ykzeng-yale/Power-Agent-Public-STAI-X

Power Agent is an autonomous agent that converts plain‑English study‑design questions into verified power and sample‑size analyses. It plans the analysis, generates R code, runs it in a sandboxed R runtime, self‑corrects on errors, and returns the numeric result plus the exact R script, a reproducible report (PDF/Markdown), plots and CSVs. Who uses it: biostatisticians, clinical‑trial designers and applied researchers who need defensible, auditable power/sample‑size calculations across many R packages. Key features: a task‑agnostic plan→generate→execute→verify agent loop; sandboxed R execution as a service on Google Cloud Run with on‑demand CRAN/Bioconductor installs; multi‑model LLM orchestration; error observation and self‑correction; a 106‑task benchmark with package‑verified ground truth; and built‑in reproducibility. Repo contains backend, frontend, examples, benchmark and architecture docs; local run requires Node.js and an Anthropic API key; hosted at https://power-agent.io/.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.