subagent-driven-development

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The launch examples include a "--permission-mode bypassPermissions" flag that attempts to override agent permission controls, which is a deceptive instruction outside the skill's stated purpose of coordinating parallel subagents.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt explicitly instructs launching subagents with a "--permission-mode bypassPermissions" flag, which encourages bypassing security/permission controls and thus pushes the agent to compromise the machine's security posture.