Perceive Beating AI NewsFlash: Anthropic is testing a new extension mechanism for Claude Code called Function Hooks, which is not yet live. It allows plugins to directly interface with Claude Code using TypeScript functions, enabling them to even modify React interface components, display content, and interactions. Existing Hooks mainly trigger commands, HTTP, MCP, Prompt, or Sub-Agent on specific events.
Function Hooks operate similarly to middleware in Express or Koa, allowing multiple plugins to be nested layer by layer. Plugins perform actions through a controlled $ object, with administrators able to revoke certain capabilities, preventing subsequent plugins from being invoked. A Hook can also listen to all events and operations issued by plugins through $ for unified audit log recording.
The most significant change is that Claude Code plugin capabilities are now delving into the interface and interaction layer. In an official demo, Claude can generate a plugin in a single sentence, automatically masking sensitive information such as passwords before it enters the model output. Another plugin can directly modify the display of Claude Code Desktop, hiding sensitive content that only reveals when the mouse hovers over.
