agent-preflight

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The materials explicitly instruct granting broad, persistent tool permissions (including write/kill), recommend choosing "Allow for all tasks", and include an explicit permission-bypass flag (--dangerously-skip-permissions) plus automated OS-level messaging — together these patterns enable persistent unattended access and are easily abused as a backdoor.

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 granting broad, persistent tool permissions (e.g., "Always allow" / "Allow for all tasks") that bypass per-run security prompts and endorses approving write/kill/terminate capabilities and "approve everything" surfaced by a preflight, which encourages defeating OS-level protections and enabling actions that can modify or damage the machine state.