detecting-dependency-confusion

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

• Third-party content exposure detected (medium risk: 0.30). In scripts/agent.py the helper parses first-party manifest files from the scanned repository (--path) to extract dependency names, then directly sends those outsider-supplied name strings into public registry HTTP requests via http_status(REGISTRY[eco].format(...)) and uses the results in its findings, so the runtime ingests outsider-originated free text (dependency names) supplied through user-controlled repo content.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill includes explicit runtime install/run steps that fetch and execute remote code (e.g., "go install github.com/visma-prodsec/confused@latest" and "docker pull ghcr.io/owasp-dep-scan/dep-scan"), so external content is downloaded and executed as part of the workflow/agent runs.