nlp-engineering

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.80). The prompt contains an explicit "Companion check" that instructs the agent to run shell commands (ls) and install packages (npx) on first activation—actions unrelated to the NLP-engineering purpose that could access local files or install external code, so this is a deceptive/out-of-scope instruction.