What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Claude Code mods let developers change how the coding assistant handles prompts, tool calls, permissions, and parts of its interface. Anthropic introduced the TypeScript-based system on October 1, 2026, for Claude Code’s CLI and desktop app. Unlike a simple settings tweak, a mod can intercept an event, alter its handling, or replace a built-in feature—but mods run with the same machine access as Claude Code and are not sandboxed.
What Claude Code mods do
A mod is a small TypeScript function packaged inside a Claude Code plugin. Claude Code emits events—for example, a tool call, a permission prompt, or an interface-rendering event—and a mod can respond before or after an event, wrap it, or take over its handling. Anthropic says multiple mods attached to the same event run in load order: the first loaded sees the event first and receives the result last. Anthropic’s launch announcement describes mods as an expansion beyond hooks, which offered more limited control.
What a mod can change
- Rewrite a prompt or tool call, or block or retry a tool call.
- Approve or deny a permission request.
- Redact sensitive information from tool output.
- Add or change interface elements.
- Replace or add features in Claude Code.
Anthropic’s example is the /diff command: its built-in implementation is now a mod that users can disable through /plugin or replace with their own. The official catalog also lists built-in AGENTS.md support and the sec-default mod. Browse the Claude Code mods catalog.
How mods differ from other customization
Choose a customization mechanism based on how much control your change needs. Settings and permission rules configure behavior; shell-command hooks run commands when events occur. Mods can keep state in a session, change event handling, draw UI, and replace built-in features. Anthropic’s tutorial distinguishes a mod loaded once, which can maintain state and draw UI, from settings hooks that pass JSON through standard input and output. Read the getting-started guide.
Recommended Free Tools
#1 Best Overall
| Mechanism | Best suited to | What it can do |
|---|---|---|
| Settings and permission rules | Configuring Claude Code behavior or enforcing command permissions | Set behavior and define permission rules. |
| Shell-command hooks | Running an external command in response to an event | Receive event data as JSON over standard input and return output over standard output, as described in Anthropic’s tutorial. |
| Mods | Changes needing session state, event control, custom UI, or feature replacement | Register event hooks, keep state while loaded, draw UI, and change or replace event handling. |
Install a mod carefully
Anthropic’s October 1, 2026 tutorial says Claude Code 2.1.287 or later is required and mods are on by default. The feature was announced for both the CLI and desktop app. Anthropic notes that the mod API may change between releases, so the declarations generated by the Claude Code version that loads a mod are the authority for that build.
- Check your version. Confirm you are running Claude Code 2.1.287 or later, the minimum stated in the October 1, 2026 tutorial.
- Inspect the plugin source. A typical plugin includes a
.claude-plugin/plugin.jsonmanifest, ahooks/hooks.jsonfile that points to a module, and a JavaScript or TypeScript module that registers event hooks. Review the actual code and publisher before installing. - Install through Claude Code. Anthropic says plugins can be installed through the Claude directory or with
/pluginin the CLI. Follow the relevant plugin’s installation instructions and confirm its source before approving it. - Check behavior and permissions. Use permission rules for hard blocks. An interface or safeguard that merely inspects command text may miss command substitution, aliases, or scripts, as Anthropic’s tutorial warns about its Blast Radius example.
- Recheck compatibility after updates. Since the API can change between releases, verify a mod against the type declarations generated for the version you are using.
Are Claude Code mods safe?
Mods are executable code, not sandboxed extensions. Anthropic’s launch announcement says: “Mods run with the same access to your machine as Claude Code itself. They aren’t sandboxed, and you should only install mods from sources you trust, the same way you’d install any code on your computer.” That access makes publisher trust and code review a practical requirement, not an optional precaution.
For Team and Enterprise, owners can allow or block plugin marketplaces in the admin console. For Claude API and third-party API plans, administrators can push managed settings to users’ machines. Anthropic says sec-default loads first on Team and Enterprise plans and on machines with managed settings, blocking risky actions such as mods overriding permission-deny rules. If administrators load their own mods first, Anthropic says they should include sec-default to retain those restrictions.
What developers could build
Anthropic’s tutorial illustrates the range with Token Weather, a context-window display; Blast Radius, an interface for reviewing or holding selected risky commands; and Replay Theater, a pane for reviewing edits from a turn. These are examples in the tutorial, not independent evidence that every safeguard catches every risky command. Anthropic also describes possible team uses such as showing CI/CD pipeline status, asking for confirmation before commands alter production configuration, and auditing calls made by other mods.
For a strict security boundary, do not rely on a mod that only examines command text: the tutorial specifically warns that the Blast Radius example can miss command substitution, aliases, and scripts. Use Claude Code permission rules when the requirement is to block an action rather than merely flag it or request confirmation.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
