panel

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill deliberately uses "dangerous" auto-approve/bypass flags and instructs sending repository diffs/files to external agent CLIs, which enables high-risk data exfiltration and remote code-execution capabilities despite superficial read-only safeguards.

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 launching agent CLIs with flags that bypass permissions/sandboxes and auto-approve actions (e.g. --dangerously-bypass-approvals-and-sandbox, --dangerously-skip-permissions, --always-approve), which encourages escaping safety controls and allows those subagents to modify the machine state despite "read-only" wording.