paper-referee-revise

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). This skill requires the model to read and reproduce verbatim content from user-provided files (complete referee comments and full manuscript excerpts), so if any of those files contain secrets (API keys, tokens, passwords, cookies, or other sensitive strings) the model will output them and thus creates an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.85). Outsider-authored free text is ingested when the skill reads referee report files at runtime (e.g., .txt/.md full content, .pdf extracted text, .docx extracted text) in Phase 1 step 7, and then includes the extracted referee comments verbatim in the LLM context for triage and per-comment revision.