EXCLUSIVE: Anthropic's Claude Desktop bundles a dormant security scan for local skills

The off-by-default control would hold skills handled through the app for review by the organization's configured Claude model; passing skills are released, warnings leave them usable with a caution, and failures are blocked.

By · Published · Updated

RUNTIMEWIRE INVESTIGATION — Exclusive

Original reporting by RuntimeWire, based on reverse engineering, testing.

Why it matters

The code points to a security gate for locally authored Claude skills, with skill text sent for review through an organization's configured provider. The control is still dormant, and the reported workflow covers skills handled through Claude Desktop while leaving plugins and skills stored elsewhere outside its review.

Reporting record

Finding

RuntimeWire's analysis of a signed Claude Desktop 3.1623.0.0 package found a dormant, off-by-default workflow that holds skills handled through the app for review by an organization-configured Claude model and then releases, cautions, or blocks them based on the result.

How we verified

Methods: reverse engineering, testing.

The signed Windows package’s scan-and-release code and managed-configuration schema, comparison with one verified October 19 Claude Desktop build, and Anthropic’s documentation of its separate hosted Enterprise scanner.

Analyzed the Claude Desktop 3.1623.0.0 package downloaded from Anthropic on October 20, including its skill-scanning workflow and managed-configuration schema; compared the setting with one verified Claude Desktop build from October 19; and consulted Anthropic’s hosted-scanning help page to describe the separate Enterprise route.

Tested versions:

  • Claude Desktop 3.1623.0.0

Reproduction

RuntimeWire was not able to independently reproduce the finding.

Company response

RuntimeWire requested comment; the company had not responded by publication time.

A Windows desktop computer with an unreadable code-like screen sits in a secured review room beneath an Anthropic wall sign.

A signed Windows build of Anthropic's Claude Desktop contains a dormant control that would hold locally created skills while a configured Claude model reviews them. RuntimeWire's analysis of the binary found a complete scan-and-release workflow. The setting is off by default and marked for a future version.…

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Reader comments

Conversation for this story loads after sign-in.