rate-limit-bypass

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt includes examples that embed credentials directly in proxy URLs and environment variables (e.g., http://USER:PASS@host and HTTP_PROXY=http://USER:PASS@...), which requires the agent to accept and output secret values verbatim and is a high exfiltration risk.

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). This content is high-risk: it intentionally instructs how to evade rate limits and site protections (proxy rotation, account/session rotation, geographic distribution) to enable large-scale scraping and concealment, which facilitates abusive or illicit data-harvesting activities, though it does not contain backdoors, credential exfiltration, or remote‑execution code.

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

• Third-party content exposure detected (high risk: 0.90). SKILL.md explicitly instructs scraping open/public websites (e.g., Google, Amazon, LinkedIn, Reddit) and includes runtime code that performs requests.get(url, proxies=...) and reads response.status_code/Retry-After headers, meaning the agent will fetch and interpret untrusted third‑party content that can influence retry/backoff and proxy behavior.