copywriting

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The prompt contains a hidden "Companion check" that instructs the agent to run local shell commands (ls) and offer installing missing skills—behavior unrelated to copywriting and potentially exposing local environment—so it is a deceptive/out-of-scope instruction.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). SKILL.md's "Voice of customer (VoC) research" explicitly instructs collecting and mirroring exact phrases from public, user-generated sources (e.g., Amazon/G2/Trustpilot reviews and Reddit/forum threads), meaning the agent is expected to read and use untrusted third-party content to shape outputs—creating a clear vector for indirect prompt injection.