coding-agent-loops

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). Explicit guidance to bypass permission checks (--dangerously-skip-permissions) and to run persistent, unattended AI agent sessions (long sleeps/tmux) enables deliberate remote code execution or unauthorized actions and can facilitate data exfiltration or backdoor behavior.

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 repeatedly instructs running the agent with --dangerously-skip-permissions and encourages unattended, persistent sessions that can modify repositories and system state, which explicitly bypasses permission/safety checks and risks compromising the host.