Anthropic's plugin marketplace ships as read-only mirror

Anthropic's community plugin marketplace went live as a read-only mirror at anthropics/claude-plugins-community, and it picked up roughly 190 stars on August 22. The repo's marketplace.json file syncs nightly from Anthropic's internal review process. Nothing in it got there without passing through claude.ai submission, automated security scanning, and manual approval first.

That process matters more than the star count. Anyone who's watched the last two weeks of research into malicious Claude skills knows why a reviewed channel is worth having. Unvetted plugin installs are exactly the kind of surface that gets abused, and a scanned, approved catalog closes that gap for anyone who sticks to it.

Installing is two commands: claude plugin marketplace add anthropics/claude-plugins-community, then claude plugin install <name>@claude-community. Same catalog, same review process, backs Claude Cowork too, so this isn't a side project bolted onto the CLI. It's shared infrastructure.

The catch nobody's fixing yet

Here's the thing worth saying plainly. A reviewed marketplace two weeks after malicious-skill research made the rounds isn't a feature launch worth applauding, it's a gap that should've been closed before the ecosystem got big enough to attract that kind of research in the first place. I'd rather see it late than not at all, but the timing tells you the review process was reactive, not planned ahead of the threat.

What I'd watch next: whether Anthropic publishes what the automated scanning actually checks for, because a black-box "passed security scanning" label is only as good as what's behind it. Until that's public, treat the badge as a floor, not a guarantee.