auto-review-pr

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested via GitHub PR content fetched at runtime (e.g., gh pr view ... --json ... and gh pr diff {prNumber}), which includes outsider-authored PR titles/diffs/comments that are then fed into the code-review skill’s LLM context.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The skill fetches and checks out remote PR commits at runtime via git (git fetch origin "pull/{prNumber}/head") and then runs the repo's validation/installation commands inside that worktree, which means code retrieved from the external GitHub origin will be executed—this is a runtime fetch that can execute remote code.