critical-thinking

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly requires copying exact text from the user's documents into the audit output and recording users' answers verbatim, which forces the LLM to reproduce any secret values present in those inputs verbatim (high exfiltration risk).

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

• Third-party content exposure detected (high risk: 0.80). The skill’s required runtime loop ingests the user-provided memo/argument text (“The document. Read it fully…”) into the LLM context for auditing and quoting, which is outsider-authored free text whenever the operating user is analyzing someone else’s document (public/provided by other parties, issues/PRs, chat/notes, attachments, etc.).