predictor-hand-skill

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

• Third-party content exposure detected (high risk: 1.00). The skill's system prompt Phase 2 (Signal Collection) and the HAND.toml tools list require performing web_search and web_fetch on public news and social sources (the "data_sources" setting includes "social"), meaning the agent ingests and interprets untrusted third‑party web/social content as part of its prediction workflow, which can materially influence actions and enable indirect prompt injection.