cmux

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The document explicitly instructs injecting repository .env secrets into agent surfaces and using per-launch flags that bypass sandboxing and permission prompts, which together enable credential exposure and remote code execution by untrusted agents.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The prompt explicitly instructs launching agents with flags that bypass approvals and sandboxes (e.g., "yolo" / "--dangerously-bypass-approvals-and-sandbox", "--dangerously-skip-permissions") and to inject environment files into workspaces, which encourages unsandboxed execution and credential injection that can modify or compromise the host system state.