adversarial-review

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill explicitly instructs sending code/diffs and repository/principles files to an "opposite model" via external CLIs (including advice to skip repo checks), which creates a clear and deliberate data-exfiltration vector for source code and any embedded secrets; there is no direct evidence of obfuscated payloads, reverse shells, or explicit backdoor RCE, but the design intentionally bypasses local safeguards and therefore is high-risk for leaking sensitive data.