codex-skill

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

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly instructs use of flags like --dangerously-bypass-approvals-and-sandbox and -s danger-full-access (granting unsandboxed network/system access) and includes commands that run arbitrary installs and system-level tmux/git operations, which push the agent to bypass security controls and modify the host state.