapter for each—each with its own formatting quirks, capability gaps, rate-limit codes, and retry logic. dsh-messaging collapses all of that into one plugin: a MessagePlatform abstraction provides a single unified send-and-receive interface, while Feishu, Telegram, Discord, and WeCom plug in as drop-in adapters. Configure once and switch between platforms without rewriting agent logic.

The standout design is capability auto-degradation. Every platform declares 12 typed capabilities (text, Markdown, image, audio, video, file, card, reaction, thread, typing indicator, push, and more), so the agent can query what is possible before acting. Unsupported features never fail silently; they trigger bilingual, user-friendly notices. Markdown that Feishu cannot render is automatically downgraded to plain text; WeCom card messages degrade to formatted text; Discord cards become embeds. Production hardening is equally thorough: message deduplication with TTL and cap, per-chat serial processing to preserve ordering, exponential backoff with jitter that respects each platform's rate-limit codes, and a streaming buffer that accumulates assistant text into a single complete message rather than firing one request per token. All of this has been battle-tested in Hermes Agent for over three months.

The plugin suits developers building AI Agents on DeepSeek Harness who need to cover two to four IM platforms from one codebase without hand-rolling format differences, retry policies, or capability checks per platform. Current limitations to keep in mind: Feishu media sends (image, audio, video, file) are not yet implemented; each instance binds to a single platform; and a test suite is still on the roadmap.