peer-review

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill routinely ingests raw diffs/file contents and requires the reviewer to quote phrase anchors and include extracted content in findings, and although it warns and scans for secrets before sending prompts to external CLIs, that pre-flight scan/confirmation only applies to external CLI paths — the self/claude reviewer path consumes and may echo unredacted secret values verbatim, creating an exfiltration risk.

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly fetches PR title/body via gh pr view "$PR" and inserts that user-generated PR content into the <untrusted_diff> block (and may send it to external CLIs/models), so the agent will read and act on untrusted third‑party content from GitHub that could carry indirect prompt-injection instructions.