What's this about?
Anthropic previewed Function Hooks for Claude Code — described as making it "way more hackable" — and is explicitly asking for community feedback on the GitHub issue before shipping. Boris Cherny (Claude Code's creator) amplified it.
What exactly is a "function hook"?
Where today's hooks run shell commands on events, function hooks would let you inject your own code into Claude Code's internal machinery — intercepting and modifying its behavior at defined points, the way editor plugins extend an editor. More power, and a bigger attack surface.
Why is this worth noting?
The HookPry paper (same day) showed the *existing* hook mechanism is the weakest link in seven harnesses. Anthropic asking for design feedback before shipping — rather than after — is the right order, and the GitHub issue is where the security conversation is happening. If hook security matters to your work, this week is the window to influence the design.