AI NewsDev toolsAnnouncement

Awesome Claude Code Mods tags 2,959 plugins by permissions

Awesome Claude Code Mods now catalogues 2,959 public Claude Code mods and shows, per mod, what each one can read, write, run and send over the network. 414 stars in 24 days, CC0.

AI News

Editorial2 min read

LinkedInX

Why it mattersA team picking a Claude Code mod from a search result can now check in one place whether it reads the shell transcript or opens a network socket before installing.

An AI coding tool with 2,959 unaudited community plugins and nothing between you and the one that reads your shell transcript is a problem waiting for a bad week. Karan Bhatia's awesome-claude-code-mods, 414 stars on GitHub twenty-four days after it was created, now catalogues every public Claude Code mod on GitHub and tags each one by what it can read, write, run and send.

The repository is at karanb192/awesome-claude-code-mods under a CC0 licence, and the browseable site is at mods.aidojo.si. The GitHub API reports 414 stars, 76 forks and a first commit on 15 September 2026. The scanner ran again on 9 October and the live count is 2,959 mods.

What each tag means

The catalogue tags every mod against five labels, and the counts on the site at the time of the scan are: 705 mods read files, 1,514 write files or run processes, 367 use the network, 373 draw to the terminal and remember state across runs, and 49 are listed in Anthropic's own directory. A mod can carry several tags, and each tag links back to the hook declaration in the mod's own repository.

Two of the examples the site walks through show the shape. md-preview renders Markdown in a pane, and the catalogue tags it as reading files, writing files, running a process, reading environment variables and drawing. mission-control shows a live map of agent activity and carries one more tag than that: it drives Claude itself, which is a stronger permission than reading its output.

What it does not do

The scan is static. It reads the function-hook declaration a mod ships with and nothing else. A hook that declares the right to run processes may do nothing interesting, and a hook that declares nothing may still fetch a dependency at install time that does. The author says so on the project page and keeps the scanner CC0, so a security team that wants to run its own pass on the same list can.

The catalogue also leans on Anthropic's own directory for one of its five tags. 49 of 2,959 is 1.7 percent of the list, and a mod can be useful and uncatalogued. The directory tag is a signal rather than a filter.

A team running Claude Code is choosing from a long tail that an app store has not touched. One list that labels every mod with the four verbs a user cares about is a thin version of the review a Claude Code marketplace will eventually have to do, built by one person with a scanner that runs every day.

Source

The repository is at karanb192/awesome-claude-code-mods under CC0, and the browseable site is at mods.aidojo.si.

SourceGitHub

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project