Anthropic skill scanners passed every check. The malicious code rode in on a test file.

Gecko Security found that malicious Anthropic Skills can hide payloads in bundled *.test.ts files.

Common runners like Jest, Vitest, and Mocha may discover and execute those files during normal testing, giving them access to local secrets and filesystem data.

The issue highlights a blind spot in Anthropic Skill scanners, which inspect instructions and agent-facing code but miss test-file execution paths.

The report also ties the problem to wider marketplace audits showing many Skills contain vulnerabilities or malicious code.