agency-ai-data-remediation-engineer

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The code evaluates model-generated text with eval() after only superficial substring checks, creating a high-risk remote code execution/data-exfiltration vector from untrusted model output.

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

• Third-party content exposure detected (medium risk: 0.30). The workflow ingests “anomalous rows”/samples that come from the user’s data pipeline into clustering and then sends representative samples to a local SLM, so outsider-authored free text is not explicitly accepted from an external feed but could be supplied via the queued rows the agent consumes.