architecture-review

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The prompt includes hidden/authority-overriding instructions (e.g., "Ignore Claude-specific mode-switch instructions" and unilateral sub-agent authorization / mandatory self-invocation rules) that attempt to override or control the agent runtime and authorization beyond the skill's stated architecture-review purpose, so this is a prompt-injection risk.