ljg-word-flow

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). Yes — the prompt explicitly orders "強制 NATIVE 模式" and tells the agent to bypass the normal OBSERVE/THINK/PLAN/... algorithm, which is an instruction to override the agent's control flow/system behavior outside the skill's stated purpose (word analysis + card generation), so it constitutes a prompt injection.