Claude Code mods are plugins that run inside Claude Code itself, on the desktop app and in the terminal, and Anthropic shipped them on October 1. A mod can change what you see, remember things across turns, and intercept commands before they run. You can set one up by describing it in plain English. The catch is that a mod runs with your full user permissions, so you should read the code and start with a read-only mod before you trust anything with more access.
Why mods are different from hooks
Until now, customizing Claude Code mostly meant hooks: external scripts that run outside the app. Think of a hook as an outside contractor. It shows up, does one job, and leaves.
A mod is closer to an inside employee. It lives in the app, shares its state, and can draw a pane beside the transcript, add a band above the prompt, show a status line, or block and rewrite a tool call. Anthropic's docs describe the handlers as functions running inside the session rather than shell commands. Creators have put it more simply: think browser extensions or game mods, but for Claude Code.
Three mods worth setting up first
Cache control. AI models bill by tokens, and Claude caches your conversation to cut that cost. When the cache goes cold, a simple follow-up forces the model to reread the whole history at full price. A cache clock mod puts a countdown on screen so you know when that window is closing. The window depends on how you pay. It is about an hour on a Claude subscription within plan usage, and about five minutes on an API key or usage credits, per Anthropic's prompt caching documentation. Check which one applies to you before you trust any timer.
Next Steps. When Claude finishes a task, this mod suggests three logical follow-up actions as clickable buttons. It saves you the effort of deciding what comes next.
Blast Radius. Claude Code can run terminal commands, which means it could delete a project folder by mistake. Blast Radius holds commands like rm -rf, git reset --hard and force pushes, shows what they would hit, and waits for you to approve. It is the digital seatbelt of the group.
If extra widgets sound noisy, remember that mods are modular. You turn on only the ones you want, and some, like Blast Radius, have no visual footprint at all.
How to build your own by describing it
Mods need Claude Code v2.1.287 or later and are on by default. The quickest route is to describe what you want and let Claude write, install and hot-reload it. The official docs use the example "make a mod that shows the current git branch above the prompt." Claude saves the files under ~/.claude/dev-mods/ and, in default mode, asks before creating each file.
If you are not sure what to build, try this audit prompt from the creators covering mods this week:
Audit my last 20 sessions. Identify my repetitive questions, common commands, and how I check your work. Suggest 5 custom mods to streamline my workflow.
Claude analyzes your own workflow and proposes mods tailored to it.
The security catch
Mods are unsandboxed code running as you. A poorly written or malicious mod can read local credential files, access environment variables, run shell commands and make network requests. For a solo creator, that could mean exposed API keys or customer data. The same access that lets Blast Radius guard your machine is what would let a bad mod harm it.
Before you install a community mod:
Run
claude plugin validate <dir>. It lists the events the mod handles and the API calls it makes without executing the code.Read the hooks module yourself. Prefer mods you generated or can read in a few dozen lines.
Trial it for one session with
claude --plugin-dir ~/mods/my-modbefore installing it permanently.Start with a read-only mod that can look but not change anything.
To switch everything off, disable the plugin in the Installed tab of /plugin, or set "disableAllHooks": true in ~/.claude/settings.json. That setting also stops your settings hooks and custom status line.
The bigger question
If the core tool can build its own interface, manage its own memory and police its own commands, how long until separate agent operating systems stop making sense? Has Claude Code become the OS? It is worth asking before you build your workflow around anything else.
FAQ
What is a Claude Code mod?
A Claude Code mod is a plugin that runs inside the Claude Code app, on desktop or in the terminal. It can change the interface, remember context across turns and intercept tool calls before they execute.
How is a mod different from a hook?
A hook is an external script that runs outside the app. A mod runs inside the session and shares its state, so it can change the interface and block or rewrite actions directly.
Are Claude Code mods safe?
Mods run with your full user permissions and are not sandboxed, so a bad one can read files and credentials. Read the code, run claude plugin validate, trial it with --plugin-dir, and start with a read-only mod.
How long does the Claude prompt cache last?
It depends on your plan. Anthropic's documentation gives about one hour on a Claude subscription within plan usage and about five minutes on an API key or usage credits.
Companion episode:
Want more practical AI breakdowns for solo builders and small teams? Get them at joebuildsai.com.