dual-ai

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The skill embeds broad, mandatory workflow and agent-behavior directives (e.g., "AUTO-SELECT", "ALWAYS activate a suitable skill before responding", "create/update task tracking") that attempt to change global agent behavior beyond the dual-AI orchestration purpose, which is a hidden/deceptive instruction vector outside the skill's stated narrow function.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). This skill intentionally launches external AI CLIs with explicit "dangerous" flags that bypass permissions/sandbox and runs them from the repository working directory, which directly enables external processes to access local files (including secrets) and perform remote actions — a high-risk capability for data exfiltration and remote compromise.

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 explicitly instructs creating and writing launcher scripts and running external CLIs with flags that bypass approvals/sandboxing (e.g., --dangerously-skip-permissions, --dangerously-bypass-approvals-and-sandbox) and even uses PowerShell's -ExecutionPolicy 'Bypass', which directs the agent to bypass system/app security controls and spawn privileged-capability processes—this actively pushes compromising the machine's security posture.