Start with the mods that change something you already do every day. If you read long replies, try a reading pane like claude-review. If you worry about risky commands, try Anthropic’s blast-radius sample. If you keep running out of context, try a meter like context-lens. Sometimes the best mod is one you ask Claude to write for you.

Mods launched on October 1, 2026. Most mods below are a few weeks old and have a handful of stars, so treat this as a first look, not a ranking. I read each README and the official docs. I haven’t run every mod here. Star counts were checked on October 2, 2026.

Disclosure: I build StashBase, a writing workspace for Claude Code and Codex. None of the mods below are ours.

What is a Claude Code mod?

A mod is a plugin with a few JavaScript or TypeScript functions that Claude Code calls when something happens: you submit a prompt, Claude calls a tool, or part of the screen is drawn. Each function can watch the event, change it, or handle it instead of Claude Code. (Mods overview)

That gives mods three kinds of power that settings hooks didn’t have:

  • Draw interface. A pane beside the transcript, a band above the prompt, toasts, a status line. Mods can also restyle what Claude Code already draws, like messages, tool output and the spinner. The permission prompt is the one thing they can’t touch.
  • Rewrite what goes in. A mod can change your prompt before Claude sees it, add context, or drop it.
  • Answer instead of Claude Code. A mod can deny a tool call, return its own result, or switch the model for one request.

You need Claude Code 2.1.287 or later. Mods are on by default. Hooks run in the terminal, the Desktop Code tab, VS Code and claude -p, but the drawn interface only shows in the terminal and Desktop.

How to install a mod

Mods ship inside plugins, so you install them like plugins. Add the author’s marketplace, then install:

claude plugin marketplace add r3al1tym/claude-review
claude plugin install review-pane@claude-review

Inside a session, the same commands work as /plugin marketplace add … and /plugin install …. If you installed from a shell while a session was open, run /reload-plugins in it. In the Desktop app, go to + → Plugins → Add plugin. (Install docs)

To see what loaded, run /plugin. To turn one mod off, disable its plugin there. To start a session with no installed mods, use claude --safe-mode.

Read this before you install anything

Anthropic is direct about it: mods aren’t sandboxed. A mod “is code that runs with your permissions. It can read and write your files, start processes, and make network requests.” It can see every prompt, rewrite it, approve tool calls before you’re asked, and spend your usage on model calls. (Mods overview)

Your deny rules don’t fully protect you either. Even where they apply, they don’t cover a mod’s own file and process calls. If you deny Read(.env), a mod can still read .env itself.

So check a mod before it runs:

claude plugin validate ./some-mod

This prints which events the mod listens to and which APIs it calls, without running it. Look closer at anything that calls $.fs, $.process, $.http.fetch, $.env.get or $.prompt.submit, or listens to tool.call or prompt.submit. A clock in your status line has no reason to read your files.

The community list awesome-claude-code-mods runs this check every night. On October 2 it covered 359 mods: 127 start processes, 95 write files, 75 reach the network and 111 see every prompt. That doesn’t make them unsafe, but it tells you how many need a closer look.

Mods that come with Claude Code

A few features are already mods, so you can turn them off or replace them:

ModWhat it does
cc-plugin-diffDraws the /diff pane. Disable it and /diff falls back to the old version.
cc-plugin-agents-mdLoads AGENTS.md as project instructions.
cc-plugin-you-should-knowA side agent that leaves short notes above the prompt. Off by default; turn it on with /plugin enable cc-plugin-you-should-know@builtin.

The source for these lives in anthropics/claude-code/mods. They’re worth reading before you write your own.

Mods for reading and writing in Claude Code

These help when the work is words: long replies, drafts, docs and the prompts themselves.

  • claude-review (review-pane). Shows Claude’s latest reply in a pane beside the conversation, rendered for reading, and follows the turn as it streams. Useful when replies are longer than your terminal is tall.
  • review-pane by Rudra Secure. Shows a diff of every file Claude changed. You comment on lines and send the comments back, and Claude answers each one. This is the closest thing to reviewing a draft with tracked comments.
  • buffer-pane. A scratch pane for writing your next prompts while a turn is still running. The buffer is kept per folder across sessions.
  • md-prompt. Renders Markdown in the prompt box as you type: code blocks, emphasis, lists, tables. It only changes what you see; what’s sent is exactly what you typed.
  • aside. /aside opens a side chat about the session. The main thread never sees it, so you can ask “what did it just change?” without adding noise to the context.
  • prismantis. Restyles replies with tables, diagrams and copy buttons, in 15 themes.

If you write in Claude Code a lot, these make the terminal easier to read. But it’s still a terminal. I find it less comfortable for reading and editing articles. I want the document open in front of me, with the agent’s edits shown as changes I can accept or reject. That’s why we built StashBase, which opens Claude Code beside your files in a document app. The writing guide has more on that setup.

Mods for keeping an eye on usage and agents

  • context-lens, quota-meter, token-ledger. Three status lines: how full the context is, where you are in your plan’s window, and what the session cost.
  • token-weather in Anthropic’s samples. A context-window “forecast” above the prompt. Anthropic shares these samples as they are, without support.
  • agent-flow. /flow shows a live tree of subagents, so you can see what’s running.
  • cctop. A dashboard pane in the style of btop.

Mods for safety and privacy

  • blast-radius in Anthropic’s samples. Holds an rm -rf or a force push and shows what it would change, with proceed and cancel buttons.
  • secret-redactor. Replaces secrets, emails and IP addresses with placeholders before the model reads them.
  • topic-filter. Swaps topics you choose for neutral placeholders before Claude reads them. Your files on disk stay the same.

These are useful, but remember the warning above: a safety mod is still code with your permissions. Read it first.

Mods just for fun

  • Mindful-Claude (86 stars, the most-starred mod repo I found). A breathing exercise above the prompt while Claude works.
  • cc-arcade. Nine games above the prompt, for long turns.

Where to find more

  • awesome-claude-code-mods: the biggest list, with nightly checks on what each mod can reach. Some older mods fail on 2.1.287, and the list marks them.
  • claudemods.ai: a community catalog with votes, mostly games and dashboards. Not run by Anthropic.
  • claude-code-templates: a large template repo with about 38 mods under cli-tool/components/mods/.

Many READMEs from September still tell you to set CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1. That was the early-access switch. It’s ignored now, and you can leave it out.

Ask Claude to write your own mod

Often there’s no mod for the small thing you want. You can describe it in a session:

Make a mod that shows the current git branch above the prompt.

Claude uses the built-in plugin-authoring skill and writes the files to ~/.claude/dev-mods/. You approve each file, then choose Enable for this session. The mod reloads as Claude edits it, so you can ask for changes and watch them appear. (Create a mod)

That mod only lasts for the session. To keep it, copy the folder somewhere and load it with:

claude --plugin-dir ~/mods/git-branch

A mod you wrote yourself is also the easiest one to trust, because you watched every file get written.