HypoGeniC — автоматическая генерация гипотез из текстов

★ 7.0 · research

hypogenic is a Claude Code skill that plans and audits the use of the ChicagoHAI HypoGeniC/HypoRefine package for LLM-assisted hypothesis generation from labeled text datasets. It covers the full workflow: installing `hypogenic==0.3.5` (Python 3.10+, uv), validating official task configs and the local run policy in `assets/run_config.example.json`, auditing HypoBench datasets for split leakage and duplicates, building a cost-bounded run plan, and requiring explicit confirmation before any external model call or dataset upload. HypoRefine enriches the iterative process with literature-derived signals, while merged hypothesis banks support text-analysis tasks such as deception detection and content classification. The output is a candidate hypothesis bank with task-prediction statistics—not experimental confirmation or causal evidence; independent scientific validation remains the researcher's responsibility. The skill is designed for teams that need a reproducible, auditable pipeline for pattern discovery in labeled text data.