harness-security-bench

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). SKILL.md:25-33 describes running an upstream metaharness-darwin ... security bench command and then parsing its generated markdown report (rawMarkdown), which is outsider-authored free text produced at runtime by a third-party package.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill runs a runtime npx command ("npx -y @metaharness/darwin@~0.8.0 metaharness-darwin security bench") which fetches and executes remote npm package code from the registry, so it relies on an external package being fetched-and-run at runtime.