Claude Code mods are small add-ons that change how Claude Code looks and behaves. Anthropic released them on 1 October 2026, and they work in the terminal, the Code tab of the Claude desktop app and the VS Code extension.
The simplest way to think about it: skills teach Claude what to do. Mods change how Claude Code itself works. A mod can add a live panel beside your chat, show a context bar above the prompt (how much of Claude's working memory your chat has used), stop a dangerous command before it runs, or get a second AI to check Claude's work. They're a bit like browser extensions, but for Claude Code.
What are Claude Code mods?
Every time Claude Code does something, like running a command, sending your prompt or drawing part of the screen, it fires an event. A mod listens for those events and can watch them, change them or take over completely. That's what lets it do things a skill or MCP never could.
| What it does | Example | |
|---|---|---|
| Skill | Teaches Claude how to do a job | A skill that writes in your brand voice |
| MCP | Connects Claude to another app | Letting Claude read your Gmail |
| Mod | Changes how Claude Code looks and behaves | A panel showing every agent that's working right now |
You don't need to code to use one. You install mods like any other plugin, and if the one you want doesn't exist, you can describe it to Claude and it writes the mod for you.
Why mods are so useful
- You can see what's going on. How full your context is, how close you are to your plan limit and what each agent is doing, without typing a command.
- They catch mistakes. A mod can hold a risky command like deleting a folder until you say yes, or have a second model review every change.
- They save usage. Mods can track spend, warn you before a limit and send simple jobs to a cheaper model.
- Less copy and paste. Your Claude sessions can pass plans and updates to each other directly.
- You can make your own. Describe what you want in one sentence and Claude builds it.
Before you install anything
- Update Claude Code. Mods need version 2.1.287 or later. Run
claude --versionto check. They're on by default once you're updated. - Check what you're using. Panels and bars show in the terminal, the desktop app and the VS Code extension. Some panels only fully work in the terminal.
- Check if it's working. Type
/plugin. A line under the tabs lists the mods that loaded, like1 mod active · agent-flow. - Turn them all off if something breaks. Start Claude Code with
claude --safe-mode, or disable one mod in the Installed tab of/plugin.
Mods are brand new, so most community mods are only a few weeks old with a handful of GitHub stars. The picks below are the ones that solve a real problem, come from Anthropic's own samples or have clear, readable code.
The 8 best Claude Code mods and use cases
1. Add a live context and usage bar
The problem: you only find out your context is full when Claude starts compacting (squashing older messages into a summary to make room). And you only find out you've hit your plan limit when it stops.
The fix: a set of mods that pin the numbers under your prompt and update after every answer.
| Mod | What it shows |
|---|---|
| context-lens | How full the context window is, how fast it's growing and roughly how many turns are left before it compacts |
| quota-meter | Your 5-hour and weekly plan limits, a reset countdown and when you're on track to hit the limit |
| token-ledger | What this session has cost, what the last turn cost and your cache hit rate (how often Claude reused earlier work instead of paying for it again) |
| budget-guard | Only appears near a limit you set. Past it, it stops tool calls and asks before the next prompt |
All four are in one collection: github.com/Arunjay4213/claude-mods. quota-meter needs a Pro, Max, Team or Enterprise plan.
https://github.com/Arunjay4213/claude-mods
Install the context-lens, quota-meter and token-ledger mods from this repo. Before installing, run claude plugin validate on each one and tell me simply what they can access. Skip the CLAUDE_CODE_ENABLE_FUNCTION_HOOKS step in the README, my version doesn't need it.Prefer something more visual? Anthropic's own sample mod token-weather draws a "weather forecast" for your context window above the prompt.
2. Add a side panel showing every agent that's working
The problem: when Claude sends out subagents (helper copies of Claude that each take part of the job), the main chat goes quiet. You can't tell what each one is doing, which one is stuck or which is waiting for your approval.
The fix: agent-flow adds a live panel beside your chat. Each agent gets a row with its status, the tool it's using right now and how long it's been running. Agents waiting on you are highlighted, and it flags any that have gone silent for two minutes. Type /flow to open or close it.
https://github.com/Charlie0113-T/claude-agent-flow
Install the agent-flow mod. Run claude plugin validate on it first and tell me what it can access, then install it and show me how to open the panel.3. Get your Claude sessions talking to each other
The problem: you have one session planning and another building, and you're the messenger. You copy the plan from one window, paste it into the other, then wait for each to finish before you can pass anything back.
The fix: Claude Code can now message your other sessions on the same computer. The planner can send the finished plan straight to the builder, and the builder can report back or ask a question mid-task, without you copying anything. Name your sessions when you open them:
claude --name planner --permission-mode plan
claude --name builderThen, in the planner session, type something like this. The @ lets you pick the other session from a list:
When the plan is ready, send it to @builder and ask it to start. Tell me when builder replies.That messaging is built into Claude Code, so you don't need a mod for the basics. A mod adds a control room on top. The ai-link mod in github.com/Creeperprol/basic-mod gives you /sessions to list every running session, /tell and /tellall to message them, and a host mode where one session hands out tasks to the others with /assign and tracks them on a /board.
https://github.com/Creeperprol/basic-mod
Clone this repo and load the ai-link mod. Run claude plugin validate on the ai-link folder first and tell me what it can access. Then show me how to make this session the host and assign a task to another session.4. Give Claude a second pair of AI eyes
The problem: Claude marks the job done, but it quietly removed an error check, or the change doesn't quite match what you asked for. You only notice later.
The fix: second-opinion sends every change Claude makes to a second, smaller model (Haiku) in the background, along with what you originally asked for. After each turn you get a ✓ or a ⚠ under your prompt. A warning comes with buttons to see the details or have Claude fix it. It looks for real bugs and removed safety checks, not style nitpicks.
https://github.com/egorfedorov/second-opinion
Install the second-opinion mod. Run claude plugin validate on it first and tell me what it can access and what it will cost me per review.It uses your own Claude plan or API key for the reviews. The author estimates about 10 cents per 100 reviewed turns on Haiku.
There's also a built-in option called You should know. It runs a side agent that leaves a short note above your prompt when it spots something you might have missed. It's off by default. Turn it on in the Installed tab of /plugin, where it's listed as cc-plugin-you-should-know.
5. Stop dangerous commands before they run
The problem: one wrong command, like deleting a folder, wiping your recent changes or force-pushing over someone's work, and it's gone.
The fix: blast-radius, one of Anthropic's own sample mods, holds risky commands like rm -rf, git reset --hard, force pushes and database migrations. It shows you what would change, with Proceed and Cancel buttons.
https://github.com/anthropics/claude-code-playground/tree/main/claude-code/mods
Clone this repo, add the claude-code/mods folder as a plugin marketplace and install blast-radius. Run claude plugin validate on it first and tell me what it can access.If you paste API keys or client details into chats, look at secret-redactor too. It swaps secrets, emails and IP addresses for placeholders before the model ever reads them.
6. Switch models automatically based on the task
The problem: your most powerful model handles everything, including the easy stuff like renaming a file or answering a quick question. That burns through your plan limits much faster than it needs to.
The fix: a mod can send each request to a different model before it goes out. Use Haiku for small, mechanical jobs, Sonnet for normal work and Opus for planning, debugging and big changes. There's no standout community mod for this yet, but it's exactly what mods were built to do, so ask Claude to make one:
Make a mod that picks the model for each of my prompts based on how hard the task is. Use Haiku for quick, mechanical jobs like renaming, reading files or simple questions. Use Sonnet for normal coding. Use Opus for planning, debugging and big refactors. Decide once per prompt, not on every step. Show which model it picked in a line under the prompt, and add a /router off command so I can turn it off.If you'd rather not run a mod, the claude-model-router-hook project does something similar with an older-style hook (a small script that runs at set moments). It warns you when your model is overkill for the task rather than switching automatically.
7. Review exactly what Claude changed
The problem: after a long turn, it's hard to tell what Claude actually edited, and giving feedback means describing line numbers in chat.
The fix: two mods make reviewing much easier.
- replay-theater, another Anthropic sample, adds a
/replaycommand that steps through every edit from the last turn, one change at a time. - review-pane shows a diff (the before and after) of every changed file. You comment on specific lines, and Claude answers each comment.
https://github.com/rudrasecure/claude-mods
Install the review-pane mod from this repo. Run claude plugin validate on it first and tell me what it can access, then show me how to comment on a change.Claude Code's own /diff command is now a mod too, so you can turn it off in /plugin and swap in one of these.
8. Keep working while Claude works
The problem: Claude is halfway through a long task and you think of the next thing, or you have a quick question, but typing it would interrupt.
The fix:
- buffer-pane gives you a scratch panel to draft your next prompts while Claude is busy. It saves per project folder, so your notes are there next time.
- aside adds
/aside, a side chat for quick questions that the main conversation never sees, so it stays clean.
https://github.com/meganemura/buffer-pane
Install the buffer-pane mod. Run claude plugin validate on it first and tell me what it can access, then show me how to open it.Make your own mod in one sentence
If none of these fit, describe what you want and Claude writes the mod. Claude Code has a built-in skill for this called plugin-authoring. Some ideas to try:
Make a mod that shows the current git branch above the prompt.Make a mod that sends me a desktop notification when Claude finishes a long task.Make a mod that adds a /standup command summarising everything that changed today.What happens next:
- Claude writes the mod and asks you to approve each file.
- Choose Enable for this session when asked about hot reloading. The mod loads as soon as Claude finishes.
- Try it. If it's not right, tell Claude what to change and it reloads at the end of that turn.
- To keep it, ask Claude to copy the mod out of
~/.claude/dev-mods/into a folder of your own. Mods Claude writes only load in the session that made them, and that folder gets cleaned up after a while.