codex-autoresearch

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill deliberately defaults detached/background/CI runs to bypass approvals and sandboxing and enforces a “never ask after go” autonomous execution model, which enables arbitrary nested codex exec sessions and shell commands with full host access—creating a high-risk surface for remote code execution and data exfiltration if misused.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly defaults background detached runs to "danger_full_access" (--dangerously-bypass-approvals-and-sandbox) and instructs launching detached runtime sessions, which directs the agent to bypass sandboxing/security controls and perform potentially uncontrolled system actions.