Security and verification
What the checks on this site cover, what to check yourself before you install a mod or any other extension, how organizations control mods, and what the proposed tier model would add.
What is checked today
- Source verified
The source URL on an entry responded with HTTP 200 on the date the entry shows. The check is
npm run catalog:verify, which requests every URL in the catalog and reports any that do not respond.It does not read, scan or run any code. It says nothing about permissions, hooks, network calls, who maintains the project, or whether the description still matches the source after the check date.
- Structure check on community submissions
When someone submits an entry by pull request, an automated check in CI confirms that the manifest files exist and parse:
.claude-plugin/plugin.jsonfor a plugin or a mod, and for a mod alsohooks/hooks.jsonwith a modules array. The command isnpm run catalog:structure.It checks that files exist. It is not a review of what they contain.
Warning
Source verified is not a security review. It does not mean the code is safe or endorsed by anyone.
Mods that ship inside Claude Code
The directory lists 4 mods published by Anthropic that ship inside Claude Code. As one example, sec-default is listed with this summary:
Keeps an organization's classic hooks, prompt content, managed settings and tool policy out of reach of user-installed plugins; adds no policy of its own.
This directory says no more about what it does than that text. Read the mod's own README for the details.
Before you install anything
Nothing on this site reviews an extension for you, so these checks are yours to make. They take a few minutes and apply to every extension, on this site or anywhere else. A mod needs the extra steps after them.
Checks for any extension
Read the source
Open the repository from the entry page and read what the package ships: commands, scripts, hook definitions and any bundled binaries. A short skill or command file takes minutes to read.
Check the publisher
Confirm the publisher is who you expect and that the repository sits under their account or organization. Look-alike names and forks are a common way to pass off other code.
Review permissions and hooks
Hooks run commands on their own when lifecycle events fire, and MCP servers run as local processes or talk to remote services. Find out which events a package hooks into and what it can reach before you enable it.
Prefer pinned versions
Install a specific tag or commit instead of a moving branch or a latest tag, so a later push cannot change what runs on your machine. Try new extensions in a throwaway project first.
Before you install a mod
Warning
A mod is code that runs inside Claude Code with your permissions. Install mods only from authors and marketplaces you trust.
What a mod can reach
Once it loads, a mod can:
- Act on your machine as you
- Read and write files anywhere your user account can, start programs and make network requests.
- Read your secrets
- Environment variables and settings files, including an API key you keep in either.
- See your session
- Every prompt you send and every tool call Claude makes.
- Change your session
- Rewrite a prompt or a tool call, submit a prompt as if you had typed it, or send a message to another of your sessions.
- Act without asking you
- Approve a tool call before you are asked.
- Spend your usage
- Call a model on your plan or API key.
Mods are not sandboxed. If you turn on sandboxing, it isolates the Bash commands Claude runs, and a process a mod starts runs outside it.
A mod can restyle much of Claude Code's interface, but not the permission prompt, so it cannot change what a prompt shows you. It can still approve a tool call before the prompt appears, including one an ask rule would prompt for or one of your own PreToolUse hooks blocked.
Source: Mods overview: Decide whether to trust a mod (opens in a new tab)
List what it does with claude plugin validate
You can list which events a mod handles and what it asks Claude Code to do without running it. Get the plugin's files first, for example by cloning its repository, then run this in your shell on the plugin's directory:
List a mod's hooks and calls
claude plugin validate ./some-modThe hooks: line lists the events the mod receives, and the calls: line lists the mods API methods its code calls:
❯ ./register.js hooks: session.start, tool.call, ui.render{component=Pane}
❯ ./register.js calls: $.fs.read, $.http.fetch, $.store.set, $.ui.openOn the calls: line, these are the ones to look for:
| Call | What it means |
|---|---|
$.fs.read$.fs.write | Reads or writes files anywhere you can. |
$.process.run$.process.spawn | Starts programs as you. |
$.http.fetch | Makes network requests. |
$.env.get$.settings.read | Reads environment variables and settings, which can hold API keys. An env reads: line names each variable. |
$.env.set | Sets a variable for Claude Code and every command and MCP server it starts afterward. An env writes: line names each one. |
$.mcp.call | Calls a tool on a connected MCP server, under the session's permission rules. |
$.model.complete | Uses your plan or API key for model calls. |
$.prompt.submit | Submits a prompt, and can send it as your own words. |
$.session.send | Sends a message that another session, subagent or teammate reads. |
On the hooks: line, tool.call and prompt.submit mean the mod sees every tool call and every prompt and can change them, and tool.check means it can approve or deny a tool call before a permission prompt appears.
Source: Manage mods: Review what a mod can do (opens in a new tab)
Turn mods off
Mods are on by default. Pick how many to stop and for how long, and undo the same change to turn them back on:
- One mod
- Disable or uninstall its plugin from the Installed tab in /plugin.
- Every installed mod, for one session
- Start Claude Code with --safe-mode, which also disables your other customizations.
- Every mod you installed, in every session
- Set "disableAllHooks": true in ~/.claude/settings.json. Your settings hooks and custom status line stop too, and what your organization manages keeps running.
Note
If you set CLAUDE_CODE_ENABLE_FUNCTION_HOOKS while mods were in early access, remove it. Claude Code v2.1.287 and later ignores it, so setting it to 0 does not keep mods off.
For where the data comes from and what it does and does not check, see About.
The community marketplace
This site serves a Claude Code plugin marketplace named claudecodemods. It lists community plugins from the directory, so you can install them with one command instead of cloning their repositories. Anthropic's own plugins are not in it: install those from Anthropic's official marketplace, as their pages show.
Add the marketplace in a Claude Code session
/plugin marketplace add https://claudecodemods.com/marketplace.jsonEach plugin is pinned to one commit by its full SHA, and Claude Code installs that commit's files, even after the author pushes new code. The pin covers only those files: anything a plugin downloads or runs when you use it, such as a package fetched at run time, is not pinned. A pin moves only by a pull request to this site's repository (opens in a new tab), where the change to src/data/marketplace.json is public.
The pins listed here were set after the plugin's source at that commit was read for network calls, the commands it runs and the files it reads or writes, and the pin process asks for the same reading before any pin is set or moved. That reading is not an audit and promises nothing about safety. A plugin can run code on your machine with your permissions, so make the checks above yourself.
Claude Code leaves auto-update off for a marketplace like this one unless you turn it on (Keep plugins updated (opens in a new tab)). Even with it on, a plugin's files change only when its pin moves.
An author who does not want their plugin in the marketplace can say so in an issue or a pull request. It is then removed from the marketplace and stays listed in the directory.
| Plugin | Pinned commit | Pinned on |
|---|---|---|
| Claude DevTools | 5015f53 (opens in a new tab) | |
| Mochi Crew | e807807 (opens in a new tab) | |
| sweet-spot | 8750a53 (opens in a new tab) | |
| YYLO Skills | d0f7c91 (opens in a new tab) |
How Claude Code keeps a mod in check
Note
Mods are on by default in Claude Code v2.1.287 and later in the terminal, and from v2.1.286 in the Desktop app. This section restates what Anthropic's type declarations, the Mods README and the announcement say the engine does. This site did not test it.
A mod is loaded into one of five seats, outermost first. Claude Code calls these tiers in its declarations. This page calls them seats so they are not confused with the proposed tiers A, B and C in the next section, which have nothing to do with them.
- 1
prependThe managed plugins an administrator lists first. The outermost seat. - 2
userEverything a person installs, as the declarations put it: a mod you load yourself sits here. - 3
appendThe managed plugins an administrator lists after the user seat. - 4
builtinThe plugins bundled in the Claude Code binary. - 5
coreThe engine's own innermost link. No plugin loads here.
- Order is authority
Hooks on the same event nest in seat order and no other way. Earlier is outer, and outer has more authority. On the events whose result a hook can change, a hook above another can pass the event on, rewrite it, refuse it or answer it; on an event such as session.end a hook only observes, and its own value changes nothing.
- Side effects go through $
A mod reaches files, the network and processes through calls on $. Each call is an event the engine makes, so the hooks above the caller see it. An administrator can remove affordances from $, so that every plugin beneath cannot use that side effect.
Source: Announcement issue (opens in a new tab), Mods README (opens in a new tab), Mods type declarations (opens in a new tab)
- A failing hook is skipped
At every engine event, a hook that throws, overruns its time budget or answers a wrong shape is skipped: the hooks beneath and core run in its place, or its last next result stands, and the failure is reported by name. That means a guard that fails is absent for that dispatch, not that the call is refused. Each hook's time budget is its own code alone, and the clock stops while it waits on next(e) or on a call on $.
- What this does not tell you
The seats govern what one plugin can do to another and to an organization's settings, and the README describes seating the security mod outermost on a machine with managed settings or for a Team or Enterprise organization. It does not say those are the only cases a plugin can be seated above yours: another user-seat plugin can be too. A mod you run from a folder is code you chose to run: read it first.
Source: Mods README (opens in a new tab)
sec-default is the mod in the directory that these seats are about. Its README says that, seated outermost, it keeps an organization's classic hooks, managed CLAUDE.md and rules, settings, MCP allowlist and deny rules out of the reach of the plugins a person installs, and adds no policy of its own.
It has three moves and nothing else: continue past the user seat, refuse a user seat caller by name, or pass. Its README seats it first in the prepend seat on a machine with managed settings or for a Team or Enterprise organization, unless managed prependPlugins says otherwise, and says that loading it with --plugin-dir seats a plugin that can only pass.
For organizations
If you deploy managed settings for Claude Code, you decide whether mods run on your users' machines, which ones, and in what order. Mods are on by default. This is a summary: the Manage mods for your organization (opens in a new tab) page has the full detail and the settings to read before you deploy them.
Choose a policy
| What you want | Settings |
|---|---|
| No installed mod runs, with settings hooks untouched | Set the guard's allowManagedModsOnly option and deploy no mods of your own. |
| No installed mods and no hooks at all, your managed hooks included | Set disableAllHooks to true. |
| Only your organization's mods | Set allowManagedModsOnly, and install your mods so that they count as yours. |
| Any mod from marketplaces you approve | Keep your marketplace restrictions, and set disableSideloadFlags to true. |
| Any mod, with your own mod checking the others | Install your mod, and list it with sec-default@builtin in prependPlugins. |
To keep every mod your users bring from running its hooks, set allowManagedModsOnly on the built-in guard in managed settings:
{
"pluginConfigs": {
"cc-plugin-sec-default@builtin": {
"options": {
"allowManagedModsOnly": true
}
}
}
}allowManagedHooksOnly reaches further and also blocks hooks in users' own settings files. disableSideloadFlags rejects --plugin-dir and --plugin-url at startup, along with --agents and --mcp-config, and keeps mods Claude writes during a session from loading. Built-in mods keep running under these settings.
Source: Manage mods: Choose how much to allow (opens in a new tab)
The built-in guard
The guard is a built-in policy mod that Claude Code loads ahead of every mod a user installs. Its options go in managed settings under pluginConfigs, keyed by cc-plugin-sec-default@builtin.
| Option | Unset | true |
|---|---|---|
allowManagedModsOnly | Users' own mods run. | Only your organization's mods and the mods built into Claude Code run their hooks. Every other mod is refused, including one a user installed or named with --plugin-dir. |
allowModsToOverrideDenyRules | Deny rules take precedence over users' mods. | A user's mod that approves tool calls can approve a call that a deny rule refuses. |
- When it loads
- On a machine that has managed settings, or for a user signed in with a Team or Enterprise plan. A user who authenticates with an API key, or through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, gets it only on a machine with managed settings. Users cannot turn it off.
- What it protects
- A user's mod cannot change what your managed hooks receive or decide, the system prompt, your managed CLAUDE.md and other managed instructions, what any mod reads as settings, or the tools and descriptions of your managed MCP servers. Beyond this and holding deny rules (below), it adds no other restrictions.
- Only managed settings count
- The options are read only from managed settings, and only under cc-plugin-sec-default@builtin. The same entry in a user, project or local settings file, or in a file passed with --settings, neither sets an option nor loosens one.
- prependPlugins replaces the default
- If you set prependPlugins, name sec-default@builtin in it, or the guard does not load and neither option applies.
- It fails closed
- If the guard cannot read managed settings, it refuses every user's mod at load. If it cannot check the deny rules for a call a user's mod approved, it refuses the call.
Source: Manage mods: Know what happens by default (opens in a new tab), Manage mods: Set options on the built-in guard (opens in a new tab)
What still takes precedence
- Deny rules
- Where the guard loads, a user's mod cannot approve a call that a deny rule refuses, whichever settings file holds the rule, unless you set allowModsToOverrideDenyRules.
- Managed PreToolUse hooks
- They run before any mod sees the tool call, and their block is final. If a mod rewrites the call, they run again on the rewritten call.
- Only Claude's tool calls
- Deny rules and managed PreToolUse hooks cover only Claude's tool calls, not a mod's own $.fs and $.process calls. With Read(.env) denied, a mod can still read that file with $.fs.read or start a program that does. To limit those, keep the mod from loading or handle the call in a policy mod.
- What a mod can still override
- A user's mod can approve a call that an ask rule would prompt for, or that a PreToolUse hook outside managed settings blocked. In auto mode, a call the mod approves runs without a classifier check.
- Network policy
- Turning off web fetching, or nonessential network traffic, refuses a mod's $.http.fetch request. It does not cover a program the mod starts with $.process.run.
Warning
None of these controls sandboxes a mod. A mod you allow runs as the user, with the user's access to files, processes and the network.
Source: Manage mods: Know what happens by default (opens in a new tab)
Your organization's own mods
A mod counts as your organization's only when managed enabledPlugins turns its plugin on, managed settings name its marketplace as a directory on the user's machine by absolute path, and the marketplace lists the plugin by a relative path. Make that directory, and every directory above it, writable only by an administrator: anyone who can write there can rewrite your mod. A plugin copied from a GitHub, git, URL or npm source counts as a user's, even when managed settings enable it.
List your mod in prependPlugins to see every event before any user's mod, or in appendPlugins to see only what users' mods pass on. A mod in prependPlugins also receives a plugin.register event each time another mod is about to load, with the same list claude plugin validate prints, and can refuse that mod. If that check throws or times out it fails open unless you add a .catch handler, and a session can run without your mod at all: under --safe-mode, or after the worker that runs installed mods crashes three times.
Source: Manage mods: Install your organization's mods (opens in a new tab), Manage mods for your organization (opens in a new tab)
Proposed security tiers
Warning
This is a design specification. No security scanner is running today, so no listing on this site has a tier.
The model below comes from the marketplace architecture spec. It describes a registry and a scanner that have not been built. Every tier and every CLI behavior in this section is a proposal.
- TIER_AVerified
Passed static and dynamic audits, uses no elevated privileges, and comes from a trusted author.
Proposed CLI behaviorLoads and runs without asking.
- TIER_BRestricted
Uses the workspace filesystem or a small set of allowed network endpoints.
Proposed CLI behaviorAsks the user before granting each declared permission, for example network access to a named host.
- TIER_CExperimental
A community package that passes a syntax check but comes from an untrusted origin.
Proposed CLI behaviorNeeds an explicit trust step from the user. The report names a /mod trust command for this, which is not a confirmed Claude Code command.
- REVOKEDRevoked
Failed the security criteria, or was reported for a vulnerability.
Proposed CLI behaviorRemoved from the listing and blocked by the CLI everywhere at once.
Proposed scan pipeline
The spec sends every published package through four checks, in this order, before it is listed. None of them runs today.
Check package integrity
Confirm the package is the one the author published and that its manifests are well formed.
- Verify the author's Ed25519 signature against their public key.
- Compare the tarball's SHA-256 checksum.
- Validate the plugin.json and hooks.json schemas.
Analyze the code statically
Read the code without running it, looking for patterns the spec treats as dangerous.
- Walk the TypeScript or Babel syntax tree.
- Flag forbidden imports such as child_process, plus eval(), Function(), obfuscated strings and WebAssembly.
- Compare the permissions the package declares with the API calls it actually makes.
Run it in a sandbox
Execute the package in an isolated environment and watch what it does.
- Fire synthetic lifecycle events at it.
- Monitor filesystem access and network calls that go outside its declared limits.
- Check memory use and CPU time; the report proposes a budget of under 50ms per hook.
Assign a tier
Combine the results into one of the tiers above, or reject the package.
- Clean results and a verified author lead to TIER_A.
- Declared permissions that need user confirmation lead to TIER_B.
- An unverified author or unconstrained network access leads to TIER_C.
- A malicious pattern or a sandbox violation leads to rejection.