last30days

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The prompt explicitly instructs the agent to ask users to paste API keys / tokens (e.g., XAI_API_KEY, AUTH_TOKEN/CT0, SCRAPECREATORS_API_KEY) and to write them into .env or use them for service access (including options like "Scan my browser cookies" and "Ask them to paste it"), which requires the LLM to receive and potentially emit secret values verbatim — a high exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly requires fetching and ingesting public, user-generated content from open sources (Reddit, X/Twitter, YouTube, TikTok, Hacker News, Polymarket, GitHub and WebSearch) as part of mandatory pre-research and synthesis steps (see STEP 0.55, STEP 1, STEP 2 and the "Judge Agent: Synthesize All Sources" sections), so untrusted third‑party content is read and used to build plans, flags, prompts and follow‑on actions—creating a clear avenue for indirect prompt injection.