gpt55-critical-auditor

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly mandates a "deep, adversarial" failure-seeking audit that asks the agent to probe for sandbox escapes, persistent mutations, and enforcement bypasses—activities that inherently encourage modifying system state or attempting privilege/escalation to prove failures.