run-parity

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md the workflow runs nr --silent eval and then streams and validates baseline.ndjson and candidate.ndjson (which contain rule/provenance text such as diagnostic message, help, url, and related fileContext fields) via scripts/compare-parity.mjs/validateReport/diagnosticsByIdentity, so outsider-submitted content from the target PR/head repository commits is ingested and transformed at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill and its run commands/reference manifests pass the git repository URL https://github.com/millionco/react-doctor.git into runtime evaluation commands (e.g., --react-doctor-repository to nr eval and in the matrix descriptor), meaning that remote repository contents would be fetched and executed as evaluator/detector code at runtime.