sst-llm-judge-ranker

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

• Third-party content exposure detected (high risk: 0.85). The skill’s comparator reads the contents of candidate files via the harness’s Read tool (e.g., in Mode 2/3 it ingests <contents of A> and <contents of B>), and those candidate files are outsider-authored documents provided by the operating user (not bundled reference material), so their free text can be included in the LLM context for pairwise comparisons.