prime-fetch

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly requires including verbatim quotes or verbatim-tokens from fetched material in the lead's emitted findings (and verifies by writing/grepping temp files), which will force the LLM to handle and output exact substrings from external dumps — including any embedded secrets (API keys, tokens, passwords) — with no redaction rule, creating an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.75). Outsider free text can enter the LLM context via the delegated sub-agent executing the runtime tool call (e.g., WebFetch/MCP search/public web pages or other users’ conversation excerpts), even though the lead is supposed to receive only a filtered digest.