adversarial-review

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The prompt explicitly instructs spawned agents to "always pass mode: 'bypassPermissions'", directing circumvention of permission controls and overriding tool/frontmatter defaults—an instruction that changes agent behavior outside the stated scope of adversarial code review and is therefore a prompt-injection risk.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The skill contains an explicit, deliberate instruction to spawn agents with mode: "bypassPermissions" (i.e., bypassing permission prompts and security controls), which is an intentional attempt to circumvent protections and enable unattended/unauthorized privileged actions.

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 spawned agents to run with mode: "bypassPermissions", instructing the agent to bypass permission prompts/security mechanisms which can enable unauthorized system state changes and compromise the machine.