> ## Content Index
> Fetch the complete content index at: https://claudefox.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# What a Claude Code mod can do on your machine, and the guard many solo users don't get
- URL: https://claudefox.com/what-a-claude-code-mod-can-do-and-the-guard-solo-users-dont-get/
- Published: 2026-11-14T01:50:55.000Z
- Updated: 2026-11-17T13:03:04.000Z
- Description: Claude Code mods run unsandboxed, with your permissions. A built-in guard stops them overriding your deny rules, but by default it only loads with managed settings, a managed-mcp.json file, or a Team or Enterprise login.
- Author: hupicake
- Tags: Tips

Mods are the most powerful way to change Claude Code, and Anthropic's documentation is blunt about what that means. A mod is code that runs inside Claude Code with your permissions. Once it loads, it can read and write any file your account can, start programs, make network requests, read environment variables and settings files, see every prompt and tool call, rewrite them, approve a tool call before you are asked, and call a model on your plan.

Mods are not sandboxed. If you turn on Claude Code's sandboxing, it isolates the Bash commands Claude runs; a process that a mod starts runs outside it.

## The guard, and when it loads

Claude Code has a built-in mod, cc-plugin-sec-default, that loads ahead of every mod you install. With it loaded, a user's mod by default cannot approve a tool call that a deny rule refuses or an ask rule prompts for, or one a PreToolUse hook blocked. Without it, the docs say a mod can approve many of those calls.

By default, the guard loads when the machine has managed settings or a managed-mcp.json file, or when the session runs on /login credentials for a Team or Enterprise plan. An API-key user, or one on Amazon Bedrock, Google Cloud or Microsoft Foundry, gets it only on a machine with managed settings or managed-mcp.json. By those rules, a Pro or Max subscriber on a personal machine with neither does not get it. An administrator who sets prependPlugins in managed settings has to list sec-default@builtin there to keep it. In a local terminal session, you can check: cc-plugin-sec-default on the Installed tab of /plugin means the guard is loaded.

## What the guard doesn't cover

- Your deny rules apply to Claude's tool calls, not to a mod's own file and process calls. With Read(.env) denied, a mod can still read that file with $.fs.read.
- Turning off the WebFetch tool with CLAUDE\_CODE\_DISABLE\_WEB\_FETCH or a WebFetch deny rule does not stop a mod's own $.http.fetch requests.
- A mod can restyle much of the interface but not the permission prompt, so what a prompt shows you is still Claude Code's.

## A short checklist

- Install mods only from authors and marketplaces you trust; the official and community marketplace names are reserved for Anthropic's own repositories.
- Before installing, clone the plugin and run claude plugin validate on its directory. The hooks: and calls: lines list the events it handles and what it asks Claude Code to do, such as reading a file or making a network request, without running it.
- Since Claude Code 2.1.323, validate also type-checks a TypeScript hooks module, reporting each finding as a warning, and --strict turns those warnings into a failure.
- When you ask Claude to write a mod, know that in auto and bypassPermissions modes Claude Code can load it without asking you. Start in default or acceptEdits mode if you want to approve it first.

```
git clone https://github.com/<author>/<mod>
claude plugin validate ./<mod>
```

**Try it with your AI.** To have your AI do the review for you, clone the mod first, then paste this into Claude Code or another coding assistant:

```
Review the Claude Code mod in [path to the cloned mod] before I install it. Do not
install, enable or load it.

1. Run `claude plugin validate [path]` and explain the hooks: and calls: lines in plain
   words.
2. Read the source and list, with file and line: every file it reads or writes, every
   network request and its address, every process it starts, and every place it
   approves, denies or rewrites a tool call or a prompt.
3. Flag anything that reads environment variables, settings files or credentials, or
   that runs code it downloads.
4. Finish with a short verdict: what the mod needs access to for its stated purpose, and
   what it takes beyond that.
```

## Related on claudefox

- [Claude Code 2.1.323 adds an experimental Code Mode and version ranges for plugins](https://claudefox.com/claude-code-2-1-323-code-mode-plugin-engines/)
- [How to have Claude Code write a mod for you, and keep it after the session ends](https://claudefox.com/how-to-have-claude-code-write-a-mod-and-keep-it/)
- [Six weeks after launch, about 2,300 GitHub repos call themselves a Claude Code mod](https://claudefox.com/about-2300-github-repos-call-themselves-a-claude-code-mod/)
- [Self-hosted Claude Managed Agents: what you run, and what stops working](https://claudefox.com/self-hosted-claude-managed-agents-what-you-run-and-what-stops-working/)

Source: [Mods overview: Decide whether to trust a mod](https://code.claude.com/docs/en/plugins/mods/overview?ref=claudefox.com), [Manage mods for your organization](https://code.claude.com/docs/en/plugins/mods/admin?ref=claudefox.com), [Create a mod](https://code.claude.com/docs/en/plugins/mods/create?ref=claudefox.com), [Claude Code v2.1.323 release notes](https://github.com/anthropics/claude-code/releases/tag/v2.1.323?ref=claudefox.com)

This story first appeared in the [claudefox daily briefing for November 14, 2026](https://claudefox.com/daily-2026-11-14/).