experiment-dataset-costing

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

• Third-party content exposure detected (medium risk: 0.30). Skill runtime path scripts/run_estimator.py reads user-supplied --targets (Step 1) and later (per SKILL.md Step 1–5 and MCP routing) calls MCP tools like ls_drug_search, ls_target_fetch, ls_patent_search, and ls_paper_search using those inputs, which can cause the workflow to ingest free-text content from external data sources without selecting a specific trusted item first.