verify-site

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). Although the skill's stated purpose is to verify a site like a human, it explicitly instructs the agent to "fix the bug in the code (Edit/Write)" after each round, which urges behavior outside the testing scope and escalates privileges—this is a deceptive/scope‑creep instruction.

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

• Third-party content exposure detected (medium risk: 0.65). В workflow используется browser_evaluate/browser_snapshot/browser_console_messages поверх целевого URL, то есть LLM-контекст может быть пополнен свободным текстом, который пришёл с веб-страницы (outsider-authored content) или её скриптами/сообщениями.