codex-worker

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill intentionally instructs launching multiple unattended Codex agents with the explicit flag --dangerously-bypass-approvals-and-sandbox, enabling agents to run without safety controls and therefore poses a high risk of remote code execution and data exfiltration if abused.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt instructs creating and deleting git worktrees/branches and tmux sessions and explicitly runs codex with the flag --dangerously-bypass-approvals-and-sandbox, which requests bypassing security protections and thus pushes the agent to modify and potentially compromise the host state.