Claude release notes: Enterprise scans skills and plugins before use
Claude Enterprise can now scan third-party skills and plugins when someone uploads or edits them, before they are used in the organization. The feature is in beta for Enterprise and applies across Claude, Claude Cowork, and Enterprise plugin marketplaces.
A skill is a folder of instructions, scripts, and resources for a repeatable workflow. A plugin bundles skills, connectors, and sub-agents into one package.
Source: Claude Apps release notes: Skill and plugin security scanning (beta).
How Claude skill and plugin scanning works
Human step: An Owner or Primary Owner goes to Organization settings > Skills and turns on Skill and plugin security scanning. It is off by default. Once enabled, Claude automatically scans new or edited third-party skills and plugins at no extra cost.
Most scans finish in one to two minutes and run in the background. Results are cached, so uploading the same unchanged skill again returns a result almost immediately.
pass: installation continues without an extra message.warn: the tool remains available after the user acknowledges a caution banner.fail: the tool is blocked. Neither the uploader nor an administrator can override the block.
Source: Get started with skill and plugin scanning.
What scanning covers and misses
Scanning applies to standalone third-party skills, third-party plugins, and skills bundled inside a plugin. It is not retroactive: material already in the organization keeps working when scanning is enabled.
It does not scan skills created with Claude, skills shared through a connected MCP server, MCP servers themselves, or hooks. Organizations using CMEK, zero data retention, or HIPAA configurations are also excluded. MCP, the Model Context Protocol, is the standard that connects Claude to external tools and data sources.
A pass means the scan did not find the type of malicious behavior it checks for. It is not a general safety guarantee. Keep a clear owner for every installed tool and give plugins and connectors only the permissions their workflow needs.
Source: Get started with skill and plugin scanning: scope and limits.
FAQ
Does Claude scan existing skills and plugins when the feature is enabled?
No. Scanning applies to new uploads and edits after activation. Material already in the organization keeps working.
What happens after a warning or failed scan?
With `warn`, the user can continue after acknowledging a caution banner. With `fail`, the tool is blocked and neither the uploader nor an administrator can override it.
Does the feature scan MCP servers and hooks?
No. Skills shared through a connected MCP server, MCP servers themselves, and hooks are outside the scan's scope.
The Forge newsletter
Get new articles in your inbox
Pick the topics you care about. No noise, at most one email a week.
We follow GDPR. Unsubscribe anytime.
