baby-extract

Extract benchmark tasks for BABY (Build A Benchmark for Yourself) from the user's own work — agent chat logs (opencode/pi), git commit history, code projects, and documents. Stage real task drafts with verifiers and reference solutions, all kept local. Use when the user says "make a benchmark from my work", "extract tasks", "把这些工作变成评测题", "build an eval from this repo / chat / git history", or "which model/agent is…